When considering the 1960s sitcoms Bewitched and I Dream of Jeannie, both of which featured women with supernatural powers navigating life with mortals, most people wouldn’t connect them with pursuing an engineering career. But Karen Panetta did. The sitcoms’ main characters—Samantha Stevens, a witch; and Jeannie, a genie—were “strong, empowered female leads using magic,” Panetta says, and they inspired her to become an engineer, as it was like sorcery to her.

Panetta, an IEEE Fellow, is dean of graduate education at the Tufts University engineering school, in Medford, Mass., outside of Boston.

Karen Panetta

Employer

Tufts University, in Medford, Mass.

Title

Dean of the engineering school’s graduate education

Member grade

IEEE Fellow

Alma maters

Boston University and Northeastern University in Boston

Like Samantha and Jeannie, Panetta has made magic happen, such as when she helped to invent the first CPU digital-twin simulator. Digital twins are computer simulation programs that track and adjust the operations of a physical device in detail. Her simulator has been adapted for several industrial uses, including by NASA to help design spacecraft.

Panetta also mentors young women to encourage them to pursue a STEM career through the Nerd Girls program she launched at Tufts in 2000. Engineering undergraduate students work on technology for socially conscious projects such as environmental cleanup, renewable energy, and the development of assistive devices to improve mobility for people with disabilities.

Panetta received this year’s IEEE Mildred Dresselhaus Medal for “contributions to computer vision and simulation algorithms, and for leadership in developing programs to promote STEM careers.” The award, sponsored by Google, was presented at the IEEE Honors Ceremony on 24 April in New York City.

Receiving the medal is particularly special to Panetta, she says, because she knew its namesake: Mildred Dresselhaus, an IEEE Life Fellow who pioneered the study of carbon nanostructures at a time when researching physical and material properties of commonplace atoms was unpopular. She was a MIT professor of physics and electrical engineering, and died in 2017.

Panetta nominated Dresselhaus for the IEEE Medal of Honor, which she received in 2015.

“Millie was a rock star,” Panetta says. “I can’t think of another medal that really encapsulates her spirit and what I’ve dedicated my life to.”

Finding a creative outlet in engineering

As a child growing up in Boston, Panetta built trapdoors and other features in her treehouse, she says.

“I also explored fashion and sewed my own clothes,” she adds. “I wasn’t very successful, but I was very creative.”

She was a top performer in math and science classes in high school, so her father encouraged her to pursue civil engineering.

“I didn’t know what an engineer was, and my father, who was a mechanic working on heavy construction equipment, only knew about civil engineers,” Panetta says. “I started taking computer programming classes at school, but knowing how to type on a keyboard and make a software program wasn’t good enough for me. I wanted to know what was inside the box.”

Her thirst for knowledge inspired her to pursue a bachelor’s degree in computer engineering at Boston University.

“My father was very disappointed that I didn’t pick civil engineering,” she says, laughing.

She commuted to school, and she struggled to find study groups for her classes, so she joined IEEE to connect with peers.

She became active in the university’s student branch, organizing events including the IEEE Student Professional Awareness Conference, which helps students learn practical career skills including résumé building, interviewing, and networking. She organized a SPAC for her branch, and IEEE Life Senior Member Jim Watson volunteered to speak at the event. It changed her life, she says.

Watson was the director of commercial and industrial marketing at Ohio Edison in Akron, where he worked for 36 years.

“He flew to Boston to speak at our event, but fewer than 20 students attended. I was embarrassed,” Panetta says. But Watson told her the important lesson was that she showed up and organized the event.

“He said I would be successful because of that,” she says. “He didn’t care about the attendees’ grade point averages, only that we were professional enough to organize the talk.

“That encouragement was the first time anyone outside of my family ever told me that I would succeed, so it was reaffirming. To this day, I still use some of the techniques that I learned in his presentation in my own classroom to teach students.”

Panetta graduated in 1986. Her IEEE membership helped her get hired for her first dream job: a diagnostic engineer at Digital Equipment Corp.

While attending the IEEE Computer Society’s annual symposium on very large-scale integration in Boston, she handed her résumé to a DEC representative, who hired her to work in Hudson, Mass.

While working full time, Panetta attended Northeastern University, in Boston, as a part-time graduate student. She earned a master’s degree in electrical engineering in 1988.

Developing the first CPU digital twin

In the early 1990s, Panetta was assigned to work with Ernst Ulrich, one of DEC’s most respected consulting engineers, she says. He was developing a new CPU using millions of CMOS transistors.

“I thought, ‘Wow, what a great opportunity,’” she says, “not realizing they assigned it to me because no one else wanted to work with him, as he set rigorous standards, expecting those who worked with him to think outside of the box and hold their own to bullet-proof new concepts.”

Panetta and Ulrich wanted the ability to test the CPU while still designing the hardware and software. That way, both would be ready to use at the same time. Typically, the hardware was developed before the software was written.

“We decided that we were going to simulate the machine to see how it was going to run—which was unheard of,” she says.

During a meeting with the company’s top engineers, Panetta shared her idea for an algorithm that could accomplish the team’s goal. She was met with silence.

“It’s going to be the engineers who better society because we know how to work together. We’ve proven that IEEE members know how to work across geographic boundaries, ethnic boundaries, and gender boundaries. And that’s a good model for the world.”

“I thought to myself, ‘Did I just say something stupid?’” she says. “But then, the top engineer looked at me and said, ‘I have been doing this for 50 years, and you, a kid just out of school, comes up with this [solution] like it’s obvious.’”

Her idea became the basis for the digital twin simulator. It used behavioral models to run software on a CPU simulation. The software passes information through the system, she says, just like it would pass information through wires or interconnects.

“We did successfully have a complete model of millions of transistors,” Panetta says. “I efficiently simulated hundreds of thousands of experiments and ran the software on this simulated model so that we knew exactly how it was going to perform on the real machine. That had never been done before.”

Her groundbreaking work led to a promotion: from computer analyst to principal software engineer.

When she began managing a team and hiring staff members, Panetta noticed the younger employees knew the theory but didn’t have the technical skills to hit the ground running, she says.

“It took the company two years to train somebody before they could really contribute technically to a team,” she says. She decided she wanted to help prepare students for jobs in industry.

In 1995 she was accepted into DEC’s Engineers and Education program, in which full-time employees who wanted to teach could take a leave of absence to complete a degree while still being paid. Participants were then placed in academic institutions for two-year stints to help students bridge the gap between classroom theory and real-world problem-solving.

After earning a Ph.D. in electrical engineering from Northeastern in 1994, Panetta began her teaching assignment at Tufts. After one year, she left her job at DEC to join the university as its first female electrical engineering professor. At the time, the department had only one female undergraduate EE student.

“I showed up to work dressed in an all-pink suit,” she says, laughing. “Other professors looked at me like I didn’t belong there because I looked different.”

She didn’t let that stand in the way of reaching her goals: preparing the next generation of students for jobs and mentoring young women who were interested in becoming engineers but who felt they wouldn’t be accepted and therefore couldn’t pursue a career in the field.

Launching the Nerd Girls program

When Panetta began teaching, she noticed that students weren’t getting any hands-on engineering experience, so in 1996 she created an internship program. It was the precursor to Nerd Girls.

At the time, she was consulting for NASA’s data visualization and animation lab in Langley, Va., translating complex information into a user-friendly animated form. The programs visualized Earth’s atmosphere and identified pollutants, their origins, and their effects on people and the environment.

Panetta needed a larger team to help conduct the research, so she asked her undergraduate students if they wanted to participate.

“Female students flocked to me because they could relate to the work I was doing, loved how their skills could benefit humanity, and didn’t see me as the classic nerd professor with no life,” Panetta said in a 2008 interview with The Institute about the program. “Eventually, the girls outnumbered the boys.”

“The research project ended up winning awards,” she added. “Tufts couldn’t believe that undergrads had a hand in it. That’s when things really turned around.”

Nerd Girls officially launched at Tufts in 2000 as a class where students work closely with industry on engineering projects. Examples have included building a solar-powered car, developing a battery for the last functioning twin lighthouse in the United States, and creating devices to help people train service animals.

“Everyone who has participated in the program graduated with a bachelor’s degree,” Panetta says. “I’m also very proud that 98 percent of participants pursue a graduate degree within three years of earning their bachelor’s.”

The program is open to all students, regardless of gender.

Creating a community at IEEE

Panetta became an active IEEE volunteer in 2004 after meeting Arthur Winston, the IEEE president at the time. Winston, an IEEE Life Fellow, was an electrical engineering professor at Tufts. He helped found the Gordon Institute, a leadership-focused engineering school at the university.

“I sat next to him on a bus, and he invited me to attend the IEEE Boston Section meetings,” she says.

Panetta eventually was elected by the section as a member-at-large—which allowed her to attend conferences and other events.

To help spread the word about the Nerd Girls program throughout IEEE, Winston connected Panetta to Mary Ellen Randall, who was chair of IEEE Women in Engineering at the time. Randall is the current IEEE president and CEO. Panetta joined IEEE WIE and was elected as its 2007–2009 chair.

In that position, she worked with Randall and Leah Jamieson, the 2007 IEEE president, to hire more staff to support the program and launch its magazine.

“At that time, we didn’t have any way to connect to members or tell the stories of women in technology,” Panetta says. “I wanted people to read the stories of women from around the globe and how they overcame adversity. So I launched the IEEE Women in Engineering Magazine in 2007.”

Panetta serves as the award-winning publication’s editor in chief, and she is a member of several other IEEE societies and committees.

IEEE is helping to change the world for the better, she says.

“It’s going to be the engineers who better society,” she says, “because we know how to work together.

“We’ve proven that IEEE members know how to work across geographic boundaries, ethnic boundaries, and gender boundaries. And that’s a good model for the world.”

What could you do if you could make a circuit trace by just bending a piece of paper? How about bridging modern technologies and traditional handicrafts while providing opportunities for learning skills in both.

As part of our interdisciplinary research into digital craftsmanship at the MEI Lab at the School of Creative Media, City University of Hong Kong, we came across research that demonstrated how to impregnate paperlike material (technically a “nonwoven textile”) with the kind of liquid metal used to make conductive ink. Initially, the impregnated material is nonconductive because an insulating oxide layer forms that encapsulates microscopic droplets of the liquid metal. However, applying pressure via shaped molds will crack open the insulating layer, allowing neighboring particles to merge, and thus creating conducting regions in the shape of the mold.

Both of us were introduced as children to origami and kirigami (similar to origami, except that cutting is allowed in addition to folding). We, along with our colleagues, decided to see if those traditional techniques could be used on the new material to eliminate the need for molds. Our goal was to allow crafters to make hybrid papercraft creations that contained easily integrated elements such as LEDs and motors.

In particular, we were interested in the possibility of combining the separate stages of creating a papercraft object and adding electrical conductors. Previous approaches to creating electrified papercraft objects relied on adding a separate flexible conductor—such as adhesive copper tape—to the paper. This increases the effort required and runs the risk of creating open circuits as the conductive material conforms to the object’s shape.

Isopropanol and a gallium-indium liquid material are used to impregnate a paperlike material that is 55 percent polyester and 45 percent cellulose. Electronic components such as LEDs and motors are held in place with masking tape. James Provost

Our first step was to see if the pressures involved in bending and cutting alone would be sufficient to create conductive traces. We became frequent visitors to our university’s materials science and engineering department to fabricate samples and then to borrow equipment to characterize their behavior.

We soon confirmed that the pressures involved in folding and cutting—ranging from 2.5 to 100 megapascals—were enough to create conductive traces. We also confirmed that normal handling of the paper didn’t accidentally create conductive paths.

We made a number of changes to the original method for creating the impregnated paper. For example, instead of immersing the paper in a mixture of isopropanol and liquid metal, we used an airbrush to spray the mixture onto the paper. That allowed us to vary how much was deposited on the paper and to use cardboard stencils to mask some areas from being impregnated, allowing folding and cutting in those regions without creating unwanted conductive traces. We also experimented with the ratios of isopropanol and liquid metal.

We became frequent visitors to our university’s materials science and engineering department.

After optimizing the mixing ratios and amount applied via airbrush, we were left with a material that reliably conducts with a resistance of 23.18 ohms per centimeter for cut edges and 4.4 Ω/cm for folded edges. The folded edges retain their conductivity even if later flattened out, and the conductivity is the same on either side of the paper. We estimate the combined cost of the paper and liquid metal (available from many online vendors) is about US $1.80 to make a 10- by 10-cm piece.

The next step was attaching electronic components to the traces. To make the connections more flexible, we cut down the rigid leads of LEDs and attached conductive thread to the stumps. We then held the threads in place using masking tape. Similarly, we connected conductive thread to the terminals of a power supply.

As our goal was to use this material educationally, we now needed to make it easy for a beginner—whether in papercraft or electronics—to try it out. We created a toolkit, dubbed LiqMetCraft. This consists of all the required materials, plus a browser-based software tool that lets the user select or create designs and then gives guidance on physical construction.

We created three versions of LiqMetCraft. The first is based on Chinese papercraft in which a piece of paper is folded into a fanlike segment and then cut to create a radially symmetric design. We provided circles of paper with a doughnot-shape impregnated region, with an untreated region that created a gap in the donut. We attached positive and negative terminals to either side of the gap. The user could specify in the software how many times they wanted to fold the disk and then draw potential cuts, receiving immediate feedback on what the unfolded disk would look like, as well as guidance on how to place LEDs.

To make our paper sample, isopropanol and liquid metal are mixed in specific ratios while being cooled by an ice bath. Sonic waves are used to ensure the liquid metal breaks up into microscopic droplets. The mixture is then applied via airbrush, while stencils prevent some areas being covered for different papercraft templates. James Provost

The second version of LiqMetCraft was based on origami. We supplied rectangular pieces of paper with two conductive regions separated by a border down the middle. The software tool provided templates for 12 origami designs, with step-by-step instructions for folding them. Once the project was completed, the user could add components, such as a motor, by taping them to the folds.

The final version supported 3D paper model making. In this case, the initial paper supplied was a rectangle with an untreated rectangular central area. By cutting this paper in half and then further cutting the halves into patterns separated by a spacer, the user could make various self-standing models. The software allowed the user to draw a pattern on screen, and then have a cutting machine produce a template for cutting the impregnated paper.

We had 42 participants, evenly divided into three groups, try out the different versions. All found it easy to use, and we were pleasantly surprised that some participants moved beyond the supplied designs to their own creations.

For full details of the current process, see our open access LiqMetCraft research paper published in CHI ‘26: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. In the future, we plan to try different substrates for the impregnating solution, as well as explore further types of papercraft, such as pop-up books. We’re also interested in developing ways to use the material to support inputs as well as outputs by constructing switches and potentiometers directly out of the material. Imagine traditional papercraft creations becoming interactive devices!

Take a moment and try to imagine your life without the wireless advances of the past three decades.

Have you lost your luggage? What a shame AirTags have not been invented. The airline representative has promised to call with updates, so settle in for a long wait by the kitchen telephone, because there are no affordable cellphones. You’ll be stuck listening to whatever is on the radio while you wait, because there are no streaming services. That’s not even to speak of all the movie plots that would have been ruined.

This is just a tiny sliver of how wireless technology makes itself felt in your day-to-day existence. The effects it has had on supply chains, infrastructure, and how the economy runs have been world-altering.

None of it would be possible without the radio-frequency integrated circuits that allow all our devices to unobtrusively send and receive information.

Now imagine what the further evolution of this technology will bring: Wide-spread autonomous vehicles, quantum communications, 6G mobile service and satellite communications. Continued momentum will depend on newer and more advanced versions of today’s RF chips.

But there’s the rub. Whereas the design of most of the world’s computing chips has been standardized into its own science, RF design has remained stubbornly in the realm of art. A dark art, even, that is mastered only through years of experience. As any sorcerer will tell you, the dark arts keep their own schedule. And that schedule is impeding progress not just in RF chip design but in every other technology that depends on it.

About seven years ago, in the wake of AlphaGo’s victory over world Go champion Lee Sedol, my students at Princeton and I began to wonder: Could AI be taught this art as well? Recent successes suggest that, to a large extent, it can. Over the last few years, our group and other leaders in the field have started to develop machine-learning-driven algorithmic methods for designing RFICs. Some of the resulting chips look more like modern art than circuit layouts. Yet in many cases, the physical prototypes bested state-of-the art circuits in terms of performance. The real achievement, however, is that it took the AI orders of magnitude less time to conceive a working design than it would a human designer.

This is not about one or two RF chips. AI-enabled design could be the future of all RF design, and maybe much more.

The Dark Art of RFIC Design

So why do these chips all have to be crafted by hand? Why aren’t RFICs designed with an algorithmic synthesis process, much as CPUs and GPUs are?

The design of RFICs is an exercise in engineering across multiple physical domains. Maxwell’s equations, operating across different spatial and temporal scales, govern how electromagnetic fields interact with active and passive devices that must be carefully codesigned for the chip to function. Alongside these are the laws of thermodynamics, which determine how heat is generated and removed during operation, as well as the mechanics of thermal expansion and contraction that dictate how reliably the chip and its packaging survive temperature changes.

Simultaneously accounting for all the physical constraints these impose makes the design space almost impossibly large. Every decision involves complex priorities that often compete with one another, preventing the optimization of any of them.

To better understand the issue, let’s walk through the steps involved, after which you’ll better understand why a single new chip design takes years and tens to hundreds of millions of dollars.

Most of the area of radio-frequency integrated circuits is dominated by complex electromagnetic structures. Human-designed RFICs, like this broadband power amplifier [1], start with templates and follow a symmetric, understandable pattern. But freed from the constraints of human-designed templates and the need for humans to even understand the rationale of electromagnetic structures, power amplifier ICs [2–5] and low-noise amplifiers [6] can take on truly wild-looking yet efficient designs. SENGUPTA LAB

Let’s say you’re an engineer assigned to design a new 28-gigahertz power amplifier for a 5G-millimeter-wave handset. (This is the type of RFIC that boosts the 5G signals on your phone and transmits them to the antenna where they can be picked up by a distant base station). Where do you start?

RFIC design has some features in common with house building. Just as the blueprint for a house dictates the number of bedrooms and bathrooms to be built and the hallways connecting them, the blueprint for an RFIC—called the architecture—establishes the kinds of elements the RFIC needs to fulfill its intended function. Instead of rooms, the architecture includes, for example, the number of stages of amplification your power amplifier needs. Instead of hallways, it shows the paths that signals must take to get through those stages.

The blueprint for RFICs is actually mostly hallway; passive elements, like inductors and transmission lines, take up far more real estate than active elements like transistors.

Here’s why. As you have probably experienced yourself, a typical CPU’s transistors overheat when faced with operating frequencies of just a few gigahertz. The frequencies RFICs can operate at are higher by an order of magnitude—28 and 39 GHz for 5G signals, 26.5 to 40 GHz and even higher for satellite communications, and 77 GHz for automotive radar. Under this onslaught, a CPU’s transistors would fail.

RFIC transistors avoid this fate because these chips cleverly manage the signal’s energy with careful electromagnetic design. This takes the form of byzantine networks of metal elements that dominate the chip’s real estate. These structures are geometrically regular, often symmetrical, and so intricately constructed they sometimes resemble lacelike filigree. But while they may look decorative, they are essential to the chip’s functioning.

Electrically speaking, these “hallways” work more like the chip’s plumbing. Like plumbing, this extensive labyrinth of passives confines electromagnetic energy only to the places it should be traveling around the chip.

The major challenge in RFIC design is putting all these elements together to ensure they work, just as constructing a house from its blueprints demands exact specs for load-bearing beams, pipes, and external walls. On an RFIC, the architecture needs to be realized with physically fabricable transistors and passive components that are connected just so, to permit the signal to travel through the chip and be processed. The way these devices are connected locally is what we call the circuit’s topology.

The RFIC Design Process

To make that power amplifier, then, your first step is to identify a candidate circuit template: The combination of structures that will meet the goals of a particular architecture with a specific circuit topology. Over the years, researchers have eased your burden by developing reusable design templates for specific functions. For example, templates suggest how many amplification stages a circuit needs (because sometimes, combining the output of two smaller amplifiers will result in better bandwidth and efficiency than you would get from a single larger one). And they suggest what the general configuration of the passive structures should be. Today there is an extensive library of such templates.

However, these can’t simply be used off-the-shelf, because each comes with trade-offs. Some have better gain at the expense of stability; some better bandwidth at the expense of efficiency; still others are more energy efficient at the expense of output power, and so on. There is rarely a clear best choice.

To arrive at the “sweet spot” where all these different parameters are balanced into optimal harmony, designers will typically lay out several different versions of the circuit, using intuitions and methods they have picked up in their years of training.

The challenge is that the decision around the architecture, circuit topology, or the electromagnetic passives cannot be done separately. One decision influences the others. So, designing an RF circuit can often feel like trying to fit an oversized carpet into too small a room—press down one corner, and another pops up.

At microwave and millimeter-wave frequencies, even the smallest misstep is the difference between a chip that works and one that doesn’t, and any number of things can go wrong. For example, when an electromagnetic wave encounters a transistor—or any other component —the path it travels must be properly “matched” to what comes next. If it isn’t, some of the energy reflects backward instead of flowing forward. Imagine trying to connect a high-pressure fire hose directly to a narrow garden hose. Without the right adapter, water will splash backward at the junction. Very little will make it through. In electronics, this is called the impedance-matching problem.

To prevent those reflections, engineers design special transitions, essentially microscopic adapters, that smooth the handoff between components. On a chip, these adapters can be surprisingly intricate. They don’t just pass the signal along; they can also split it, combine it, or distribute it across multiple paths with carefully controlled timing and strength.

Once you’ve done the architecture, plumbing, and everything in between comes the moment of truth. Have all the choices you have navigated through the enormous design space resulted in an RFIC that meets its specifications? If the specifications are not met, you will have to go back, either redoing the topology or the entire architecture, and repeat the whole process. So get ready for months of time- and resource-heavy simulation and iteration. Perhaps you now see why, for decades, a core belief has persisted in the RFIC community: “RF design is an art.” It was said that only an experienced designer—with an artisanal understanding of how the pieces make up the whole—could master the subtleties of analog and RF design. Unfortunately, this entrenched notion has long held back algorithmic innovations in the field just when we need them most. Traditional, artisanal RFIC design is hitting its limits as the complexity of these systems inexorably grows.

AI for RFIC Design

While RFIC designers continued their battle against their “oversized carpet” problem, a series of interesting developments emerged in allied disciplines. Across a range of other previously intractable problems like protein folding and climate modeling, AI has been able to successfully navigate multidimensional complex spaces. This gave us the incentive to look deeper into AI for RF. After all, the combinatorial complexity of protein folding is not that different from the nature of the design space in our domain.

We were not the first to think of using artificial intelligence to speed up parts of RFIC design. Researchers had previously trained machine learning algorithms on circuit templates in the hope of speeding up the normal optimization processes. While this approach was undoubtedly faster than humans at optimizing templates, it still relied fundamentally on libraries of existing designs invented by humans.

We didn’t want that. We wanted to break free from the restrictions of prefabricated topologies. Because while a designer’s experience and hard-won heuristics are crucial to building a working design, they also place fundamental limits on it. Furthermore, such an approach would necessarily require simulation steps as part of the optimization cycle, and even the fastest simulations use a lot of computing resources. Worse still, in many advanced cases, such as for broadband designs, there are no existing templates.

But if we didn’t start with templates, where could we start?

The goal here was to allow algorithms to determine—entirely from scratch—every parameter for architecture, constituent circuits, and electromagnetic passives. This approach differs fundamentally from conventional optimization, which is limited to determining the parameters—like transistor dimensions and passive component geometries—that optimize structures originally devised by humans.

In our new approach, the architecture begins essentially from nothing and is progressively assembled through successive iterations. The system explores the design space by generating myriad candidate circuit combinations and mapping the resulting performance trade-offs as it navigates this landscape. Because the process is not biased by prior human design choices, it can produce completely novel circuit topologies that look markedly different from those created by human designers.

In some ways, the approach echoes AI systems such as AlphaGo Zero, which achieved superhuman performance not because it was trained on games played by humans but because it explored the rules by playing against itself. Similarly, our algorithm develops new circuit architectures by exploring and evaluating its own design strategies. In so doing, it learns to understand circuits, electromagnetics, and the close codesign they need to achieve the end-to-end design of RFIC.

Inverse Design for RFICs

To realize this capability, we proceeded in two stages. First, we developed a reinforcement-learning (RL) framework that determines the optimal system architecture, circuit topology, device parameters, and even the properties of the electromagnetic interfaces that connect different circuit elements. In this stage, the algorithm effectively defines how signals should propagate and interact across the system.

The algorithm trains very similarly to how a computer learns to play a game. If you let it play enough times, it can learn to play better by observing the relationship between the actions it took and the score it achieves. In a similar way, the RL agent here learns to design effective circuits by playing with a set of combinations, and over time, it can map the space between the circuit performance to its architecture, topology, and parameters. This training takes a few days to a week, but once trained, the agent can design circuits very quickly

The next step was to determine the physical structure of the IC’s electromagnetics—the plumbing—that can create the desired properties of the passive elements, which are characterized by a set of metrics called scattering parameters. These measure if a signal entering a component actually moves forward—or is reflecting backward, being wasted, as in our previous example with the fire hose and the garden hose.

Deriving the structure from the desired scattering parameters is an example of an approach called inverse design, which appears across many areas of engineering. In structural engineering, for example, one might collaborate with an architect on a physical goal—such as creating large interior spaces with high ceilings—and then determine the arrangement of arches or buttresses that can support it.

Generative AI for Electromagnetic Networks

But RF integrated crcuits pose a particular challenge for inverse design: The process must account simultaneously for circuit behavior and the electromagnetic responses of the interconnects and passive elements that link them together. But it has to figure that out without doing a lot of artisanal iterating.

So we replaced our RF circuit simulator with an AI-based emulator. This AI model can predict the behavior of electromagnetic fields going through any structure—even totally arbitrary two-dimensional shapes—without having to compute the underlying physics from scratch, as simulation tools do. It would predict the solution of Maxwell’s equations and tell you the scattering parameters for any structure you showed it, without actually doing the math. With such an AI in hand, what a time-consuming electromagnetic solver normally takes minutes or hours to accomplish is reduced to milliseconds.

We chose to build our emulator around a convolutional neural network—a machine learning model that has been remarkably successful for image processing. Such networks can extract spatial features from any structure, and it turns out that the image of a structure contains a lot of spatial information that can accurately predict its electromagnetic performance. Then we trained it on a vast number of random pixelated structures whose scattering parameters had been labeled.

Once we had our inverse-design RL and suitable AI emulator, we essentially had an end-to-end AI designer. So we asked it to design us a power amplifier.

Unconventional RF Architectures

In 2023, we published this proof of concept—a power amplifier targeting the millimeter-wave band, specifically spanning 30 to 100 GHz, which covers most of the relevant 5G and radar frequencies. The final design achieved the best combination of wide bandwidth, output power, and efficiency then reported for a silicon-based power amplifier—meaning it could amplify a large amount of data across a wide swath of frequencies—while maintaining record efficiency.

The structure of the IC’s electromagnetic pathways was unlike anything any human would ever consider. Since the AI is not trained on human designs, the layout that emerged looked more like an arbitrary pattern or perhaps a QR code than the regular symmetrical structures we are used to seeing.

One unexpected insight revealed by this prototype, and our research generally, is that there’s no evidence that the templates we’ve historically relied on are even close to optimal for modern design goals. It’s not that a human designer can never come up with a better design. But with the removal of the templates and the time to synthesize cycle upon cycle of optimized circuits, it is now clear that AI-driven synthesis could break traditional design barriers and push the limits of RFIC capabilities.

Our 5G amplifier had only one input port and one output port. Adding more inputs and outputs to a design is not straightforward. Every port electromagnetically couples to every other port, so the scattering parameters quickly add up. Two ports give you four scattering parameters. Four ports, 16 scattering parameters. The math gets ugly fast. Could our model keep up?

We next trained our model on larger classes of electromagnetic structures with many input and output ports. In 2024, we published work showing that multiport integrated circuits are no problem for these AI algorithms either. Where previously multiport electromagnetic simulation required days or weeks of toil, this model evolved new structures in minutes. Since then, a plethora of work in the space by research communities across the globe have demonstrated the power of inverse design in RFIC.

Combining the reinforcement learning framework with the inverse design, we now had the ability to create an RFIC from specifications all the way to a fabrication-ready layout. We’ve so far shown this is true for RFICs ranging from low-noise amplifiers to subterahertz and broadband power amplifiers. The hope is that this will work just as well for other circuits.

Making AI Designs Interpretable

Our goal was to make RFIC design better and easier, but we didn’t want to make it beyond human understanding. Chip testing and debugging is a long, arduous process, sometimes even more so than design. Engineers often prefer ICs to have interpretable structures, so that if a problem crops up, they can understand how the chip works well enough to debug it.

To create structures that are more interpretable, we turned to diffusion models, which you may know from their remarkable ability to generate realistic images from text prompts.

AI-driven synthesis could break traditional design barriers and push the limits of RFIC capabilities.

Imagine you go to your favorite image-generation engine and ask it to create a painting of the sky in the style of Picasso, Van Gogh, or Michelangelo. You will get images that capture the essence of their brushstrokes, their use of colors, and their framing. All are pictures of the sky nonetheless, but in different styles.

Electromagnetic design is similar in that multiple structures can have very similar electromagnetic responses. Instead of using text input, we used scattering parameters as our input, and the electromagnetic structure of an RFIC chip as our output. As part of the inputs to the diffusion model, we created a dial that sets the spatial frequency of the final structure. By turning the dial, a designer can direct the model to synthesize structures with low (classical-looking and interpretable), medium (mazelike structures), or high (pixelated or arbitrarily-shaped) spatial frequency.

From prompts to output, the entire process took about 6 minutes. With this diffusion model, algorithms can now both discover novel architectures and accelerate the creation of conventional, so-called classical ones.

All an RFIC designer needs to do is specify virtually any valid set of scattering parameters. As long as they are physically realizable under Maxwell’s equations, the model pops out a corresponding structure as if it were a vending machine.

The Future of AI-Driven RFIC Design

The results of our investigations have drawn the attention of the RF community. The traditional bottom-up design process is clearly beginning to reverse.

But there are still questions: How generalizable are these methods? Can they consistently deliver truly high performance? Can we get to a place where AI produces designs that maximize every conceivable trade-off, holistically optimizing every parameter to its most ideal physical state? We want to take this strategy beyond RFIC design and invent other kinds of circuits that are different from anything humans have ever done.

These are exciting and ambitious prospects, but we are not there yet. AI can hallucinate a design that creates bad circuits that don’t work. This means verification methods need to remain under human oversight. And, while hallucinations are rare, it would still be good to reduce their occurrence.

History suggests that meeting these dreams of the future will take much more data than we’ve been using. Before the creation of the ImageNet repository—a repository of 14 million varied, human-annotated images—image-recognition models didn’t function well in the real world. The datasets they had been trained on were too tiny to be effective. ImageNet’s massive amounts of training data ushered in a revolution that led to AI that can generalize and recognize images in the wild. The rest was history.

If the goal for RFIC and analog design is a universal foundational model—something that learns the governing laws of electromagnetics and circuit behavior—then we also need data.

The good news is that this data is plentiful. Around the world, countless engineers at companies and academic labs simulate nearly identical RF circuits and passive structures every day. The bad news is that it’s all locked away behind nondisclosure agreements.

Open ecosystems have propelled other areas, and we think the RFIC community should do the same. There had been some movement toward this. Natcast, the operator of the U.S. CHIPS and Science Act’s R&D program, would have bolstered shared infrastructure and innovation for the next generation of wireless, sensing, and defense technologies. Unfortunately, both the organization and the program it ran specifically for machine learning and RFICs have been closed.

But the momentum Natcast’s effort sparked hasn’t died out. Building on our early work, groups across the community have already demonstrated remarkable advances. AI-driven IC design is part of a much broader technological shift. From biology and materials science to automotive and aerospace engineering, AI is reshaping how complex systems are conceived and optimized. Deeper collaboration between AI researchers and chip designers will unlock the field’s full potential. It’s by no means a foregone conclusion, but if we get this right, this genie won’t stay in its bottle.

5G telecommunications, according to industry hype when 5G first launched in 2019, was going to be all about buzzy applications like mobile augmented reality and autonomous vehicles. But the surprise plot twist came when replacing home cable internet turned into 5G’s most widely adopted new application.

Fixed wireless access (FWA) now serves over 14 million U.S. customers, and contributes 28 percent of worldwide wireless traffic. Fixed wireless access is what the term sounds like: broadband internet delivered over a cellular radio link to a stationary location—no cable, no fiber, no trenching, no satellite broadband antenna pointed at the sky. What makes FWA distinctive is that it repurposes the same towers, spectrum, and 5G infrastructure that was built for mobile devices.

One U.S. Federal Communications Commission (FCC) commissioner has called FWA 5G’s killer app. And that’s true not just in the United States either. Jio, India’s largest carrier, is also one of the world’s largest FWA providers, with over 9 million customers as of last year.

Carriers discovered they could repurpose surplus 5G capacity, while also exploiting a usage pattern quirk: mobile traffic starts to drop after 8 p.m., just when home internet usage peaks. The result is broadband, delivered via traditional cellphone towers, at a lower cost than fiber deployment. For these reasons FWA provides real price competition to cable broadband, while reaching underserved rural and suburban communities.

Fixed Wireless Access Repurposes Ambitious 5G Infrastructure

FWA is cheaper to deploy than fiber, and for most homes and small businesses, fiber’s gigabit speeds are overkill anyway. And since FWA uses the same wireless networks built for cellular service, FWA works anywhere that receives a steady cellular signal.

As cellular networks extend into rural and underserved areas, FWA’s coverage map expands with them. In these remote locales, the other main viable broadband alternative typically comes from satellite services like Starlink—which are, compared to FWA, more expensive, with higher delays, and lower bandwidth.

While most FWA deployments use currently underused microwave bands, some FWA deployments use electromagnetic spectrum that 5G launched but that mostly failed with mobile users. Millimeter waves operate at frequencies 10 to 40 times higher than 4G’s spectrum, offering high data rates from their wide available bandwidth.

However, there are good reasons 5G mobile users today don’t generally use millimeter wave spectrum. Millimeter waves can’t penetrate buildings. Plus, they lose signal strength within a kilometer or two of the transmitter. Millimeter wave antennas are also a real drain on cellphone batteries compared to microwave and radio wave tech.

Yet none of these challenges applies to a fixed station with a clear line of sight to a nearby tower. FWA home units (called customer premise equipment or CPEs) outperform 5G handsets by a significant margin. That’s mostly because of hardware. CPEs carry larger, more sensitive antennas than a typical cellphone, paired with more capable transceivers. CPEs also tend to be plugged into wall outlets, making battery concerns a non-issue.

Another 5G technology that did not gain traction in mobile wireless is Multi-User Multiple-Input Multiple-Output (MU-MIMO). A base station with MU-MIMO uses an array of antennas to serve multiple users on the same frequency simultaneously.

However, maintaining a MU-MIMO signal involves tracking each user individually—a problem that quickly becomes overwhelming with enough mobile users. FWA is different, however. Static CPEs, with their steadier downlink traffic loads, are an ideal match for MU-MIMO technology.

So, FWA internet service not only uses mostly fallow spectrum but also uses 5G spectrum more efficiently than do 5G mobile users—for whom, of course, these 5G technologies were originally designed!

How FWA Became 5G’s Surprise Killer App

Not long ago, the high-bandwidth use cases for 5G made for an impressive list: millisecond latency for autonomous vehicles, mobile augmented reality headsets with extensive high-speed data needs, and massive machine connectivity for an expanding internet of things (IoT).

These applications have all stalled. Autonomous vehicles pose challenging—and still unsolved—problems unrelated to spectrum allocation. Augmented and virtual reality technologies have yet to create meaningful spikes in bandwidth demand. And the IoT has, to date at least, fragmented across an array of competing standards.

Mobile carriers had built dense 5G networks for mobile customers whose needs rarely saturated the network’s capacity. Home broadband usage peaks in the evening hours, precisely when cellular networks are quietest.

FWA sits at cellular networks’ crossroads of supply and demand.

The Advent of 6G Will Only Expand FWA’s Reach

In December, the telecom standards body, the Third Generation Partnership Project (3GPP), issued its latest 5G specification—Release 20, the final “5G only” update. So, although 6G is still years away (its first specifications are expected in early 2029), engineering decisions that will define 6G are being made today. And FWA is not on the margins of that conversation; FWA is currently considered an established day-one use case.

6G wireless technology promises to expand FWA’s reach—not only via spectrum but also via geometry. Instead of following 4G and 5G’s connectivity model—strong signals near towers and weak signals far away—future 6G networks will let homes connect to multiple towers simultaneously, using a technology called distributed MIMO (multiple-input, multiple-output).

Where 5G’s version of MIMO (a.k.a. massive MIMO) concentrates user communication with dozens of antennas at a single tower, distributed MIMO uses antennas across multiple base stations and coordinates them to deliver signals to your home from multiple directions simultaneously.

The practical result: because no single tower is responsible for any given connection, the “edge” of a cell network—that outer boundary where signal strength falls off and service degrades—no longer represents a hard limit on who gets well served. A home that would once have been too distant from a tower, or blocked by terrain, could now be within reach of several base stations working together.

6G may eventually adopt distributed MIMO technology for mobile users, when synchronization challenges and other signal engineering hurdles are solved and deployed for real-world cellular networks. The jury, as of 2026, is still out on whether the full distributed MIMO problem will be solved once the 6G standards start to be set in place, within three years.

As demand for FWA grows, carriers will also deploy increasingly capable millimeter wave infrastructure for fixed customers first—the stationary CPE use case that millimeter wave best suits. The dense millimeter wave antenna infrastructure that FWA requires is the same infrastructure that future mobile applications will eventually inherit. AR glasses, AI-powered wearables, and other bandwidth-hungry applications originally promised for 5G are not canceled—they are waiting for the infrastructure to arrive.

The pathway to FWA is being prepared at lower frequencies, too. There is growing interest today in the largely unoccupied FR3 band, which spans roughly 7 to 24 gigahertz, situated between crowded low/mid-bands and the much higher millimeter wave frequencies.

Recent field trials by Nokia have demonstrated FR3’s viability for both cellular and FWA applications. FR3 is emerging as one of the more promising near-term frontiers for extending FWA coverage beyond its current footprint.

None of this was the plan. No carrier executive in 2020 stood on a stage and announced that 5G’s defining achievement would be delivering living room broadband to rural homes and suburban subdivisions underserved by cable.

FWA became 5G’s killer app because the engineering economics made it happen. Surplus wireless capacity met unmet consumer broadband demand, with the physics of a stationary receiver doing the rest.

That is not a criticism of the engineers or the carriers. It is simply how technology sometimes advances—sideways, through gaps nobody was trying to fill.

But FWA’s model of prioritizing unconnected users may in the end prove to be telecom’s on-ramp to everything else. Fix the digital divide first. Tomorrow’s sci-fi future appears set to follow close behind.

By most accounts, the United States appears poised to fall woefully short of meeting new electricity demand over the next five years as data centers and domestic manufacturing proliferate.

Ian Magruder

Ian Magruder is the founder of Utilize Coalition and previously served as director of market mobilization at Rewiring America, an affordable electrification advocacy group.

Building new power plants and transmission lines may seem like the obvious solution, but there are other options, says Ian Magruder, founder of Utilize Coalition, a nonprofit based in Washington, D.C. The U.S. uses only about half of its grid capacity, and a lot more power could be tapped by deploying a spate of newly available technologies.

Backed by Google, Tesla, HVAC systems manufacturer Carrier, and several other companies, Utilize Coalition advocates for more thorough use of grid capacity through policy change and new technologies. Magruder spoke with IEEE Spectrum about those efforts.

Why does the United States use only half of its grid?

Ian Magruder: Most studies have found that average utilization rates are between 40 and 55 percent across different geographies. And the reason is that we’ve built our grid to meet peak demand. We have to ensure that on the hottest summer day or the coldest winter morning we have enough power. But in many parts of the country, we really only hit peak a few days a year, and it’s really only a few specific hours within those days.

It didn’t used to be this way. What’s changed?

Magruder: Over the last 20 years we’ve seen the gap between average use and peak use grow wider. There are a variety of reasons for that. Grid operators have become more conservative following major blackouts and reliability events. And with more variable-generation sources such as wind and solar, grid operators are building in more capacity. But this also presents us with an incredible opportunity to get more out of the grid using new technologies.

What technologies are being deployed to address the problem?

Magruder: Pairing battery storage with energy generation is a key part of this, as are other kinds of distributed energy resources, like managed [electric vehicle] charging and smart thermostats. I would also say that transmission technologies that safely maximize the current in power lines, increase conductivity, and optimize power routes all play a critical role here. And then there’s demand flexibility, which is when utility customers adapt their power use to accommodate the grid during peak hours. Some really good work is being done around flexible data centers.

Is grid underutilization also happening elsewhere in the world?

Magruder: It’s a global phenomenon, but it varies widely by country. European grids face similar dynamics as [those in] the U.S., and in some places utilization is even lower. But Australia and the United Kingdom are further ahead in measuring and managing utilization with new technologies.

What’s the downside to overbuilding our grids?

Magruder: Mainly cost. Electricity rates have gone up, and we [at Utilize Coalition] think it’s because utilization has gone down. A report that we released earlier this year shows that a 10 percent increase in grid utilization could save Americans over US $100 billion over the next decade.

Imagine sitting down at your desk and logging in for a performance review, with an AI system analyzing the conversation. You’ve been working long hours, balancing deadlines, and your manager asks how you’re doing. You say you’re fine, and maybe even smile, but there’s a hint of hesitation and your voice wavers. As you shift your posture, your shoulders slump.

These are subtle cues that to the human eye might hint at underlying stress. But to an AI model that’s been trained only to categorize emotions as “happy” or “sad,” such nuances are likely lost. It logs the words and a smile and moves on—and unless your human manager intervenes, the fact that you’re tired, unfocused, and maybe a couple of days from burnout never enters the equation.

“Emotion AI,” which estimates how people feel based on facial expressions, voice tone, and behavior, seems to be suddenly everywhere; it’s being used in employee well-being and recruitment interviews, education platforms, and driver-monitoring systems. Technology call-center platforms such as NiCE and Genesys use AI to detect when a customer sounds frustrated and prompt agents in real time to slow down or respond with more empathy. Giant companies like Meta and startups such as Hume AI are developing more-expressive voice AI systems that can detect emotional cues in the person they’re “talking” to and adjust how they communicate.

What’s more, hundreds of companies already offer virtual AI companionship apps, a fast-growing market that may be worth an estimated US $555 billion by 2035—and robot buddies have also entered the picture. Intuition Robotics’s ElliQ, for example, is a small device vaguely resembling a white desk lamp that’s now being used to engage older adults in conversation in hopes of reducing loneliness.

But while the field of emotion AI is advancing at a rapid clip, most existing systems are focused on detecting a limited number of signals to label one specific emotion at a time—which is insufficient if you’re trying to understand the human condition. In the real world, human signals and emotions are contextual, overlapping, and constantly changing. A laugh can signal joy, nervousness, or both; a raised voice might signal enthusiasm just as easily as frustration. To make the job of emotion detection even more difficult, reactions differ greatly from one individual to the next, depending on demographics, cultural background, and countless other variables.

In other words, there’s a gap between what we’re expecting AI to pick up on and what AI can actually deliver. That’s the gap a new field of research—what we call human-context AI—is working to close. Instead of looking at just one input and labeling it, human-context AI increasingly has the capacity to take stock of an individual’s personality and character, and to track emotions in real time while combining multiple inputs, including facial dynamics, voice, tone, language, and behavior. Crucially, responses are also evaluated in the context of a specific environment, such as a performance review or professional coaching session. The result? Computers are learning to read the scene, rather than just the screen.

The Origins of Emotion AI

The story of emotion-sensing AI began almost three decades ago in the MIT Media Lab, where the American electrical engineer and computer scientist Rosalind Picard coined the term “affective computing.” Her work introduced the radical idea that computers could be taught to recognize and respond to human emotions.

Picard’s early experiments focused on single modalities: facial expressions, tone of voice, and physiological signals, such as skin conductance or heart rate. The goal was to give machines a window into human feeling, helping them become more empathetic. It was an exciting vision, but back then the science and hardware weren’t ready. Computing power was limited, sensors were crude, and datasets were narrow and biased.

Josie Norton

Over the next decades, researchers and companies got better at measuring the many ways in which humans express themselves. In the 2010s, sentiment analysis—the processing of large volumes of text to suss out emotional undertones—began to reach the mainstream. At the same time, marketing firms, including my company, Neurologyca, began using video and webcams to measure and catalogue customer reactions. Biometric devices and activity trackers, such as Fitbits and Apple watches, also became ubiquitous, generating new streams of data about people’s sleep, step counts, stress levels, and more.

Unsurprisingly, scientists soon confirmed that larger volumes of personalized data led to greater accuracy in reading human emotions. In 2019, researchers at Cornell demonstrated that combining multiple types of signals improves emotion sensing. Their system joined physiological data, such as brain activity measured by electroencephalography (EEG) and heart rate, with visual cues like facial expression, outperforming systems that relied on just one input. Around the same time, Picard and her team at MIT found that humanoid robots trained on data unique to a specific person were substantially better at reading that person’s reactions and feelings than robots acting without personalized data.

More recent studies align with these findings. In 2024, scientists in South Korea showed that fusing physiological, environmental, and personal data to recognize emotion resulted in a 32 percent error reduction. Another paper, published in 2025, demonstrated that user-specific information significantly enhances emotion recognition performance.

Today, our devices know who we are; our habits and tendencies, likes and dislikes. They’ve also gotten smaller and more efficient. Tiny, low-power cameras and microphones embedded in phones, laptops, and virtual-reality and augmented-reality devices can detect dozens of human signals simultaneously, from eye movements and micro-expressions to breathing rhythms, voice modulation, and posture. Advances in computing have also made it possible to integrate audio, video, biometric, and text data, often without even transmitting raw data to the cloud. And researchers at Stanford, Cambridge and MIT, and Kyoto University, in Japan, as well as the Software College of Northeastern University in Shenyang, China, are exploring how fusing such inputs can refine the sensitivity and accuracy of human-machine interactions.

And yet, despite so many breakthroughs, machines still can’t reliably interpret emotion or even physical stress. Just last year, a survey published in the Journal of Psychopathology and Clinical Science revealed that stress scores on smartwatches rarely, if ever, matched the level of stress that users were experiencing. In fact, a quarter of those surveyed reported feeling the direct opposite of what their smartwatches were reporting.

Why the disconnect? We’ve gotten very good at capturing signals, but not at interpreting them. A fitness tracker might infer from your heart rate that you’re stressed and recommend easing off training, but it doesn’t know if your increased heart rate is due to excitement, tiredness, or an extra cup of coffee. Gauging emotions in real-world settings is even more difficult. To solve this complex problem, machines need context.

From Neuromarketing to Emotion-Sensing AI

My company, Neurologyca, was founded in Spain in 2015, and started out in neuromarketing. Working with major European brands and conglomerates, our cofounder, Juan Graña, had realized that companies lacked solid data on consumers. At the time, most customer feedback came through surveys, which posed questions such as, “On a scale of 1 to 10, how joyful does this car advertisement make you feel?” or “Which emoji best describes your mood?” Naturally, these overly simplistic tools led to high levels of self-reporting bias, as people often misjudge or misstate their own reactions.

To get around this problem, Neurologyca set up labs, using neuroscience and cognitive science to more accurately capture human responses to products, logos, advertisements, and experiences. In addition to using biometric tools such as heart monitors, eye trackers, and EEG, we recorded millions of video frames of human reactions, logging each specific context and the resulting facial and bodily movements. To do this, we mapped over 790 points of reference, including corners of the mouth, size of the eyes and pupils, blink rate, and angling of the head. All of this data was collected and stored anonymously under strict European privacy standards.

Next, we paired this information with findings from decades of neuroscience and behavioral science studies on how biometrics, speech patterns, and human movement are related to emotion—research we continue to gather from academic institutions across Europe. We also created a database of situational contexts—for example, “watching a dog food commercial” or “hearing a new song”—and the human feelings they engendered.

In our work with companies, not only did this approach allow us to recognize nuanced emotions, it also let us identify which reactions indicated positive or negative outcomes. Take, for example, the context of horror-film trailers: Our research helped us figure out that the most successful elicit a very specific mix of emotions, namely a little bit of fear, a little bit of anxiety, but also some joy. With this knowledge, we could quickly rate viewer reactions to help a film company figure out how to tweak its trailer for the desired impact.

Neurologyca

Within a few years, we discovered that a model trained on our database could accurately evaluate emotion using just a webcam. We stopped needing to host focus groups in rooms full of equipment. Instead, we were able to do such things as sending out a new perfume sample to paid participants around the world along with a link. When people opened the link, it turned on their cameras, allowing us to record their faces as they sniffed the perfume for the first time. Suddenly, we had expanded our reach: Rather than using small focus groups in one or two countries, we could quickly assess 1,000 people across the planet, comparing how someone in Japan, India, or Germany might feel about a certain product.

About four years ago, as AI was becoming pervasive, we realized that our models had applications well beyond neuromarketing. Importantly, these models are grounded in directly observed human behavior rather than inferred patterns or loosely labeled open datasets. Looking beyond brands and companies, we established that our model could be integrated into AI systems to help them understand human emotion at a much more granular level. In other words, we could provide a layer of context.

For Empathetic AI, Context Is Key

When we talk about “a layer of context,” we mean three different types of context. The first is situational or environmental context; for example, a performance review, a telemedicine session, or a horror-film viewing. The second is personal context, which includes an individual’s specific history, goals, and baseline state. The third is behavioral context, which covers the individual’s reaction over the course of the event or interaction by evaluating real-time changes in attention, confidence, engagement, and cognitive load.

Most systems today focus on only situational context, although some are starting to include personal context. Very few include behavioral context or combine all three in a meaningful way. What we’ve built at Neurologyca is a logic layer that fuses the three and translates them into structured, machine-readable information that allows AI systems and agents to respond more effectively. Our technology is being used to enhance systems in development, as well as some that have already been deployed, including driver-safety apps like Netradyne, home assistants like Amazon Alexa, and health-care AI platforms like Sully.ai.

It works as follows: Situational context is determined by the platform or application, be it a professional coaching session, a meditation app, or a driver’s safety monitor. Personal context already lives within each respective platform—or if not, it can be created through sharing of personal data or monitoring via camera. (Most wellness and professional-development apps, for example, contain each user’s profile, history, and prior sessions.) Last but not least, behavioral context is collected and analyzed in real time using our models. In the end, our logic layer fuses these three streams of information.

Our system doesn’t assign fixed weights to the three contexts. Instead, it provides a continuous calibration, with the balance shifting depending on the specific situation. For example, a pause in speech might signal uncertainty in a performance review, but something entirely different in a relaxation setting. If signals are ambiguous or overlapping, our system reflects that uncertainty through lower confidence scores rather than forcing a definitive interpretation.

What’s more, our system can work without ever sending raw data to the cloud, thereby easing privacy concerns. In many cases, video, audio, and biometric signals never leave the device. Instead, our lightweight models extract information locally and share only what’s necessary. Cloud systems, meanwhile, are used for training, pattern analysis, and model improvement. The result is a hybrid architecture: edge-based processing for speed and privacy combined with cloud-based learning for continuous improvement.

The result? By incorporating context, AI systems are beginning to interpret aspects of the human state as interactions unfold, dynamically adapting to emotions rather than reacting after the fact. The range of potential applications is broad and still evolving. Picture a professional-development platform that uses a human avatar to perform a mock interview and then provide feedback and tips on how to appear more confident, likeable, and well-informed. Or a meditation app that knows exactly how well you slept and how anxious you’re feeling, and can recommend an appropriate breathing meditation. Or a humanoid robot teacher that can tell when a student is confused or bored and step in to get them back on track.

Avoiding Potential Dangers on the Road Ahead

There have long been debates about the ethics of emotion-sensing AI. Some critics question whether systems should attempt to infer human feelings from external signals at all. They argue that reducing people to measurable outputs risks oversimplifying human experience while opening the door to manipulation, surveillance, and unfair judgments in workplaces, schools, and public spaces.

We take those risks extremely seriously. In fact, our technology aims to reduce the dangers of oversimplifying human emotion. Human-context AI is not based on the assumption that a machine can definitively know what someone is feeling. Rather, it is an attempt to move beyond simplistic labels by incorporating situational, personal, and behavioral context, while explicitly representing uncertainty when signals are ambiguous or incomplete.

That said, ethical concerns regarding implementation are real and have shaped the kinds of projects we pursue. We would never, for example, accept military engagements to help with interrogations. Not only for ethical reasons: Emotion AI cannot reliably detect deception, and claiming otherwise would be overstating what the technology can actually do. And while our technology can be used to gauge crowd behavior and predict things like when a football stadium is at risk of becoming destructively rowdy, we don’t want our technology deployed for surveillance. In short, we believe that using our logic layer on anyone who hasn’t opted in would be intrusive and ethically problematic.

In Europe, our systems are designed to comply with the EU AI Act’s restrictions on emotion recognition in workplaces and schools; as we expand into the United States, we apply jurisdiction-specific guidelines while maintaining the same core ethical commitments.

We also don’t advise companies to become overly reliant on our technology. Hiring and firing decisions should not be based on our outputs alone. Instead, our logic layer is designed to support human understanding and surface emotions that might otherwise go unnoticed.

Let’s return to the scenario of the performance review. Never mind basic AI—all humans, and even great managers, miss things during conversations. There’s a lot happening at once, as people process what’s being said, how to respond, and the greater context of the situation. These days, many exchanges also occur virtually or via video, adding more distractions while shared context is stripped away.

While we would never claim that our models understand humans better than their fellow humans, we believe we can offer an added layer to help managers capture and interpret behavioral signals that might otherwise get lost, providing greater visibility into how a conversation is unfolding.

Our model can track patterns moment to moment, picking up, for example, a shift in engagement, an instance when something didn’t land, or a change in how someone is behaving. The model won’t tell the manager what these moments mean or what to do about them; it simply makes them easier to see and follow up.

Human-context AI is at an early stage. The use cases, the adoption patterns, and the actual impact are all still evolving. At the same time, emotion-sensing systems are quickly being incorporated into real products and platforms. And without context—without knowing why people feel the way they do—AI risks misunderstanding us in critical moments.

Artificial intelligence is the transformative, strategic technology of the early 21st century. It is significantly reshaping practically every aspect of our lives, including in ways that probably no one anticipated. Its rate of adoption and impact have been unprecedented when compared with other technologies.

AI as a distinct field was formally established in 1956 at the Dartmouth Summer Research Project on Artificial Intelligence, proposed by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon. In their August 1955 proposal for the research project, the scientists introduced the term artificial intelligence and envisioned machines capable of simulating human intelligence.

AI is the “science of making machines do things that would require intelligence if done by men,” as defined by Minsky. The professor received the ACM Turing Award, which is often called the “Nobel Prize in computing.”

Since AI’s humble beginnings 70 years ago, it has evolved significantly in its capabilities, gained prominence, and earned widespread adoption across many areas including business, education, finance, health care, industry, and the military.

IEEE’s contributions to the progress and adoption of AI throughout its journey are substantial and multifaceted.

As we celebrate AI’s 70th birthday, understanding its history, current status, limitations, and concerns is key to harnessing it for good.

The technology’s roller-coaster evolution

Although AI emerged as a distinct field in 1956, its intellectual roots extend back further. The ideas and theories that underpin AI predate modern computers such as the ENIAC, unveiled in 1946.

In 1943 Warren Sturgis McCulloch, a neurophysiologist and cybernetician, and Walter Pitts, a logician working in computational neuroscience, were inspired by the human brain. The two devised mathematical models of artificial neurons, demonstrating that artificial neural networks could perform logical computation.

Frank Rosenblatt, a Cornell psychologist, later advanced those ideas by developing the perceptron, an early neural network that laid the foundation for modern machine learning and deep learning.

A major milestone came in 1950, when celebrated computer scientist Alan Turing posed the question, “Can machines think?” In his 1950 landmark paper “Computing Machinery and Intelligence,” published in Mind, he explored the nature of machine intelligence. He introduced the “imitation game,” later known as the Turing test, as a practical means of evaluating it. The test remains an influential concept in AI and the philosophy of intelligence, as I discussed in my article “The Turing Test at 75: Its Legacy and Future Prospects,” published in IEEE Intelligent Systems.

Claude Shannon, recognized as the father of information theory, explored the potential of machines for complex reasoning tasks in his 1950 article “Programming a Computer for Playing Chess,” published in Philosophical Magazine.

In 1956 AI became a formal discipline, inspiring scientists to explore and advance it further. John McCarthy developed Lisp in 1958, and it became the dominant programming language for AI research and development. In 1959 Arthur Lee Samuel, a computer science professor at Stanford, introduced the term machine learning to describe programs that could improve their performance through experience.

In the early 1980s, renewed enthusiasm and government funding fueled the development of symbolic AI, a rule-based expert system (also known as a knowledge-based system) that encodes domain-specific knowledge as sets of rules. A notable example was MYCIN, designed to diagnose infectious diseases.

Although successful in limited domains, expert systems’ inherent limitations have restricted their broader adoption. Expert refers to a computer system that mimics human experts in a specific domain. It was popular in the early days of AI, and subsequently disappeared with advances in AI such as neural networks and machine learning.

AI’s journey was marked by periods of soaring expectations and disappointing progress, known as “AI winters,” during which funding, interest, and confidence declined. Analyses of the episodes revealed recurring causes and insightful lessons for the field.

A new phase of growth—often described as “AI spring”—emerged in the 2010s with advances in deep learning, the rise of large language models, the transformer architecture, and generative AI (GenAI).

“The imperative before us today is not only to advance AI’s capabilities but also to ensure that it remains human-centered, trustworthy, ethical, and dedicated to enhancing human well-being and societal progress.”

Unlike earlier approaches that processed information sequentially, a transformer model analyzes an entire sequence of text or audio, assessing the importance of each word or component relative to others, enabling dramatic advancements in GenAI and its applications.

Ashish Vaswani, a former computer scientist at Google, and his colleagues at Google Brain introduced the transformer architecture that underpins today’s generative AI systems in their influential 2017 paper “Attention Is All You Need.” Vaswani and Sam Altman—chief executive of OpenAI, which offers ChatGPT—are widely regarded as the masterminds behind the GenAI revolution.

AI reached new heights with the public release of ChatGPT in 2022, followed quickly by a wave of chatbots and generative AI tools that accelerated global interest.

More recently, the rise of agentic AI systems capable of increasingly autonomous operation has expanded AI’s capabilities and impact.

AI’s 70-year journey reflects an extraordinary interplay of vision, experimentation, setbacks, innovation, and impact.

For further information and diverse perspectives on AI history, check out my curated collection of articles.

Strengths and promises

AI’s pragmatic strength lies in its ability to process information, recognize patterns, and perform cognitive tasks at an unprecedented speed and scale. It can analyze vast amounts of data, extract insights, and identify trends or anomalies that are difficult for humans to detect. The programs can automate routine tasks and repetitive knowledge work, improve productivity, and reduce costs.

Chatbots and other forms of GenAI can answer queries and rapidly create text, images, videos, music, software code, educational materials, and other content on the fly in response to a user’s prompts, accelerating information-gathering, innovation, and decision-making. AI summarizes, translates, and rephrases text effectively and can assist in idea generation. It also facilitates natural-language interactions, making technology more accessible to nonexperts and the diverse global community. Its multimodal capabilities enhance its usefulness across diverse domains. Additionally, it can serve as a powerful collaborator, augmenting creativity and problem-solving capacity rather than replacing human intelligence.

AI is transitioning from standalone tools to autonomous, goal-driven systems. Agentic AI systems that can plan, act, and adapt with minimal human oversight are on the rise, enabling large-scale impact.

The 400-page AI Index 2026, published by the Stanford Institute for Human-Centered AI, reveals the technology’s enhanced capabilities and unprecedented adoption rates, outpacing those of the telephone, the television, the personal computer, and the Internet.

For a deep exposition on the current state of AI, read this analysis from IEEE Spectrum, which also published the “Great AI Reckoning” special report.

Weaknesses and concerns

Along with its benefits, AI presents significant risks and concerns. They include biased, discriminatory, and harmful responses; a lack of transparency and explainability in decision-making; privacy violations from data collected for AI training; and cybersecurity vulnerabilities including AI-powered attacks.

AI systems can hallucinate, generating confident but incorrect or fabricated information. Moreover, AI can facilitate and amplify the spread of misinformation, deepfakes, and manipulated content, undermining public trust and driving the algorithmic manipulation of public opinion. The flattering, people-pleasing, or affirming behavior known as AI sycophancy can be harmful as well.

Overreliance on AI could erode human judgment, critical thinking, and decision-making skills. And autonomous systems can make errors with serious consequences in critical domains including defense, health care, and transportation.

The technology’s development and deployment, therefore, must be guided by informed understanding, sound judgment, and responsible governance. In assessing AI’s suitability for any application, its capabilities, advantages, limitations, and risks must be carefully and holistically considered.

IEEE’s contributions

IEEE has not merely documented and disseminated AI’s progress. It has actively fostered, standardized, and guided it toward further advances and responsible use for the benefit of humanity. IEEE maintains a hub for information on its AI activities that is a valuable resource for researchers, developers, regulators, and users.

IEEE publishes 11 AI-focused journals that advance the frontiers of knowledge, including IEEE Intelligent Systems. In its AI at 70 commemorative issue, Intelligent Systems identified the 10 most influential AI articles published since 2000. The magazine, produced by the IEEE Computer Society, has inducted 10 pioneers into its AI Hall of Fame, honoring their contributions and impact on technology and society.

To foster AI research and development, since 2006, the magazine has recognized the field’s rising stars through its AI’s 10 to Watch awards. The biennial awards spotlight outstanding contributions of young researchers and professionals. Nominations for this year’s awards are open until 1 July.

Since the early days of AI, the IEEE Computer, Computational Intelligence, and Systems, Man, and Cybernetics societies have been among those that have fostered AI research and practice. The Computer Society offers a guide to becoming an AI developer.

IEEE and its societies sponsor more than 100 AI conferences annually. The conference archives are available in the IEEE Xplore Digital Library.

The IEEE Learning Network offers more than 200 courses across AI-related areas.

The IEEE Standards Association has developed more than 100 AI-related standards. Its CertifAIEd program promotes ethical design and deployment of autonomous intelligent systems.

The Institute has featured several IEEE members who have developed AI-driven applications, such as Abhishek Appaji, who has created tools to help detect psychiatric disorders.

Shaping AI’s future

The history of AI helps us understand the motivations behind developments and inspires and guides us toward the next phase of the technology’s innovation and revolution. AI’s trajectory is bound to be shaped by the collective choices we make now and in the future.

As Turing wrote in his 1950 landmark article, “We can only see a short distance ahead, but we can see plenty there that needs to be done.”

The imperative before us today is not only to advance AI’s capabilities but also to ensure that it remains human-centered, trustworthy, ethical, and dedicated to enhancing human well-being and societal progress.

Salome Mikadze-Struk is no stranger to adversity. The daughter of refugees, she built a software-development business as an undergraduate at the height of the COVID-19 pandemic and kept it running despite the outbreak of war in her native Ukraine. Now, she’s drawing on her experiences to mentor tech-startup founders and speak publicly about the importance of resilience in entrepreneurship.

Mikadze-Struk was studying at Georgetown University, in Washington, D.C., when COVID-19 struck. Classes went online, and she moved back to Ukraine. In the midst of that disruption she saw an opportunity to develop her business idea, called Movadex, by tapping Ukraine’s pool of talented young engineers. Then Russia invaded in early 2022, during her final semester. Taking online classes from bomb shelters and helping employees evacuate to safer parts of the country was surreal, she says, but the team kept the company afloat and she graduated later that year.

In 2023, Mikadze-Struk took a hiatus from her business to pursue an MBA at Stanford University, which she completed this year. In her precious spare time she’s been advising startups and giving talks, using her unique perspective to promote the need for resilience in entrepreneurship—something she thinks is increasingly important in the software industry as AI coding tools upend old business models.

“You need to be okay with risk, you need to be resilient. You need to be okay with disruption and okay with uncertainty,” she says, “because this is inevitably going to be part of this industry for the foreseeable future.”

An Early Focus on Education

Mikadze-Struk’s parents had settled in Ukraine after fleeing conflict in the Abkhazia region of Georgia in the early 1990s. “They left everything behind,” she says. “You can look on Google Maps and zoom in on where their houses were and it’s all rubble.”

Despite this backstory, Mikadze-Struk says she and her sister had a conventional middle-class upbringing in Kyiv. Her father ran a small shop and her mother was a stay-at-home mom. Her parents placed an emphasis on education and encouraged her to study hard and take part in extracurricular programs such as Ukraine’s Junior Academy of Sciences, which introduces students to research.

“They weren’t rich, so they knew that our way to make it in life was not through investments, but through merit-based accomplishments,” she says.

When Mikadze-Struk was 14, her family discovered the newly launched Ukraine Global Scholars program, a nonprofit that helps talented students secure scholarships abroad. The program helped her win a full scholarship to the Emma Willard School, a private girl’s school in Troy, N.Y.

Discovering Tech

After graduating high school in 2018, Mikadze-Struk was accepted to Georgetown to study business administration. But it was outside the classroom that her career direction began to take shape. She won a startup competition with a medical device she had developed for a school project and, while the business idea didn’t go anywhere, it sparked an interest in entrepreneurship.

Ukraine’s software industry was booming, and she began attending startup events and competitions in her home country the summer before starting college. There she met her eventual cofounder Nor Newman.

Despite both being just 18, they saw a gap in the market. The pair noticed many founders had strong ideas but lacked the technical expertise to realize them, while talented engineering students often struggled to gain real-world experience. Newman had begun informally connecting startups with his college friends, but the pair soon saw commercial potential. “We realized we could actually create our own startup studio and help startups as a team, versus just connecting people,” says Mikadze-Struk.

Then, when the COVID-19 pandemic struck in early 2020, halfway through her sophomore year, it brought both disruption and opportunity for Newman and Mikadze-Struk. While travel restrictions and lockdowns made life complicated, there was also a surge of companies looking to move their business online. “COVID really skyrocketed everything we were doing,” she says.

Sensing an opportunity, Mikadze-Struk and Newman incorporated Movadex in Ukraine in early 2020. From the start, they decided to focus on not only providing engineering talent, but also helping startups with product development. Many times, says Mikadze-Struk, a founder’s vision for the software doesn’t line up with what users actually want. “What really helped us grow is not just the engineering or quality of code, but rather a holistic approach to creating a product and actually getting into the brain of the user,” she says.

Navigating Adversity

Back in Ukraine, Mikadze-Struk had to juggle this booming business with studying remotely—taking classes at night and working during the day. It was exhausting, she says, but it also allowed her to immediately apply what she learned in business classes to building her startup.

Having successfully navigated the pandemic, Mikadze-Struk was dealt another wild card. In early 2022, Russia invaded Ukraine and her life was again turned upside down. It was particularly traumatic for her family, having already been forced from their home in Georgia once by war.

In 2023, Mikadze-Struk took an extended leave from her company to pursue an MBA at Stanford.Christie Hemm Klok

“For my parents to experience their daughters going through all the same things they had gone through was really heartbreaking,” she says. “But at the same time, because I’d heard so much about their story of resilience I had power in me to not fully break down.”

On the day of the invasion the founders told employees to take the day off and emailed clients to warn of potential disruptions. The next couple of days were spent checking on staff and evacuating as many as possible to their headquarters in Lviv, in Western Ukraine.

By the following Monday the business was back up and running. Soon afterward, they partnered with the Lviv IT Cluster business association’s nonprofit arm to help resettle refugees from the eastern part of Ukraine, where strikes were focused, and offer job placements. Throughout this period, Mikadze-Struk was also completing her final year at Georgetown remotely. “Half of my senior year was actually spent in bomb shelters,” she says.

Promoting Resilience in Entrepreneurship

That summer, Mikadze-Struk graduated with a bachelor’s degree in business administration and learned she had been accepted onto Stanford University’s MBA program. In 2023, she took an extended leave from Movadex and moved to California. She also gave birth to her daughter in 2024.

Balancing studies and parenthood was already a full-time job, but she continued to engage with the startup ecosystem by volunteering as a startup mentor and public speaker. Now, after graduating from Stanford, she is stepping back into a more active leadership role at Movadex, where she hopes to drive the company’s expansion into the United States. She also wants to develop a stronger focus on helping customers understand and implement AI in their businesses.

While AI is undeniably disrupting the tech industry, Mikadze-Struk, now an IEEE Senior Member, is fundamentally optimistic about its impact. “The way AI democratized access to building software and to prototyping…is just mind blowing,” she says.

But it will require a significant shift in mind-set for engineers, especially junior developers hunting for jobs. They need to “fall in love with AI” and embrace it as a powerful copilot, she says. As these tools increasingly take over the nuts-and-bolts work of coding, engineers also need to nurture higher-level skills like systems thinking and architectural design.

Perhaps most importantly, given the rapid pace at which the technology is evolving, engineers need to nurture their adaptability and resilience. “It’s both exciting and scary, because you don’t know what tomorrow will bring.”

Large language models have moved out of the research lab and into engineers’ daily workflow. LLMs serve as reasoning engines that can orchestrate complex tasks including identifying vulnerabilities in source code and transforming fragmented project discussions into rigorous technical specifications.

While the general public uses AI tools to write email and plan vacations, technical professionals use LLMs as core architectural elements that are fundamentally changing how digital infrastructures are built and maintained. As the AI models move into mainstream engineering practice, the demand for technical expertise is rising.

The LLM technology market is expected to grow by about 33 percent every year through 2030, according to MarketsandMarkets. The rapid expansion suggests that proficiency in implementing and securing the models is transitioning from a niche into a core requirement for technologists.

More than just a better search engine

To use LLMs effectively, technical professionals must move beyond treating them as conversational robots. At a fundamental level, the AI systems are built on the transformer architecture, a framework that replaced the older method of processing data in a fixed, sequential order. Unlike earlier models that analyzed information one step at a time, transformers use self-attention mechanisms to ingest vast datasets simultaneously.

For technical professionals, LLMs are core architectural elements that are fundamentally changing how digital infrastructures are built and maintained.

Relying on such LLMs without understanding their internal logic creates a significant reliability risk. To build tools that work consistently, developers must understand the core principles that govern how the models process information and generate results. By mastering how a model processes information and how its internal settings influence the result, developers can move away from a trial-and-error approach toward a more precise one to ensure the AI tool handles complex data reliably.

Four ways LLMs are changing jobs

Here are areas that integrate large language models.

Moving past basic prompts. Developers are using application program interfaces (APIs) to connect LLMs directly to their databases and software tools. Employing the APIs allows AI to perform work such as executing code or searching through internal repositories.

Fixing the “hallucination” problem. LLMs are at risk of hallucinations, which are generated facts or code that looks correct but actually is wrong or broken. To fix the problem, retrieval-augmented generation (RAG) forces AI to look up information in a trusted source such as a company’s database.

Prioritizing data security. When using AI with proprietary code, security is a major concern. Engineers must learn how to set up “private” instances of the models to ensure that sensitive company data stays within a secure cloud environment and is not used to train public versions.

The future of collaboration. By automating repetitive coding tasks and summarizing thousands of pages of documentation, LLMs let engineers spend more time on high-level designs and solving important issues.

Online course program helps with mastering the tech

The gap between people who use AI and those who understand how to build with it is growing wider. To help technical professionals stay ahead, IEEE offers a five-course online program, Large Language Models Demystified, available through the IEEE Learning Network.

The program, developed by IEEE Educational Activities in partnership with the IEEE Computer Society, is built for people who want to understand the “how” and the “why” behind the technology. Rather than just teaching basic prompting, the curriculum dives into the engineering behind generative AI, including:

• Evolution, impact, and hands-on exercises: the shift from statistical methods to modern transformers, including hands-on model optimization.
• Understanding transformer architectures: the mathematical core of self-attention and positional encoding, implemented in NumPy and Python.
• Architectural analysis and implementation: advanced LLM design with practical model-building exercises.
• Training and modeling with PyTorch: end-to-end pipelines in PyTorch, leveraging parameter-efficient techniques such as low-rank adaptation and quantization.
• Optimization, alignment, and deployment: performance scaling, reinforcement learning from human feedback (RLHF), group-relative policy optimization, RAG, and agentic AI.

Upon completion of the program, participants earn professional development credits and a digital badge from IEEE to verify their expertise.

Enroll in the course program on the IEEE Learning Network.

Organizations looking to prepare their teams to work on LLMs can connect with an IEEE content specialist to discuss group enrollment and tailored training paths.

In 2018, Amazon brought me in as the lead UX Sound Designer for Astro, its first consumer home robot. Astro used cameras and other sensors to map and navigate your home and workplace, and could proactively patrol, check up on loved ones, and transport small items using its built-in cargo bin. While there was a well-defined feature set and form factor, initially there was no character direction. In fact, even before Astro had a name, there were two main questions—was it simply Alexa on wheels, or was it a robot with its own character?

The Astro team was divided. One option was to focus on Alexa, and treat the mobile robot simply as an added utility. Along with the majority of the UX team, I argued for Astro to not focus on Alexa. Our belief was that a thing that moves through your home and turns toward you with intent can never be just an appliance. People would ascribe character to it whether we wanted them to or not, and so the only question was whether we shaped that character or let it happen by accident.

Ultimately, Astro became Astro rather than Alexa, and user testing backed up our decision. People didn’t see the robot as Alexa. They saw it as its own character, and that’s what they wanted it to be. Alexa on the device felt somewhat strange and creepy, but building Astro its own voice was too slow and expensive in 2018. So, we settled on Alexa as a supporting character that handled any actual talking, while Astro was the main character, communicating as much as it could without words, through sound, motion, and facial expressions.

I had been brought on to the Astro team to define the robot’s sound design language and voice. But there was no one to flesh out the robot’s actual character. You cannot make a single real decision about a character without defining it first. Every choice about how Astro moved, sounded, paused, or reacted was a character choice, and those choices required all disciplines working together. As sound lead, I was weaving together sound, motion, and character, and how they played together inside each story moment. The animators, who programmed Astro’s motion and facial expressions, were extraordinary at what they did, but the emotional arc they were animating came from the sound (and therefore character) work first. So I stepped into that role, which is where my real work started. What I learned about building character for robots applies to nearly everything being built in embodied AI right now.

Character Is a Design System

Developing a character for Astro meant answering questions that had never been asked about a product at Amazon: What is the emotional range of this robot’s baseline state? How does this robot communicate uncertainty without eroding trust? Where is the line between being expressive and annoying? What are the vulnerabilities of this device’s character?

These are design questions. They have real answers, and every team working on the product has to build from them. For example, Astro’s emotional range was designed to be relatively small at first. We never wanted Astro to get too sad or too angry. It could play sad, but would snap out of it quickly and end the reaction on a high note to keep things positive.

Character leaks out of every seam and can create a disjointed experience if not defined correctly. Even if it’s just animation timing that’s slightly off, or a response that’s technically correct but contextually tone-deaf, users feel every one of these inconsistencies, even if they can’t name them. Watch what happens at the beginning and end of this Sing sequence:

Astro goes from nothing, into the emotional moment, and then lands back on nothing. No buildup, no cooldown, no sense that the feeling came from somewhere or had anywhere to go. I pushed hard for better character stitching, the transitions in and out of expressive moments that make a performance feel continuous rather than assembled, but it never got implemented. The moment itself works. But without the stitching, it reads as a clip playing on a robot rather than coming from within the robot character itself.

Story and Sound at the Beginning

We had decided that Astro would have no spoken dialogue, but it had something that functioned the same way: a vocabulary of sounds, tones, and rhythms that acted as its voice. This vocabulary became the leading output of the character’s personality. The robot’s motion and facial expressions were built around it.

Astro’s wake-up sequence is a great example. Waking wasn’t just a boot animation on the screen; it was an entire performance. Slow and humble at first, the robot oriented itself quietly, then stretched its screen, checked its wheels, and finally, with an upward gesture toward its telescoping mast, it popped it up slightly, and did a little dance of joy. Sound, motion, and eyes hit every beat together in full choreography.

The character’s output in that sequence was first written as a story. Astro is waking up in its new home for the first time. Its main aspiration is to be part of a family, so this is the moment it has been waiting for, this is its purpose. Being the responsible character that it is, it wants to make sure everything is good to go before it introduces itself and starts learning its new home.

This narrative came first because it drove every other decision that we made. After the story was written, sound gave that story a metaphorical voice: the excited tones, the pacing as it checked its wheels, and the bright melodic phrase as Astro looked up at its new family for the first time and introduced itself. Once the sound was laid down, the animation team did their thing with motion and facial expressions, taking cues from the emotional arc the sound had established. Motion didn’t lead—it followed the feeling of the story and the sounds, the same way an animator follows a recorded vocal take.

That wake-up sequence became one of the most-discussed moments in early user testing. People described it as “alive.” What they were responding to wasn’t any single element. It was all three channels (sound, motion, and facial expressions) expressing the same defined character in harmony.

Context Is Where Character Becomes Real

The most compelling characters are defined not by a fixed disposition but by how they respond to their environments and the people in them. They’re still recognizably themselves even as they adapt. This is what I call contextual character. A robot living in a home doesn’t occupy a single emotional state. It moves through rooms with different energy, encounters people in different moods, operates at different times of day, and responds to an endless range of social situations it was never explicitly designed for.

We got close to a contextual character output with Astro’s sound. When a specific piece of environmental context was fed in, the system adapted beautifully, and Astro felt completely alive. But every state like this was still a prediction we made by hand—a situation we had to imagine in advance and design a response for. A random home throws more situations at a robot than anyone can possibly predict, so there was always a longer tail of moments the system was never prepared for.

The difference between a product people describe as “smart” and one they describe as “aware” often comes down to this. Smartness is capability. Awareness is context. Presence is character. And character is always in reaction to the people around it, to its environment, to its own evolving state. That’s what makes it feel like something is emotionally present with you.

This is where AI changes the game for character design in ways that go well beyond what was possible with Astro. AI-driven adaptation doesn’t require the contextual predictions that we relied on. It learns the specific rhythms, preferences, and emotional context of the people it lives and works with. The character doesn’t just respond to context. It grows into it.

What Industry Is Missing

The character and soul of the impending wave of embodied AI products appears to almost always be an afterthought. And character defined late is character defined by default. It becomes the sum of a thousand small decisions made by different people thinking about anything but character. People project character onto devices whether you plan for it or not, especially if those devices move—a robot that moves is already a character. If nobody has designed this character, the result will be products that feel like nothing, or worse, feel confusing and not trustworthy. Technically impressive, but lifeless.

We did not get this fully right with Astro. So many things were moving in parallel that character was rarely treated as a utility, and it made sense why. When you are building a first-of-its-kind product, the things that are the loudest are the ones that break, the deadlines, the costs, the features a customer can point to on a box. Character is quieter than all of that. It’s easy to assume it can come later. On a team as large as the Amazon Astro team, it’s lucky to get any idea onto the road map when it is competing with a hundred others that all feel more urgent in the moment. None of this came from people not caring. It came from character being the kind of thing that is hard to prioritize until you see what its absence costs you.

My Asks to Product Leaders

If you are building a product that will share physical or conversational space with people, three things are worth considering:

Define character before you define interactions. You need a defensible character with enough emotional logic to answer hard questions consistently. Find answers to character questions early, and have every discipline build from the same foundation.

Build story and sound into the character pipeline, not the production pipeline. Story and sound developed alongside character definition has the chance to inform motion, expression, and interaction logic. This requires a different kind of collaboration, and a different kind of hire.

Design for adaptation, not just consistency. A consistent character is necessary, but the products that will matter most in people’s lives are the ones that deepen through use. The infrastructure to support that is more and more accessible, but the design thinking to take advantage of it is still rare.

An expanded version of this story is available on Medium.

The rapid evolution of the global engineering landscape requires continuous education. For one week in April, the IEEE community focuses on its educational frameworks. IEEE Education Week, which just concluded its fifth year, provided a comprehensive overview of the resources available to professionals and students.

From 11 to 19 April, the organization supplied a variety of live and virtual events, online resources, and promotions that champion the cycle of lifelong learning.

IEEE President Mary Ellen Randall kicked off the week with the keynote: “Inspiring Tomorrow’s Innovators: How IEEE Educational Resources Can Open Pathways Into STEM.” The event served as a central point for programs that run throughout the year.

“Education Week allows different units to share resources with members and the public, covering everything from preuniversity programs to advanced professional training,” says Jamie Moesch, managing director of IEEE Educational Activities.

Coordination across the organization

The event relied on the cooperation of 120 IEEE partners. Involved organizational units included the IEEE Communications Society, the IEEE Education Society, and chapters and sections from around the world, including in Brazil, Colombia, and India. They produced 114 events, 23 resources, and 11 special offers.

“These collaborations help members remain current in a changing technological environment,” says Timothy Kurzweg, vice president of IEEE Educational Activities. “The goal is to provide accessible tools that assist members in both their own professional development and their efforts to mentor new engineers.”

“The week allows different units to share resources with members and the public, covering everything from preuniversity programs to advanced professional training.” —Jamie Moesch, managing director of IEEE Educational Activities

The participation metrics reflect a broad geographic interest. The IEEE Education Week website recorded more than 4,770 visitors, with primary engagement coming from India, Nigeria, and the United States. Nearly 240 digital badges were issued to people who completed educational quizzes.

To encourage participation, organizers enlisted 72 volunteer ambassadors to promote the week’s activities across their local networks and share key resources on social media.

Available educational tools

Here are a few of the virtual events held during Education Week—most of which are available on demand:

• Celebrating Excellence: The EPICS in IEEE Contributor Awards and Service Learning Showcase.
• Classroom to Startup: Uniting Academia and Industry.
• IEEE’s Role in Shaping AI-Ready Engineering Education Globally.
• Leveraging IEEE Standards to Enhance Engineering Service Learning Projects (EPICS in IEEE).
• Mastering the Modern Job Market: The Power of IEEE Microcredentials.
• TryEngineering Volunteers Making an Impact in STEM.

The Education Week website highlights resources and offers shared by IEEE organizational units, including:

• A half-off discount for members on IEEE e-learning courses. The catalog covers such topics as computing, power and energy, and telecommunications.
• IEEE Communications Society on-demand webinars. Learn the latest trends and innovations.
• IEEE Women in Engineering career-focused, upskill, and reskill webinars. The presentations cover a variety of topics including agentic AI, leadership, and robots.
• IEEE Innovation at Work. The e-newsletter covers emerging technologies, education, and training for technical professionals.
• IEEE Learning Network. Hundreds of continuing education courses, all in one place.
• IEEE TryEngineering lesson plans. The easy-to-use, engaging activities and plans help teach engineering concepts to preuniversity students.
• IEEE TryEngineering collections. The lesson plans and multimedia resources, developed with partners and IEEE technical societies, are designed to introduce technical topics and deepen student understanding.

Individuals who were unable to attend the live sessions can find the archived content on the IEEE Education Week website.

The website also accepts donations for education-related funds managed by the IEEE Foundation.

Updates and technical resources continue to be shared through the #EducationAtIEEE hashtag on social media channels.

Planning for IEEE Education Week 2027, scheduled for 3 to 11 April, is underway.

This article is crossposted from IEEE Spectrum’s careers newsletter. Sign up now to get insider tips, expert advice, and practical strategies, written in partnership with tech career development company Parsity and delivered to your inbox for free!

I’ve sat on both sides of the interview table several times over the past decade. You might be surprised to hear that I’ve often been just as nervous interviewing candidates as I was when being interviewed!

Nearly all the interview advice out there is about the candidate’s side, but understanding the other side can also help you prepare. Let me show you what I’ve seen firsthand, and what I’d bet is happening at the company you just interviewed with.

If you recently got rejected after an interview, this might explain what actually happened.

One caveat, because I’ve been on the receiving end of this: A couple of my recent interviews were run entirely by AI. These were screening rounds, but a growing share of job seekers now report being interviewed by a bot somewhere in the process. Everything below assumes you reached a person.

Most teams have no standard prep

You might assume companies train people to run interviews. Many don’t.

In practice, your interviewers may be much less prepared than it seems. Their prep might look like this: “Here’s a rubric from three years ago, figure it out.” Or: “Let’s grab a conference room between meetings and decide what to ask.”

The questions are often whatever the interviewer personally studied when they were job hunting. These days, they may be generated with an LLM the morning of.

Then the panel negotiates. One person wants to quiz candidates on data structures and algorithms for a role in which they design websites. Another insists system design is essential for a junior level position. People default to what was done to them and assume it’s normal because it was normal to them.

What’s normal to the spider is chaos to the fly.

“Scoring” that isn’t really scoring

After an interview, some processes I was part of had one simple scale to score candidates: yes, no, strong yes, strong no.

The result is predictable. Like the candidate? Strong yes. They rubbed you the wrong way but answered everything correctly? Somehow a soft yes at best.

Structured scoring with defined criteria measurably reduces this. The research backs it, and the rare times I saw it used well, it changed my own assessments. Yet many teams I worked on never used this approach.

Prestige bias and politics

Even with a strong scoring system, bias and office politics can change the outcome.

For instance, I once interviewed someone I was strongly against hiring. It was clear they didn’t know what they were doing, and they’d be running critical infrastructure. I gave a strong no with objective reasons, scoring notes, specific examples from the technical round.

Leadership pulled me into a meeting right after and asked why. I walked them through my notes.

What I didn’t know: Several of them already knew the candidate personally. They liked them. They wanted them hired. I said the decision was theirs, my assessment hadn’t changed, and wished them luck.

I’ve also watched a strong resume short-circuit an entire loop. The team saw a top-tier company name, skipped the standard technical rounds, lobbed a few softballs, and basically welcomed the candidate in.

But once this engineer got started, it turned out to be a poor fit. And it wasn’t the candidate’s fault. They were set up for failure, because nobody checked whether this person could do this job at this company.

In both cases, it didn’t work out.

What you can actually control

You could read all this and decide the system is broken or rigged.

The broken part is fair. The rigged part isn’t. People who are genuinely good at interviewing pass more often. It’s messy, but it’s not a lottery.

You can’t fight bias, politics, or a sloppy process. That’s like being mad at the weather. You can only play the two cards you’re dealt: your technical ability and your behavioral presence.

Most candidates obsess over the technical side and forget the behavioral rounds exist. But product managers, designers, and cross-functional leads—people with zero technical background—will judge you entirely on whether you can tell a clear story and seem like someone worth working with. If you’re unlikeable in the room, you’ve roughly halved your odds at every stage.

So here’s the unglamorous advice that actually works: put yourself on camera.

Talk through a project you led, a mistake you made, a hard problem you solved. Record it. Watch it back. Cringe. Do it again.

Think out loud, under pressure, with another human watching.

If you keep failing interviews, the fix isn’t always more technical prep. It’s getting better at being in a room with other people who are potentially more nervous, less prepared, and more biased than you ever imagined.

The process is broken. You can still win.

—Brian

NSF Experiments With New Kind of Science Funding

A new initiative from the U.S. National Science Foundation plans to distribute $1.5 billion of funding over 10 years to independent research organizations, which it calls “X-Labs.” The program is meant to support work being done outside of academic institutions, starting with two areas: scientific instruments for sensing and imaging, and interconnects and integrated photonics for quantum systems.

Read more here.

7 Ways New Engineers Can Flourish in the Age of AI

We’ve said it before, and we’ll say it again: AI is changing the engineering profession. So how can you stay in demand as the field’s tools evolve? A senior engineering manager at Walmart Global Tech offers seven quick tips.

Read more here.

Collection: Career Advice for Engineers, From Engineers

For even more expert tips, check out the new career advice collection from The Institute. These articles feature guidance written by working engineers, meant to help those in all stages of their careers stay at the forefront of their profession. Discover tips for technical presentations, dive into a specific career path like cybersecurity consulting, and more.

Read more here.

Musicians are accustomed to getting paid each time their creative work is used. Across vinyl/CD sales, streams, radio, cover versions, and those numerous niches like karaoke, there are agreements in place about what “use” means. Underlying this is a simple economic principle: The more something is used, the more money it makes.

Generative AI has complicated the definition of use. On the one hand, you could argue that the use of a piece of musical training data happens just once, at the point of training. On the other hand, creators would be right to complain that the creative essence of their work lives on in the structure of the model, used every time the model produces an output.

Now, companies like Sureel and SoundVerse are working to re-create the essential economic principle that motivates creativity in an era of AI. Such initiatives aim to turn the generative AI industry from one guilty of “the biggest act of copyright theft in history” into one that coexists harmoniously with hardworking artists.

Music Royalties for the AI era

Sureel, a startup Warner Music Group just acquired, has partnered with the Swedish copyright agency STIM to explore the potential for music creators to get paid when their music is used to train generative AI tools. Sureel’s software labels online media, such as a music file, with instructions determined by the owner. The instructions specify whether an AI company may use the media freely in training, limit its influence in any given training set, or avoid it altogether. The software then tracks how the AI company uses the media in training and sets licensing fees accordingly.

Meanwhile, the founders of the AI music company SoundVerse “[reject] one-time royalty buyouts as insufficient and [advocate] for ongoing participation of artists in the AI lifecycle,” they wrote in a 2025 white paper. They argue that each time a generative AI system produces an output, certain pieces of training data play a greater role than others. If the system outputs music resembling jazz, the jazz in the training set has arguably contributed more than, say, the folk music. You can therefore differentially reward each piece of training data for each output.

Sureel’s copresident Benji Rogers told me, “Attribution isn’t about re-creating the old economics. It’s about measuring, for the first time, the thing the old economics only approximated.”

Such influence attribution needs to do more than superficially measure how similar a training data point is to the AI output. The challenge is to attribute causality, or a relationship between the training data and the trained AI, Sureel CEO Tamay Aykut says.

Even if the AI industry achieved that, however, it might encourage people to create music designed to maximize training-data royalties. While all creative markets lead to new incentives (music streaming, for example, has driven songs to have shorter intros), the industry could do without another economic structure that is easily gamed, in which someone’s reverse-engineered pastiche diverts royalties away from original works of creative expression.

RELATED: Generative AI Has a Visual Plagiarism Problem

Inferring the influence of a particular piece of music on a generated piece of music, if a well-defined problem at all, may involve more advanced information theoretic principles, or modeling the actual historical role and impact of individual works. Aykut proposes that in carefully designed attribution systems, more unusual and unpolished musical works could even have more inherent value than radio standards.

Simon Gozzi, head of business development at STIM, says the company is in the process of seeing how Sureel’s attribution reports could underlie licensing agreements between musicians and AI companies. Could generative-AI attribution strategies not only sustain the economic logic that “popularity pays,” but also motivate musical experimentation and diversity? It’s a compelling concept when public sentiment rightly fears generative AI’s threat to cultural vibrancy, pushing power toward tech companies, deskilling creative workers, shrinking revenue in the creative sector, and filling the internet with slop. “Attribution is one of the few credible tools we have,” Rogers says.

There’s a window of opportunity to debate and establish approaches to paying for AI training data that serve a vibrant and sustainable creative sector.

The technical problem of training-data attribution is both complex and ill-defined. Just as a simplistic attribution strategy based on measuring similarity might motivate people to reverse-engineer the canonical works of a genre to capture royalties, a more complex attribution strategy based on some information theory of originality might be easily gamed or fail to reward human cultural production.

For creative workers, there’s good reason to fear that even with the best intentions, AI attribution will only compound the baroque and opaque arms races that they are already weary of navigating. Some voices within the music AI sector are also skeptical. Drew Silverstein, president of SourceAudio, says, “Attribution would seem to be the obvious answer, but it’s flawed in AI, so we have to look at other models.” He advocates simple negotiated agreements with an agreed or annually recurring price at the point of training.

Meanwhile, the copyright lawsuits that have dominated the generative AI revolution are beginning to give way to an increasing number of privately negotiated agreements, such as those between Universal, Warner, and major AI companies to work together on training models with copyright consent. Although little is certain, these agreements may have considerable influence over the industry norms that arise.

Right now, there’s a window of opportunity to debate and establish approaches that pay for AI training data while also sustaining a vibrant creative sector. Sophisticated engineering solutions will have a role to play, but they need to take into account the cultural complexity of the challenge, and enable fairness and transparency through good design.

Making AI Training Pay Off

It remains to be seen whether monolithic generative models such as Suno actually have as much credibility as first touted. In many creative applications of AI, there’s a renewed focus on smaller customized models that are tailored for specific human creative expressive needs such as IRCAM’s RAVE model or Jen’s Style Filters. Meanwhile, more mainstream “end user” creative applications may be shifting towards a focus on fan engagement. OpenAI’s sudden dropping of Sora, despite being in negotiations with Disney and Suno’s recent emphasis on building fan-engagement experiences that draw directly on the work of artists, following its deal with Universal, both point to teething troubles in the creative AI sector.

A move to smaller, more targeted models and applications would give more room for creator alliances. For example, collectives of musicians might band together to provide the training data for a smaller custom model, for which revenue splits might be egalitarian or based on other principles of fairness.

The same may possibly be true of hybrid model architectures and structured training regimes where different data sources are used at different points in the training process, as well as retrieval-augmented generation, which mixes context-specific information with training data to improve results. An approach that produces worse results but enables fairer or more transparent paths of attribution may be more successful if it brings creators on board with more lucrative royalty flows and even clear credits.

Also, no matter how sophisticated an attribution algorithm is, it will always be grounded in human decisions, ranging from the wise and the fair to the arbitrary and corrupt. Ask a music industry insider to explain how the percentage split between recording and songwriting royalties is determined, and you’re in for a long answer. At best, the machinery of training data attribution will enable open and informed discussion about what makes our creative and cultural sectors fair and vibrant. At worst, it will conceal already opaque private agreements in complex black boxes.

This is where national policies are vital. Attribution must be “multi-layered and auditable, open to expert and regulatory scrutiny,” Rogers says. Crafting such policies will take expertise from computer science, musicology, law, and economics. AI-competitive governments will be able to boost their cultural and creative sectors by supporting institutions that fulfil this purpose.

Even the most neoliberal economies look beyond markets to sustain cultural expression, whether through public arts funding or measures like local music quotas for radio. As the economic impact of generative AI in the creative sector takes form, taxation, redistribution, and active support of cultural infrastructures may still be the most effective way to support positive social outcomes. Taxing big AI and redistributing that revenue back to the creative workers that contributed to the industry’s wealth is, after all, another “AI attribution strategy.”

On 19 April 2026, the Honor Lightning humanoid robot ran a half-marathon in 50 minutes and 26 seconds, beating the human world record by 7 minutes and the best robot time from 2025 by almost 2 hours.

How did Honor do it? Is there some magical technology or technique that unlocked this performance? How did the company beat the significantly better-known Unitree (which reportedly had to supply its robot with an ice backpack to try and complete the race without overheating)? My doctoral thesis involved building and controlling hopping and running robots, and since then I’ve tried to design and build efficient commercial legged robots, giving me a decent idea of the constraints involved. In this article, we take a look at the fundamental underlying constraints to try and answer these questions.

The Physics of Running

Running consists of alternating phases of a leg pushing against the ground (“stance phase”) and the body flying through the air (“aerial phase”). In the aerial phase, the body falls due to gravity, losing vertical momentum. The leg in stance phase pushes against the ground to redirect the vertical momentum upward, while the other leg swings forward to reposition for the next foothold.

Electric motors use energy to produce torque—the higher the torque, the more energy is lost as heat. Adding a gear train after the motor amplifies its torque and reduces its speed. A large reduction helps with torque production, but since the rotor of the motor itself has to spin faster, it becomes very sluggish at accelerating its output. This is obviously bad for the swing phase described above. These competing effects mean that for a particular motor, there is usually a sweet spot for the gear ratio:

The power consumed by a robot leg is minimized at an optimal gear ratio (30:1 in this example).Avik De/Datawrapper

How Honor Did It

While the Lightning’s motor specifications are not published, the hip and knee motors roughly have a 110-to-150-millimeter outer diameter. For an approximate set of motor parameters, I looked to the ILM115x25 motor due to its relevant size and detailed specifications.

We can use a simple physics model to estimate the power consumption for running at 7 meters per second (the Lightning’s average half-marathon speed) as gear ratio varies:

The light blue curve shows how to pick the optimal gearing (45:1). The dark blue curve shows how much heat will be produced in the knee motor, ~150W for the optimal gearing.Avik De/Datawrapper

We see that the drivetrain is not magical: with a gear ratio chosen for this task (we’ll return to this below), the approximate robot power consumption would be a very reasonable 400 watts.

However, the dissipated knee power ( typically the main thermal limiting factor) is approximately 150 W. This is almost an unavoidable consequence—running at human speeds with a humanoid-size robot will inevitably generate this amount of heat! Over a prolonged period, keeping the motor from overheating would be a challenge, but the Lightning has a trick up its sleeve:

According to Honor, the liquid-cooling pipes penetrate deep into the motors like capillaries. The high-power liquid pump has a heat-exchange flow rate of more than 4 liters per minute. Each of the four drive motors in the lower limbs is equipped with an independent liquid-cooling circuit.

Liquid cooling is not new, but it’s definitely not a commodity. It has shown up in research periodically, and on the commercial side Apptronik tried it for a few of its prototypes but (to my knowledge) does not use it on its main Apollo platform. Basic air-convection-based cooling would not continuously be able to extract 150 W out of the knee motor, and so the cooling technology is a key enabler of this type of performance.

Why Others Couldn’t Compete

Why did Honor’s competitors, including more established and widely shipped humanoids such as from Unitree or Agibot, not compete as well?

We can use the same model to generate an equivalent energetics plot for walking at 1.5 m/s, a much more modest but potentially more common activity for a commercial humanoid robot:

The solid and dashed light blue lines show a running-optimized design, while green lines show a walking-optimized design. The optimal ratio for walking is much lower (30:1 vs. 45:1). However, the power dissipated in the knee motor while running [dark blue] is much higher at 30:1 vs. 45:1—the price to pay for running with a walking-optimized design.Avik De/Datawrapper

The plot adds a new green curve for the walking power, and the optimal gearing is significantly different!

Let’s say you design your robot to excel at the normal walking task and choose the green design with 30:1 gearing. The knee motor power to run a half marathon is over 300 W (red arrow), more than two times what we had with the running-optimized design. It wouldn’t be so surprising to need ice packs!

Conversely, visually following the green curve shows that the running-optimized robot wastes more power for walking. Using larger motors sized for running increases the weight of the robot and wastes power when it is standing or walking. The larger motors also pose practical issues like bumping into objects while operating in homes or factories.

Closing Thoughts

Honor’s half-marathon performance was an impressive engineering effort and result. It didn’t need any magical leaps in technology, but the deployment of the capillary motor cooling solution is a notable advance without which this running pace would have been unsustainable. The cooling, weight optimization, and robustness advances may well be useful for more practical purposes like carrying heavy payloads down the line.

The Honor Lighting robot [right] has much larger motors driving its legs than the Unitree H1 robot, making it a more efficient runner but a less efficient walker.Left: Wei Zhiyang/Zhejiang Daily Press Group/VCG/Getty Images; Right: VCG/Getty Images

However, the Lightning is not as well-suited to other tasks as a robot designed for greater versatility. Engineering is always characterized by trade-offs, and making the correct ones separates good products from great ones. With consistently improving AI language models, this very human skill is becoming the most valuable one an engineer can have.

The news coverage seemed to overly focus on the fact that the human half-marathon record had been broken by a robot. Machines and humans have very different capabilities and constraints, so why should we ever have expected the half-marathon time for a robot and human to be related? As in Deep Blue’s 1997 defeat of Garry Kasparov in chess, where it couldn’t physically move the pieces, the Honor robot’s capabilities are much narrower than a human running elbow to elbow with other runners while visually navigating the course without GPS. Comparing the robot runner to a human runner is just an apples-to-oranges comparison, which only risks diminishing Honor’s engineering achievement on one hand and human athletic achievement on the other.

Nearly 750 million people face hunger today, according to the U.N. World Food Program. And by 2050, global demand for food is expected to increase by 50 percent from 2010 levels, the World Resources Institute says.

A smart agriculture special-issue report recently released by the IEEE Smart Agri-Food Initiative says meeting the demand will require technology to expand food production. The report highlights research, case studies, and new ways of applying technology to inform farmers, engineers, and policymakers.

Leading the initiative is IEEE Fellow John Verboncoeur, chair of the smart-food program and professor of electrical and computer engineering at Michigan State University, in East Lansing.

“Food security is becoming a systems-engineering problem,” Verboncoeur says. “We’re no longer talking only about tractors and irrigation. We’re talking about sensing, communications, computation, automation, and sustainability all working together.”

Although not formally trained as an agriculture scientist, Verboncoeur’s first involvement with smart agriculture was as an undergraduate at University of Florida in 1985-86, where he helped develop an SmartAg aeroponics system for NASA for the International Space Station. It used mist to spray the plants’ roots and lightweight pneumatic structures to hold the vegetation in place.

He has also chaired the executive committee of Michigan State’s SmartAg Initiative since it launched in 2017. He chaired the program’s leading interdisciplinary efforts to apply engineering and digital technologies to farming and food systems.

Verboncoeur connects the shift of using engineering as a force multiplier for farming to lessons learned from the IEEE Smart Village program, which supports projects and organizations bringing electricity and educational and employment opportunities to remote communities. Agriculture, he argues, requires the same systems-level mindset.

“The challenge isn’t just inventing technology,” he says. “It’s making systems practical, affordable, and deployable.”

From digital twins to autonomous harvesting

A central theme across the Smart Agri-Food Systems report is the convergence of automation, data analytics, and sustainability.

One paper, “Smart Agriculture, Precision Agriculture, Digital Twins in Agriculture: Similarities and Differences,” addresses the confusion regarding how researchers and practitioners define and apply the technologies to farming.

The paper was written by Dilan Onat Alakuş, a research assistant in the software engineering department at Kırklareli University, in Türkiye, and Ibrahim Türkoğlu, a software engineering professor at Fırat University, in Elazığ, Türkiye.

Unclear terminology can lead to inefficient investment and poor adoption of the technologies, the two authors say. They note that agricultural methods based on traditional practices and intuition lack a thorough analysis of their environmental and economic impacts.

They describe how three technologies can benefit farmers:

• Smart agriculture systems integrate sensors, artificial intelligence, robotics, and analytics to improve efficiency and sustainability at scale.

• Precision agriculture focuses on location-specific decisions. Farmers use GPS-guided equipment to map fields, deploy drones to monitor crop health, and install field sensors that track soil moisture and nutrient levels in targeted zones. The tools allow farmers to apply water, fertilizer, and pesticides only where needed—which can reduce waste and lessen environmental impact.

• Digital twins create virtual replicas of an agricultural area. The resulting models simulate the farmstead, crops, and irrigation systems, allowing growers to test scenarios and predict outcomes before implementing changes.

The authors emphasize that the categories overlap in practice. A digital twin might draw data from precision agriculture systems and feed recommendations into smart agriculture platforms.

Clearer distinctions help farmers select appropriate tools and avoid unnecessary complexity and costs, they say.

“This study contributed to conscious agricultural practices by differentiating agricultural technologies,” they wrote, adding that clearer definitions can increase productivity.

Smart farming in practice

The report shifts from theory to application in a paper describing bustani, which means my garden in Arabic. The Bustanica project in Saudi Arabia is an automated hydroponic vertical farming system developed by researchers at the Prince Mohammad Bin Fahd University, in Al-Khobar, Saudi Arabia. The “Bustani: A Microcontroller-Based Automated Hydroponic Vertical Farming Solution” paper was written by Hussah Alotaibi, a computer engineer at Saudi Aramco, the country’s national oil company; Abul Bashar, Widad Karsou, and Shehvar Khan, researchers in the university’s computer engineering and computer science department; and Salahudean Tohmeh from the university’s robotics laboratory.

The Bustanica system combines hydroponics with aeroponics, in which plant roots hang in the air and receive nutrients through a misting system. Together, the approaches allow crops to grow in compact indoor environments, using far less water than traditional methods.

The method integrates IoT sensors that continuously monitor water chemistry and reservoir conditions.

The system grows crops in controlled indoor environments. A closed-loop design recirculates water to reduce waste. Sensors measure pH levels, nutrient concentration, and water levels. An Arduino Mega processes the sensor data. A NodeMCU ESP8266—a low-cost, open-source IoT platform—handles Wi-Fi communication and cloud connectivity.

The system sends the data through Google’s Firebase cloud platform, which acts as a real-time bridge between sensors and control systems.

A mobile app lets users monitor and control the system remotely. It displays real-time data on lighting, nutrient levels, and water pump activity. When conditions move outside optimal ranges, automated dosing pumps adjust the levels as needed.

Engineering can’t solve all the world’s problems. But it absolutely has a role to play in helping the world feed itself.” —John Verboncoeur, chair of the IEEE Smart Agri-Food initiative

The system operates as a feedback loop, collecting data, transmitting it to the cloud, analyzing the conditions, and automatically triggering adjustments.

LEDs simulate sunlight. Ultrasonic sensors measure water levels. Electrical conductivity sensors track nutrient concentration. During testing, the system maintained stable environmental conditions and adjusted dosing dynamically as readings changed.

The authors describe the outcome as “a fully functional and automated vertical sustainable farm that creates desirable growing conditions, along with an Android application that provides real-time monitoring and notifications.”

Beyond automation, bustani reflects a broader shift toward merging agriculture with consumer technology and smart-home systems. Future plans include integrating the Amazon Alexa virtual assistant and machine learning tools for plant disease detection and growth analysis.

Robotics and labor challenges

The “Toward an Efficient Tomato Harvesting Robot” paper addresses autonomous harvesting, a long-standing challenge in agricultural robotics. Tomatoes in the field vary widely in size, shape, and ripeness, and they can bruise during handling. The paper was written by IEEE Senior Member Hyoung Il Son—a professor of biosystems engineering and robotics at Chonnam National University in Gwangju, South Korea—and his graduate students Jongpyo Jun, Jeongin Kim, and Jaehwi Seol.

The paper describes how robotics is increasingly being used to target crops once considered too delicate or variable for automation.

The researcher combined 3D machine vision, robotic arms, suction-based grippers, and rotating cutting tools to build a harvesting machine capable of operating in unstructured outdoor environments. The system aims to reduce reliance on manual labor while improving harvesting efficiency and consistency.

Agriculture as a systems problem

Verboncoeur says the developments highlighted in the papers reflect a broad transformation in how engineers view the agricultural industry.

“Agriculture used to be seen primarily as managing the challenges of planting, watering, and fertilizing plants, and using machines to make the process less labor-intensive,” he says. “Now it’s also a data problem, a communications problem, an energy problem, and a resilience problem.”

Another featured paper, “Sustainable and Smart Agriculture: A Holistic Approach,” examines how technology can address environmental and demographic pressures. The paper was written by Surender Singh and Sannihit , researchers at the computer science and engineering and the civil engineering departments at Chandigarh University, in Mohali, India.

Farmers must increase food production while reducing environmental damage from depleting water resources, overapplication of fertilizer, deforestation, and greenhouse gas emissions, the authors say. They describe smart farming as “a revolution in food production” that can allow farmers to generate higher yields from existing resources through connected technologies and data systems.

The authors highlighted the issue of rapid urbanization. By 2050, they report, nearly 70 percent of the global population will live in cities, increasing pressure on food supply chains and distribution systems.

Wireless sensor networks will play a central role in the transformation, the researchers say. The networks use small, connected devices to monitor soil moisture, temperature, humidity, light intensity, and crop conditions. The system transmits the data to cloud platforms, where machine learning models analyze trends and recommend actions.

The authors emphasize that decision support, not automation alone, drives the greatest value of crop harvest. Farmers can integrate the information into crop management strategies to improve productivity while reducing their environmental impact.

They also note increasing collaboration between industry leaders such as Caterpillar, CNH, John Deere, and Kubota and technology companies including Bosch, Google, Intel, and Microsoft. Challenges remain, however, in communication reliability, sensor cost, and scalable data infrastructure, the authors say.

SmartAg beyond the farm

The implications of the tech advances that make farming more efficient extend beyond agriculture. Many of the same technologies—remote sensing, wireless sensor networks, AI analytics, and cloud platforms—support transportation, energy, and industrial systems.

The convergence explains IEEE’s growing involvement. Modern agriculture now combines electronics, communications, computing, and control systems.

Agriculture requires that integration, Verboncoeur says: “The challenge isn’t just inventing technology. It’s making systems practical, affordable, and deployable.”

What’s next for smart agriculture?

The special issue marks an early stage for the IEEE Smart Agri-Food initiative, which plans to develop standards; create structured ways for farmers, researchers, governments, and agribusinesses to work together; and devise deployment strategies for smart systems.

Future research is likely to focus on interoperability between platforms, data sharing, and scalable deployment models. Digital twins are expected to play a larger role as computing power and sensor density increase. Simulating agricultural systems before applying changes in the field will become commonplace, experts predict.

Adoption depends on more than technical capability, though. The central tension moving forward lies between innovation and practicality.

“Farmers face challenges in adopting such technology due to cost, electricity availability, communication infrastructure, and vulnerability of connected devices,” Singh and Sannihit wrote.

Smart agriculture offers improved efficiency, in addition to reducing the inputs of water, fertilizer, and time that would otherwise be spent on tasks machines can handle autonomously. But the benefits matter only if systems function reliably across diverse environments—from industrial farms to small, family-run operations in food-insecure regions.

For IEEE, agriculture now sits within core engineering domains. The stakes extend beyond technology itself, Verboncoeur says.

He adds that: “Food insecurity affects stability, health, education, and economic development. Engineering can’t solve all the world’s problems, but it absolutely has a role to play in helping the world feed itself.”

A half century ago, a scrappy crew at the University of Massachusetts Amherst erected a wind turbine on Orchard Hill, the highest point on campus. It was a frugal production, cobbled together from the rear axle of a Ford truck, a donated generator and microcontroller, a steam pipe, and various handcrafted steel and fiberglass parts, including its 4.5-meter blades.

The team of UMass engineering grad students, faculty advisors, and one precocious undergrad built it to prove that wind energy could keep rural homes toasty in New England’s frigid winters, as a way of trimming U.S. oil dependence—a national imperative in the aftermath of the 1973–1974 energy crisis. To illustrate the point, they also assembled a modular home there on Orchard Hill, and outfitted it with heaters that would be powered by the turbine.

In 1975 and 1976, a crew from the University of Massachusetts Amherst designed and constructed the 25-kilowatt wind turbine that kick-started the U.S. wind industry. Sandy Butterfield

It worked—too well. “We had to open up the doors in the dead of winter. It was just too damn hot,” recalls Michael Edds, who designed the turbine’s electrical system and served as the project’s first resident engineer. Fittingly, they dubbed the turbine the “Wind Furnace.”

The turbine maxed out at 25 kilowatts—puny compared to modern machines that generate up to 26 megawatts, but more than most energy experts expected from wind technology in November 1976. Back then, wind power still conjured up images of quaint Dutch mills and creaky prairie water pumpers. Crafty engineers would soon show that wind power could be so much more. And it all began with the brilliant, commanding, and often polarizing UMass professor leading the Wind Furnace project: William Heronemus.

A retired U.S. Navy captain, Heronemus had joined the UMass faculty in 1967. He’d earned Bronze Stars for valor in World War II, designed and built nuclear submarines, and liaised with the British Royal Navy on the Polaris missile. UMass had recruited Heronemus to do ocean engineering, but the energy crisis and his growing misgivings about nuclear power shifted his attention to renewable energy.

Heronemus, photographed circa 1973, publicly advocated for the buildout of wind turbines, both onshore and off, at immense scale. Robert S. Cox Special Collections and University Archives Research Center/UMass Amherst Libraries

By 1972, Heronemus was advancing detailed designs to deploy wind turbines at immense scale. That year, at the Marine Technology Society’s annual gathering in Washington, D.C., he presented schemes for building thousands of them across the Great Plains as well as a vast grid of massive floating turbines transecting New England’s continental shelf. Wind power, he contended, could generate nearly a fifth of U.S. electricity needs by the year 2000. Never mind that the technology for such an enormous buildout had yet to be commercialized. Espousing grand schemes made Heronemus a quixotic figure.

He also vigorously attacked the commercialization of nuclear power, creating enemies within electric utilities and U.S. government agencies that saw nuclear technology as the future. They didn’t appreciate his claims that a cleaner energy future via wind was ready to be tapped, and that the push for nuclear power and its radiological risks was unnecessary. As author and energy analyst Peter Asmus put it in his 2000 book, Reaping the Wind: “William Heronemus was a dangerous man suggesting an audacious departure from the status quo.”

The UMass Amherst wind turbine generated most of the energy to heat a modular home through the cold, windy winters on Orchard Hill. Solar thermal panels provided some heat during windless periods. Robert S. Cox Special Collections and University Archives Research Center/UMass Amherst Libraries

What happened on Orchard Hill in 1976 marked Heronemus’s turn from provocateur to changemaker. The success of the experimental turbine set off waves of technological and industrial developments that forever changed the energy landscape. Within a few years, the students he trained and the entrepreneurs he inspired were building the world’s first modern wind farms and leading the Great California Wind Rush—the market that turned wind craft into an industry that’s still growing fast half a century later.

Globally, annual wind generation more than tripled between 2015 and 2025, according to data from Ember Energy, a think tank based in London. It will best nuclear’s global output by the end of this year, Ember predicts. And it all started with Heronemus, says Robert Thresher, longtime former director of wind research at the National Renewable Energy Laboratory (NREL) in Golden, Colo. (a U.S. Department of Energy lab rebranded late last year as the National Laboratory of the Rockies). “In my mind he was the father of the people that went out and really made the industry what it is today,” he says.

William Heronemus and the History of Wind Power

I got to know Captain Heronemus posthumously, interviewing his contemporaries and sifting through boxes delivered to the UMass Amherst archival research center’s 25th-floor reading room. During three visits there since 2023, I have discovered clues to his life, thinking, and research process amid the writings where he pitched his big ideas to the world. His papers include proposals to governments, utilities, and deep-pocketed philanthropists and investors, including Jane Fonda and Goldman-Sachs. Papers reveal the internationalism and commitment to service that took Heronemus on renewable-energy consulting trips to Pakistan, Cuba, Côte d’Ivoire, and beyond. Records show meetings with corporate powerhouses like Boeing and Grumman Aerospace and calls on politicians, including the senator and presidential hopeful Ted Kennedy. Postcards from former students exude gratitude.

Heronemus sits with a mock-up of a multirotor turbine in his cramped office in Marston Hall, UMass Amherst’s main engineering building. Robert S. Cox Special Collections and University Archives Research Center/UMass Amherst Libraries

I learned that Heronemus turned his attention from ocean engineering to energy a few years after arriving at UMass, when he saw the growing string of nuclear power plants going up along the Connecticut River, which flows past Amherst en route to Long Island Sound. The U.S. government had picked nuclear power as an antidote to the 1970s oil crises, and Northeast utilities had jumped in big. But Heronemus and other UMass engineers worried that the riverside reactors’ waste heat would threaten the river’s ecosystem and bounty.

The advent of cooling towers to blow off heat into the air addressed the thermal pollution concern but created another: water depletion. (Nuclear plants consume about 60 million gallons of water per day, per reactor, on average.) And Heronemus perceived other nuclear power liabilities, stemming from his experience with nuclear propulsion on Navy ships. As a design engineer and head of construction and repair for a shipyard, he valued the military’s zero-accident standard for reactors but also knew the high cost of adhering to it. He argued that building expanded versions of the Navy’s pressurized water reactors to power cities and factories couldn’t be both safe and economical.

In 1971, Heronemus designed an offshore turbine with three rotors, but the first big multirotor prototype wouldn’t be built for another four decades. Robert S. Cox Special Collections and University Archives Research Center/UMass Amherst Libraries

He predicted—accurately, as it turned out—that costs would rise sharply as the nuclear industry addressed safety and environmental concerns. “Each plant costs more than its predecessor. The shipyards involved with nuclear reactors came to that conclusion years ago,” he wrote in a 1973 research proposal. He also argued that the risks inherent in nuclear reactors and their radioactive waste were unnecessary given Earth’s abundant solar and wind energy resources. He broadcast those views wherever and whenever he could: before congressional committees, at U.S. Atomic Energy Commission hearings, at academic conferences, in media interviews, and even at Rotary Club luncheons.

At a 1973 licensing hearing for the proposed 820-MW Shoreham Nuclear Power Plant on Long Island, N.Y., for example, Heronemus called affordable nuclear energy a “myth.” He detailed, in its stead, a floating wind power system that could be moored off Long Island and sized to deliver more than four times as much electricity as the Shoreham plant. Each of the 640 floating platforms would carry six rotors and crank out up to 12 MW, some of which would power electrolyzers to generate hydrogen. The hydrogen would be fed to power plants or fuel cells to produce electricity when the wind wasn’t blowing. This seemingly futuristic idea drew on his Navy experience with water-splitting electrolyzers, which supplied the oxygen that enabled subs to remain submerged for months at a time, and NASA’s use of hydrogen fuel cells to power the Apollo missions.

More than five decades later, his vision for offshore wind power is big business. Floating platforms are now widely accepted as the future of offshore wind, as necessity pushes the industry to build in deeper waters. Testing began on the first floating electrolysis platforms in 2023, and multirotor turbine prototypes are in development in China, Norway and Scotland.

The UMass Amherst Wind Turbine Legacy

Photos in the UMass archives invariably capture Heronemus in jacket and tie, usually standing bolt straight. That commanding affect, plus his World War II veteran pedigree, Cold War engineering credentials, and his informed, pugnacious attacks made him a hard target for his adversaries in the nuclear establishment. He certainly wasn’t your typical antinuclear activist.

Wielding his Cold War engineering credentials and often dressed in a suit and tie, Heronemus fought hard against nuclear energy, arguing that wind was a far safer and cost-competitive resource.Robert S. Cox Special Collections and University Archives Research Center/UMass Amherst Libraries

But brutal candor in public settings probably won him as many enemies as friends. Consider his presentation at the IEEE Power and Energy Society’s 1974 winter meeting, where Heronemus suggested scrapping the utilities’ then nuclear-focused research arm, the Electric Power Research Institute. That stance no doubt created discomfort for the engineers in attendance who were involved in EPRI projects, or who aspired to be.

It’s hard to say whether Heronemus’s campaign slowed nuclear development. The industry was already struggling with cost overruns when, in 1979, a reactor at Three Mile Island in Pennsylvania partially melted down and slammed the brakes on further expansion.

What is certain is that Heronemus spurred investment in wind power. When he started talking up wind in the early ’70s, even fellow travelers in the fledgling renewable energy movement were writing it off. As future White House science advisor John Holdren opined in a 1971 Sierra Club book: “There are few places in the world where the wind is strong enough and steady enough to make harnessing it for the large-scale production of power at all interesting.”

Heronemus dreamed up networks of wind turbines over and along highways after driving down the Garden State Parkway to a conference in Cape May, New Jersey. Ellen Heronemus

Heronemus countered the naysayers by quickly forging expert consensus around wind power’s immense potential, playing a key role as the sole wind expert on a 1972 federal panel on renewable energy. That joint National Science Foundation–NASA panel concluded that, in fact, wind could meet up to 19 percent of projected U.S. power demand by the year 2000.

Congress listened, sort of. After most Persian Gulf states restricted oil shipments to the United States in 1973, congressional appropriators dedicated US $1.8 million to wind-power research and development for 1974—up from zero—and by 1976 it had bumped that to $22 million. (For comparison, Congress gave nuclear power $714 million in 1976.)

Heronemus’s vision for a massive highway wind-power scheme was inspired in part by the wind-power advocate Percy Thomas, who in the 1940s and 1950s “talked a lot about how fresh New Jersey winds are,” he told the New York Times in 1974. “I got to thinking about what Thomas had said and how wind energy could be captured there.” Ellen Heronemus

The bulk of the funding for wind power flowed to big aerospace firms and to NASA, financing an ultimately fruitless attempt to leap straight to megawatt-scale wind turbines. UMass struggled to grab a slice of the leftovers to pursue Heronemus’s offshore wind system. Professors and students who worked with Heronemus told me they felt they’d been blackballed as payback for his activism and antagonism.

UMass finally caught a funding break when Heronemus dialed back his ambitions and proposed the 25-kW unit for Orchard Hill. A $130,000 federal grant landed in early 1975, and $150,000 more the following year. It was a “trivial” sum, according to team member Sandy Butterfield, who would later become chief engineer for wind-turbine testing at NREL. “They gave us just enough to fail,” says Butterfield.

A crane erects the “Wind Furnace” in November 1976. Sandy Butterfield

But the project triumphed, resulting in Wind Furnace 1, or WF-1 (pronounced “woof one”). The young engineers behind it credit their success to the confidence, sense of mission, and structure that Heronemus gave them. The self-described “hippies” called Heronemus “the Captain” out of both affection and respect.

As team member Edds puts it: “What showed in his demeanor and his actions was discipline, and it sort of rubbed off on us. We didn’t always dress like the Captain, but we knew we had to be disciplined, to be prepared, and just do the job.”

From Helicopter Rotor to Wind Turbine

Team WF-1 got a quick start, thanks to earlier, privately financed work by a couple of doctoral students, including Forrest “Woody” Stoddard. Stoddard had been designing helicopter rotors for the U.S. Air Force when Heronemus invited him to come work on wind power in 1972. Stoddard set about adapting helicopter-rotor theory to the closely related wind rotors, and his aerodynamics modeling proved essential to the engineering of the entire machine.

Woody Stoddard [far right, in hat] designed the fiberglass blades with Ted Van Dusen. The team assembled the blades in a campus shop, and when it was time to squeegee epoxy from the blades, it was all hands on deck. Robert S. Cox Special Collections and University Archives Research Center/UMass Amherst Libraries

As WF-1’s de facto chief designer, Stoddard likely supported the team’s early choice to mimic a helicopter’s ability to “pitch” its blades. To fly forward, a helicopter continuously adjusts the lift created by each blade, turning the airfoil on its long axis to reduce lift as it swings past the front of the aircraft. Doing so tilts the nose down and moves the vehicle forward. In WF-1’s case, blades pitched to regulate torque, helping get the rotor spinning in low winds and then easing off to protect the machine in dangerously high winds.

Repurposing a truck axle to mechanically couple WF-1’s rotor and generator was one of several design elements borrowed from engineers at McGill University in Montreal. Production of WF-1’s fiberglass blades got started at UMass in 1974 under the direction of doctoral student Ted Van Dusen. A competitive rower, he had a side hustle making ultralight composite boats—a trade that had stalled his doctoral work at MIT but was an accelerant for WF-1.

The federal funds in 1975 allowed Heronemus to really spin up the project and recruit a squad of students to engineer the balance of WF-1’s components. They made good use of the UMass engineering machine shop and received guidance from faculty, including mechanical engineering professors Duane Cromack and Jon McGowan. But it was the dozen or so students who really cranked out the parts.

Most were master’s students, like Butterfield, who designed the blade-pitching mechanics. Edds, the team’s only electrical engineer, had come to UMass to learn ocean engineering, only to be diverted into handling WF-1’s generator. Louis Manfredi, another ocean engineering student, teamed up with master’s student Jim Sexton on the nacelle housing the generator and drivetrain. Fred Antoon adapted the truck axle. Brian Kuhn did drawings.

WF-1 contained a mechanism that pitched its blades to regulate torque in response to wind speed, a feature that became an industry standard.Sandy Butterfield

An 18-year-old freshman, Dan Handman, came aboard and soon made himself indispensable. When he approached Heronemus to introduce himself, Heronemus handed him three months’ worth of anemometer readings punched into recording paper, and told him to turn it into 15-minute averages. Figuring there had to be a more efficient method for analyzing wind speeds, Handman asked around and found a wind-averaging machine from an earlier student project. A month or so later, he’d installed it in a cabinet near Heronemus’s office and wired it to an anemometer on Orchard Hill.

Handman’s primary role on WF-1 was setting up its computerized control system, which tracked wind speed and sent commands to Butterfield’s pitch mechanism. The controls also tracked the generator’s speed and adjusted the current to its rotor windings, in accordance with calculations by Edds. Tweaking the current ensured that power demand from the electric heaters installed in the home below didn’t stop the rotor in weak winds.

Sandy Butterfield, part of the 1970s “UMass Mafia” team that built WF-1, became a wind-power entrepreneur and a top engineer at the National Renewable Energy Laboratory in Golden, Colo. Sandy Butterfield

The finished WF-1 really cranked up the heat, some of which was stored by heating water in tanks in the modular house’s basement, to be circulated through baseboards in windless periods. It turned out WF-1 was unusually efficient at capturing wind energy because its rotor could change speed with the wind, keeping the blades close to an aerodynamic optimum.

This varying rotor speed meant that the frequency of the electric power WF-1 produced also varied. Turbines linked to power lines must strive for the opposite—a steady output that synchronizes with the grid’s frequency—primarily 50 or 60 hertz. But it suited the home’s low-tech heating scheme just fine. (Electronic converters let today’s turbines have it all by ingesting a variable wave and outputting a new wave that’s synced to the grid.)

The Great California Wind Rush

In 1977, with WF-1’s success in hand, Heronemus projected that 3 million homes like the one on Orchard Hill could soon slash U.S. heating oil demand by 90 million barrels a year. That never happened, but an industry was born, starting with a Burlington, Mass. startup called US Windpower—the first “credible” U.S. turbine manufacturer, according to Thresher, who is now an emeritus researcher at the National Laboratory of the Rockies.

Belgian-made WindMaster turbines erected at Altamont Pass signaled the internationalism of the California wind rush. UMass team member Woody Stoddard conducted engineering analyses of many early designs deployed there.Bettman/Getty Images

Boston-area entrepreneurs Russell Wolfe and Stanley Charren launched US Windpower with Stoddard and Van Dusen after visiting Heronemus in 1974 and liking what they heard. They adapted WF-1’s design to make it suitable for grid-connected operation, building and breaking prototypes before erecting the world’s first grid-connected wind farm in 1980—20 turbines on a mountain in New Hampshire. California’s water authority placed an order for 100 MW of wind power, and in 1981 US Windpower began installing hundreds of turbines in Altamont Pass, east of San Francisco.

As more firms jumped to California, drawn by state government incentives, WF-1’s creators and the next cohort of UMass grads assumed important roles in the nascent market. Seven joined Energy Sciences, a startup cofounded by Butterfield. More joined U.S. Windpower. Stoddard left that company to start a consulting firm and ended up advising some of Denmark’s modern wind pioneers, which rapidly expanded thanks to the California market. Those early Danish firms made relatively simple, sturdy machines that subsequently scaled up and dominated globally for several decades — until China embraced wind power.

The California wind power boom peaked in 1986, after which energy prices collapsed and incentives faded. Most manufacturers were bankrupted by equipment failures and financial challenges, making the 1990s a tough time for wind power’s pioneers. Many UMass wind engineers, like Butterfield, joined Thresher’s operation at NREL, culling everything they could from the California experience.

“An entire generation of U.S. wind engineers got their graduate training, at least in part, using the Wind Furnace.”—Harold Wallace

There, Heronemus’s protégés became known as the “UMass Mafia.” Thresher says it attests to the crew’s impact: “There were others. But that UMass Mafia were really leaders in the field. I think that’s the heritage we got from Bill Heronemus. Those people were so impactful and the education they got [with Heronemus] was the key.” What Heronemus began at the university became the UMass Wind Energy Center, which has awarded over 300 graduate degrees.

WF-1 now rests in the Smithsonian Institution’s collections in Washington, D.C. It earned its place there, as Smithsonian’s only modern wind turbine, because it represents wind energy’s revival, according to Harold Wallace, Smithsonian’s curator for electricity collections. “An entire generation of U.S. wind engineers got their graduate training, at least in part, using the Wind Furnace,” he says.

Heronemus didn’t get to witness the production of the massive offshore machines that he foresaw. He lost his long fight with cancer in November 2002, at the age of 82, even as former students and family members were racing to patent his multirotor and floating turbine designs.

Had he lived longer, the Captain would almost certainly have railed against current U.S. energy policy. The U.S. government has never backed wind power as generously as he’d hoped. Wind supplied 10 percent of U.S. generation last year—that’s half the share in Europe—with offshore turbines providing only a tiny sliver. Federal support for wind power has been in a stop-go cycle since Ronald Reagan’s administration, and it’s hit a low again under President Donald Trump, who has vowed to stop wind power cold. As Trump boasted to oil executives in January: “We have not approved one windmill since I’ve been in office, and we’re going to keep it that way.”

Under Trump, stop-work orders have disrupted offshore projects from Massachusetts to Virginia, contributing to a nearly $600 million loss in 2025 for GE Vernova’s wind business. GE Vernova is the only major wind turbine manufacturer remaining in the United States, and it too can be traced back to Heronemus via a US Windpower patent.

In stark contrast, European and Asian countries have been going big on offshore wind and are now developing floating wind farms to push into deeper waters. China might be the one to finally conjure up Heronemus’s favored wind design: floating platforms bearing massive multirotor machines. In 2024, Zhongshan-based turbine maker Ming Yang Smart Energy Group deployed a two-rotor offshore prototype. The company says its next iteration will generate a whopping 50 MW—a twin-headed beast that would be the world’s most powerful wind machine.

That will be a bittersweet moment for the U.S. wind industry and Captain William Heronemus’s UMass Mafia, for whom such massive machines are a dream come true. Joanne Carroll, a retired member of the UMass Mafia, says she remembers the very moment, her freshman year, when Heronemus’s dream became hers. While he was lecturing in Introduction to Engineering about the hidden costs of coal-fired power, Heronemus walked to the window and said: “‘But out there there’s wind, and you can harvest that energy,’” Carroll recalled. “And I remember thinking: That’s what I want to do with my life.”

The author would like to give special thanks to UMass professor emeritus James Manwell for his assistance with this story.

Yen-Ling Kuo always wanted to understand how things worked. When she was growing up in Taiwan, reading the story of Michael Faraday in elementary school piqued her curiosity about the natural world. During that time, she was introduced to Logo, a computer program with a turtle cursor to help children learn basic coding through hands-on experimentation.

It was Kuo’s introduction to programming logic.

Yen-Ling Kuo

Employer

University of Virginia in Charlottesville

Title

Assistant professor of computer science

Member grade

Member

Alma maters

National Taiwan University; MIT

In high school she learned the capacity computers held. She could write programs that completed tasks independently, she realized.

“Once I discovered how powerful computers could be,” she says, “I knew I wanted to focus on using them to solve real-world problems.”

Kuo, an IEEE member, never lost her interest in the “how” behind processes and tools. Her curiosity, combined with a stint working at a Silicon Valley company, led her to focus on innovations that live at the intersection of cognitive and computer sciences.

Kuo, now an assistant professor of computer science at the University of Virginia in Charlottesville, last year received the IEEE Robotics and Automation Society’s inaugural Outstanding Women in Robotics and Automation Early Career Contribution Award. The award is part of the IEEE-RAS Women in Engineering’s Outstanding Women in Robotics and Automation (WiRA) Paper Awards, which promote excellence and recognize the impact that female researchers have on robotics and automation fields at different stages in their academic careers.

Kuo’s winning paper, “Diff-DAgger: Uncertainty Estimation with Diffusion Policy for Robotic Manipulation,” demonstrates a novel method to help robots better identify and estimate uncertainty when faced with scenarios on which they’ve not been trained. The method reduces the amount of human supervision, improves a robot’s rate of successful task completion, and opens up a path to introduce more complex models with bigger data demands into interactive robot learning.

She says her research will help people working in the robotics and automation fields more efficiently collect the data needed for effective model training.

Silicon Valley’s impact

Kuo earned bachelor’s and master’s degrees in computer science at the National Taiwan University, in Taipei, in 2009 and 2012. As she was nearing completion of her master’s degree, she did what many computer science graduates do: She pursued a summer internship at a tech company.

She spent the summer of 2011 at Google’s campus in Kirkland, Wash., working on the company’s comparison ads project.

When her internship ended, she joined the MIT Media Lab as a visiting student, working on the Open Mind Common Sense project with Henry Lieberman.

As she was considering pursuing a Ph.D., a call from Google changed her plans. The company offered her a full-time role as a software engineer.

“I viewed the job offer as a positive development,” she says. “I believe it can never hurt your future research career to get some real-world experience under your belt.”

She was hired in 2012 and helped build techniques that incorporate computer vision and natural language processing to improve the customer shopping search experience. She led the company’s Shop the Look initiative, a predecessor to Google’s current AI-powered shopping experience. The project connected social media content with search results, something the company had struggled to do in the past.

Kuo and her team were tasked with building a connection between the natural language people use to describe an item and an image that matches the searcher’s intent. It was at a time when the neural network—using deep learning models to power Google products—was gaining momentum at the company. Integrating neural network tools into her work was a requirement—which raised questions for Kuo.

“I was applying the neural network tools,” she says. “But I didn’t have 100 percent certainty about how they actually worked.”

She considered how she could become more knowledgeable about deep learning models. It was a full-circle moment. She decided that after nearly four years at Google, it was time to earn a Ph.D. in computer science. She returned to MIT in 2016.

The question that changed everything

Boris Katz, one of Kuo’s Ph.D. advisors, is a principal research scientist and the head of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)’s InfoLab. He also led the creation of the START Natural Language System, the world’s first Web-based question-answering system.

When the two met, Katz asked Kuo why she wanted to pursue a doctorate degree. She explained her interest in understanding how neural networks work and in using that knowledge to connect the physical world with human language.

He suggested she attend a summer course at MIT’s Center for Brains, Minds, and Machines, a research initiative that ran from 2013 through 2025. CBMM’s objective was to bring together computer scientists, cognitive scientists, and neuroscientists to understand how human intelligence works. The goal was to use the resulting insights to establish an engineering practice to build artificial intelligence systems.

For Kuo, it was a chance to better understand human intelligence and identify ways it could be replicated in machines.

“It was an opportunity for me to interact with other scientists and gain insight into how people learn, understand, and figure things out in the world,” she says. “I saw it as a very useful and inspiring way to incorporate those ideas into my own research work.”

During her Ph.D. studies, she was a research assistant at CSAIL. The experience helped shape her doctoral research, which focused on building AI systems that apply past learning to new situations. She developed machine learning models to support the efforts, including language understanding and social interactions.

She completed her Ph.D. in computer science in 2022 with a minor in cognitive science.

After graduation, she continued her work and collaboration at CSAIL, particularly on projects that involved the “theory of mind” concept.

Theory of mind spurs innovation

Theory of mind isn’t new, having originated with primatologists studying chimpanzees in the late 1970s. The theory recognizes that others have their own thoughts, beliefs, and perspectives. It’s a skill that allows humans to infer someone’s mental state and predict their behavior without verbal communication.

“It’s like when college roommates are moving into their dorm. They may not talk too much, but they work together naturally to coordinate their activities and accomplish goals,” Kuo says. “They can infer and mentally interpret each other’s behaviors and signals to make decisions and complete tasks without words.”

She brought her theory of mind research to the University of Virginia when she joined as an assistant professor in 2023.

Kuo conducts her research in UVA Engineering’s multidisciplinary cyberphysical Link Lab. Her broad focus is on developing computational models that help robots interpret both direct data and silent signals, from language and movements to a person’s gaze. If successful, it could give robots the same sort of physical and theory of mind reasoning capabilities that power physical and social interactions among humans.

“There are no computational frameworks yet available that will translate this kind of understanding into a robot efficiently,” she says.

She adds that the process to get there begins with improving how robots learn to perform tasks.

The evolution of robot learning

Historically, one way robots learned was to mimic humans. A researcher would manually guide a robot through a task, like cutting an apple, and it would repeat the movements. The robot was successful until the environment changed, such as when its hand was in a different position or the apple was at a different angle. The robot was then faced with a situation for which it hadn’t been trained. Without any data available to help it correct course, the robot would start making small errors that eventually led to a full system crash.

This diagram describes how the robotic gripper’s visual perception and tactile sensing prevents a potato chip from breaking.Xuhui Kang, Yen-Ling Kuo, et al.

To solve the problem, researchers developed the dataset aggregation (DAgger) method. As a robot performed a task, a researcher was on standby to provide real-time corrections during unexpected scenarios. The correction data was continuously added to the robot’s model, teaching it how to recover from mistakes.

To reduce the human monitoring effort, robot-gated DAgger was created to enable bots to query humans when the machines became uncertain.

The most popular approach to make the query decision is to train multiple models to consider when determining a course of action. If the models all agree, the robot proceeds. If they don’t agree, the robot is likely to get stuck and ask for help.

Although the multiple model approach was widely adopted, it has limitations. Practically speaking, as models become more complex, it is hard or impossible to train multiple copies. A more fundamental issue is that disagreement among models doesn’t always imply uncertainty; it could just mean there are different ways to accomplish a task.

The Diff-DAgger solution

That is the gap Kuo’s research team closed with the novel Diff-DAgger research. The approach builds on diffusion policy, a technique that helps robots account for different ways a task can be performed.

The new method repurposes diffusion loss, the signal a robot uses to improve its model during training, as a real-time confidence check. During task execution, the robot computes the signal and compares it against values from its training data using a statistical test. The signal spikes when the robot faces an unfamiliar situation and is uncertain how to proceed. The signal stays silent when the robot’s current action is close to what it learned before.

The spike represents the robot’s ability to self-diagnose and predict an imminent failure. Human intervention is triggered only when the signal spikes. No spike means the robot can be left to complete its decision-making process on its own.

Kuo’s team achieved significant results: Failure prediction rates were improved by 39 percent. Task completion rates were increased by 20 percent, and tasks were completed nearly eight times faster.

Her research at UVA gained attention from the National Science Foundation, which honored her last year with a Career Award, the foundation’s flagship grant for early-career researchers. The five-year US $665,000 grant supports her research that builds computational models for human-robot interactions through theory of mind reasoning.

She also received the Toyota Research Institute’s Young Faculty Researcher Award to teach cars to reason about interactions on the road and with the driver.

As service robots and self-driving vehicles become more available, such works are likely to make interactions between humans and robots more intuitive and useful.

Kuo ultimately wants to build more robust robots that are able to integrate into a social space with humans by engaging with us through grounded interactions, she says.

The impact of IEEE

Like many IEEE members, Kuo was introduced to the organization as a student. In 2018 she submitted her first paper, “Deep Sequential Models for Sampling-Based Planning,” to the IEEE/Robotics Society of Japan International Conference on Intelligent Robots and Systems while pursuing her Ph.D. at MIT. Her IEEE involvement grew alongside her professional career.

“It was a natural segue to transition from student to a full IEEE member,” she says. Today she is an active volunteer with the IEEE Robotics and Automation Society, a reviewer for submitted papers, and a presenter and panelist at conferences.

She says one of the best parts of attending conferences is having the opportunity to engage with students. She also enjoys participating as a panelist at luncheons, she says, because it gives her one-on-one time with student attendees. She can share her knowledge and offer insights as they prepare to embark on their career.

Her goal in the coming years, she says, is to broaden her involvement with IEEE initiatives and branch out to other technical committees. Sharing knowledge and learning from others is essential to anyone’s career growth, she says, and “IEEE offers a great opportunity for both.”

“Space computing, the final frontier, has arrived,” Nvidia CEO Jensen Huang declared at the Nvidia GTC conference in March.

Indeed, the idea of data centers in orbit has gone from science fiction to a serious spending category. Elon Musk’s SpaceX has acquired xAI (also Musk’s) and is planning a constellation of space-based data centers. Google, not to be outdone, announced Project Suncatcher in partnership with Planet, planning to launch two satellites equipped with Google Tensor Processing Unit (TPU) AI chips by early 2027. Startup Starcloud has already filed a proposal with the Federal Communications Commission for an 88,000-satellite constellation for orbital data centers. As Starcloud’s filing suggests, these companies are all proposing fleets of satellites numbering in the thousands, each housing a rack or multiple racks of AI-grade GPUs, interconnected with each other through free-space optical links and communicating back to Earth via microwave links, either directly or through other satellites.

Proponents tout the many wonders of computing in space: abundant solar energy, free cooling, and freedom from Earth-based disturbances like earthquakes, floods, and protesters. But a sober look at the physics of space-based computing paints a much more nuanced picture.

Free cooling is perhaps the biggest misconception. Space is cold, but it also has no atmosphere. That means the best heat-removal mechanisms, conduction and convection, are off the table. The only option is radiation. To prevent a chip from overheating in space, a large, costly surface area is required to dissipate the energy and then radiate it.

Solar energy is abundant, but collecting it with functional solar panels that maintain perfect alignment toward the sun is a complex task requiring extensive attitude control systems. On top of that, ionizing radiation in space from cosmic rays and other sources poses a unique challenge, degrading the solar panels, the radiative coolers, and the chips themselves. Because regular maintenance in space is difficult, redundancy has to be built in at launch, and cost estimates have to account for efficiency degradation over time.

At ABI Research, where I work as an aerospace analyst, we did a rough total-cost-of-ownership comparison between a data center on Earth and one in space. It showed that the cost to launch and run a GPU in space for a year is at least an order of magnitude higher than the same feat in a terrestrial data center. Our model was simple, assuming an Nvidia H100 server rack launched with the requisite-size solar panel and radiator on a spacecraft akin to Starcloud’s pilot launch. We assumed SpaceX’s Starship was used at a highly optimistic launch cost per kilogram of US $44, and a terrestrial energy cost of $0.20 per kilowatt hour. This is a simple back-of-the-envelope calculation, but it does signal something real.

From our perspective, the cost of delivery and space hardening of the payload makes general-purpose space-based data centers difficult to justify economically today, despite the fact that data-center builders in many regions are scrambling for electric power. However, there are niche applications where the much higher costs of computing in space could be justified. Examples include preprocessing data from Earth-observation satellites, real-time detection and tracking of hypersonic missiles, and active collision avoidance in the increasingly crowded low Earth orbit. Even for these, though, contending with fundamental physics will still be a demanding challenge. And a technologically compelling one, too.

The Cooling Challenge in Space

Cooling is where physics separates the science from the fiction. The governing equation for radiative cooling, the only type of cooling available in space, is known as the Stefan-Boltzmann Law. It states that the amount of power you can radiate is proportional to the area of the radiator times its temperature to the fourth power. For a space systems architect, the implications of this law are brutal. In orbit, the only variable we can control is area. This restriction creates a geometric penalty, or a “physics tax,” for cooling in space: The more power you need to reject, the bigger the area of the radiator you need to bring along from Earth.

The only cooling method available in space is radiation, and the radiator area required is derived using the Stephan-Boltzmann law. For a single chip drawing 700 watts, like Nvidia’s popular H100 GPU, the area required to keep it at 20 °C is just under 3 square meters, and it goes down to 1 square meter for an operating temperature of 85 °C. However, as the radiator surface is exposed to ionizing radiation, its emissivity decreases, and after 5 years in space the required area increases by about 40 percent.

To understand how big this baseline area is in practice, I used the Stefan-Boltzmann law to model the heat-rejection area needed to keep a single chip that draws 700 watts of power—such as the H100 GPU chip, an AI stalwart—at a constant 60 °C, usually considered the sweet spot for GPU longevity and stability. I further assumed that the radiator is perfectly facing deep space, at a chilly background temperature of 3 kelvins. By this calculation, a single chip would require 1.4 square meters of radiator surface.

To put this into perspective, consider that a common AI rack can hold approximately 32 GPUs (four H100 server boards). With CPUs, memory, and networking equipment, this rack would draw around 40 kilowatts of power. This single rack includes 2.5 terabytes of memory—enough capacity to serve over 20,000 concurrent users or run 16 simultaneous instances of Llama 3, an open-source AI model. But to cool this thermal load in a vacuum, that single rack would require an 80-square-meter radiator, roughly the size of a pickleball court. For an aggregate 100-megawatt data center, you’d need at least 2,500 of those radiators.

And that’s the best-case scenario. Additional problems are hidden in the low Earth orbit environment itself. Space exposes radiators and their coatings to a chemically hostile brew of ultraviolet light and atomic oxygen, quite the opposite of a clean-room environment. Over a LEO satellite’s typical 5-year lifespan, these elements degrade the radiator’s surface properties and lower its ability to shed heat.

Including this degradation in the model reveals that as the radiator degrades from a “fresh” state to an “end-of-life” state, the physics demands a further penalty. To maintain that same 60 °C operating temperature for the GPU chips, the required surface area jumps from about 1.4 square meters per chip to nearly 2.0 square meters. In other words, the physics tax rises by 40 percent. Therefore, you must launch at least 40 percent more radiator mass, endure higher atmospheric drag, and sacrifice valuable launch volume just to survive the degradation of the thermal coating. This increase adds significantly to the launch cost and further erodes the economics of a space-based data center.

The Silicon Challenge in Space

Solving the heat problem is only part of the battle. The other significant challenge in low Earth orbit is ionizing radiation, which affects the computing hardware itself. Today’s satellites typically use radiation-hardened processors, which are very reliable but also much more expensive, and they perform poorly compared to commercial off-the-shelf processors.

A standard rad-hard chip doesn’t have the processing power to run a modern large language model (LLM). As a result, satellite operators aspiring to launch a data center have no choice but to make a risky compromise: to use hardware meant for terrestrial use. In order to achieve the necessary compute density, orbital data centers must use the same Nvidia H100s or Google TPUs found in terrestrial server farms. The problem is that these chips are “soft” targets in space. High-energy particles can flip bits in memory or cause “latch-ups” in logic that fry the circuit.

Three big proposals for orbital data centers vary in satellite size, number, and cooling plans.

One possible option is to shield the computers from radiation with thick, absorbent panels. However, the shielding would add significantly to the already heavy satellites. The other option is to compensate for the radiation damage with redundancy. Indeed, edge computing architects are moving toward software-defined resilience, where instead of one perfectly hardened computer, operators fly a cluster of imperfect, commercial ones whose total cost could be as low as one-tenth to one-hundredth that of the rad-hard model.

This redundant approach is used in many spacecraft, including Artemis II, which recently carried astronauts around the moon, as well as SpaceX’s flight computers and the Hewlett Packard Enterprise edge servers for the International Space Station. By running three (or more) instances of the same calculation on three different nodes and comparing the answers, the system can detect a corrupted processor. If a node fails, the “orchestrator” reboots it while the others continue the mission. While this ensures resiliency, it also means that some fraction of the compute capacity is dedicated to redundancy, further increasing the costs.

The Energy Challenge in Space

An often-touted advantage of space-based data centers is the seemingly unlimited supply of free, clean energy from the sun. Solar energy in orbit is indeed abundant, at 1,361 watts per square meter. Of course, capturing that free energy is made possible only by the very costly launching of large solar panels into orbit. And those solar panels also degrade over time due to radiation exposure, typically losing 1 to 3 percent efficiency per year.

Let’s say a solar array collects 1 MW of power to run an AI cluster. The laws of physics demand that the satellite must eventually radiate 1 MW of waste heat. Because the square area needed to generate the solar power—around 400 W/m2—and to reject the heat—around 450 W/m2—are nearly equivalent, every square meter of power generation now demands approximately another square meter of cooling. The radiator needs to be a structural equal, not merely a passive coating on a surface used for something else.

As Elon Musk recently noted in Davos, the most efficient radiator is one that never sees the sun. By orienting the spacecraft so the solar panels face the sun and the radiators face the deep vacuum of space, efficiency skyrockets for both. But there’s a catch: Maintaining this perfect three-way alignment—panels to sun, radiator to the void, antennas to Earth—requires complex, high-torque attitude control systems. So this configuration means more payload and more computing power. Plus, these control systems are complex components with many failure modes, which is not optimal in a situation where maintenance is difficult.

The Killer Apps for Computing in Space

Given all these challenges of deploying massive radiators for satellites in the hostile environment of space, why build data centers in space at all?

While training or inference on LLMs in space doesn’t seem economical today, there are other, very compelling applications for computing in space. Here are two: solving the downlink bottleneck from Earth-observation satellites and enabling collision-preventing maneuvers in the increasingly crowded low Earth orbit.

The latest Earth-observation satellites, equipped with hyperspectral and synthetic aperture radar sensors, are used for a range of important reconnaissance missions, such as battlefield intelligence, tracking the global shadow fleet of ships carrying contraband, and assessing earthquakes or infrastructure failures down to the millimeter. These systems can generate hundreds of terabytes of raw data per day that must be transmitted to Earth. However, the radio-frequency “pipes” used to downlink the data are congested, and the ground infrastructure cannot absorb the sheer volume of raw data.

Another immediate, mission-critical application for in-space computation is protecting the orbital environment. With over 17,000 satellites in orbit, the overwhelming majority of which are in low Earth orbit, avoiding collisions between these satellites is crucial. As NASA astrophysicist Donald Kessler pointed out back in 1978, a single space collision could cause a cascading effect that renders the entirety of LEO unusable.

RELATED: Have We Reached a Space-Junk Tipping Point?

According to SpaceX’s recent annual report, the Starlink constellation executes a collision avoidance maneuver every 2 minutes on average. Each maneuver already relies on onboard AI systems but still requires most of the processing to happen on the ground.

As low Earth orbit gets increasingly populated, collision avoidance will have to break the traditional ground-loop model. In the megaconstellation era of space, the OODA (observe, orient, decide, act) loop must happen onboard, thereby reducing the analysis turnaround from minutes to milliseconds.

The problem is that the flight computers standard on satellites are not built for this level of processing. The complex probability models required for maneuvering cannot currently be implemented by onboard computers in conjunction with their navigation systems. Clearly, more powerful computers are needed.

This is the true economic justification for moving compute to space: to move insight generation there. By placing high-performance computing adjacent to the sensors, we can process terabytes of data in orbit and downlink only the relevant data in real time, and we can do the computations necessary to avoid satellite collisions in real time.

The Future of Computing in Space

So, assuming that some form of computing is inevitable in low Earth orbit in the foreseeable future, how will the heat be handled? The industry is currently experimenting with two main classes of solutions to cope with the Stefan-Boltzmann law.

One creative option is to use origami-inspired radiators, the kind used for the James Webb telescope. Companies are developing flexible, high-conductivity composite radiators that fold into a tight cube for launch and unfurl into enormous yet lightweight thermal wings in orbit.

Another possibility is to use liquid-droplet radiators. This concept proposes removing the rigid radiator structure completely and instead spraying a stream of coolant oil directly into the vacuum of space. The fluid travels through an open loop, exposed to the near-absolute zero of the void, maximizing radiative surface area before being caught by a collector and pumped back into the ship. It sounds like science fiction, but as the heat loads climb into the megawatts, liquid-droplet cooling may be the only way to cheat the mass limits of this exponential reality.

Options for Future Radiator Design

Our rough total-cost-of-ownership model uses optimistic versions of current numbers, such as launch cost, chip cost, and power use. A critic might point out that future technology will improve, both in efficiency, purpose-built designs, and costs.

Sure, the technology is bound to improve. But the critical factor isn’t just launch cost; it’s the computing power per unit mass and electric-power economics. Radiators and solar arrays can consume 65 to 70 percent of total satellite mass, and space-grade photovoltaics run orders of magnitude more expensive than terrestrial equivalents.

Chris Philpot

Even as launch costs fall, the mass and cost burden of power generation and thermal management will remain a fundamental problem.

Current space-grade solar panels rely on germanium substrates, whose supply is concentrated in China. It will be extremely difficult to scale up availability of these substrates. A transition to radiation-tolerant perovskite solar panels or a similar alternative could change the economics significantly, but that possibility is five years away or more. The technology will get cheaper, but the bottlenecks of power and thermal architecture will remain.

Recognizing the thermal reality of cooling in space forces us to shift how we view satellite operations. We are moving away from the “launch and forget” era toward an era of “autonomous logistics.” As our thermal model demonstrated, the harsh environment of space steadily attacks the hardware. UV radiation degrades thermal coatings; cosmic rays degrade silicon. In a traditional satellite model, when the radiator degrades or the memory fails, the satellite becomes space junk. For a multimillion-dollar data center, that disposal model is potentially ruinous.

To make the economics of orbital computation work, the infrastructure must be serviceable and the rockets to launch them reusable. The orbital domain will require automated servicing vehicles capable of swapping out degraded radiator panels and upgrading fried servers. In these ways, the future of the orbital data centers is dependent on the innovations of an emergent in-space economy.

There’s a good argument to be made that the need for space-based computation is less of a hype cycle and more of an enabler for the new space economy. Look no further than SpaceX’s recent regulatory filings proposing a constellation of up to a million satellites in low Earth orbit. At such a scale, routing all raw data back to Earth is physically impossible; the network itself must become the data center.

However, the winners in this sector will be determined by the systems architects who most cleverly accommodate the thermodynamics and the companies with sufficient vertical integration to take on the massive costs of operating data centers in orbit. Ultimately, the physics tax is universal. Whether managing heat rejection in the vacuum of low Earth orbit or managing power density in a hyperscale facility in Northern Virginia, the constraint is never the silicon. It’s the thermodynamics.

An examination of how socially assistive wellness robots could support the seven dimensions of senior wellness, and how a framework can measure their autonomy.

What Attendees will Learn

1. Why the senior care crisis exceeds incremental automation. Demographic pressure, workforce shortages, and a daily wellness-programming gap all strain traditional care models.
2. What defines a wellness robot as a category. The seven ICAA wellness dimensions and eight properties separate these robots from companion and medical devices.
3. How autonomy can be measured with CRAS. This six-level scale, modeled on the SAEJ3016 driving standard, evaluates four care dimensions.
4. What maps the road to full autonomy. The paper examines technical capabilities, clinical evidence, and a three-phase roadmap toward the early 2030s.

The EPICS (Engineering Projects in Community Service) in IEEE program, administered by IEEE Educational Activities, has launched the Excellent EPICS in IEEE Contributor Awards. The recognitions honor the program’s outstanding students and faculty volunteers in Excellent Team Leader and Excellent Faculty Advisor categories.

The awards recognize individuals whose leadership, mentorship, and commitment have meaningfully advanced the impact of EPICS projects. Candidates must demonstrate clear, measurable contributions that elevate both the student experience and the outcomes delivered to community partners. Reviewers also consider other awards, publications, presentations, and professional achievements that reinforce the nominee’s credibility and leadership.

Recipients must demonstrate outstanding project management and documentation, strong mentoring and collaboration, and high-quality outcomes.

Here are this year’s recipients.

Team Leader Award

Surattana Kakay is a computer engineering student at Rajamangala University of Technology Thanyaburi (RMUTT), located in IEEE Region 10 (Asia Pacific). Kakay, an IEEE student member, was honored for guiding her team in the design, development, and implementation of the Automatic Water Level Control System project, which aids rice farmers in Thailand.

As the team leader, Kakay played a pivotal role in transforming the student initiative into an operational, community‑centered solution. Her inspiration was purpose-driven, she says.

“My motivation was to apply engineering to real agricultural challenges, like water scarcity and climate change,” she says. “I wanted to bridge advanced technology with the tangible needs of local farmers.”

She managed the project end to end—coordinating workflow, assigning tasks based on team members’ strengths, and ensuring each phase of development aligned with the technical road map she created. She served as the primary liaison between the student team, the Pathum Thani Rice Research Center, and farmers to make sure the system was practical and user‑friendly, and that it addressed community needs.

“Watching students grow as they design solutions that improve lives has been both inspiring and deeply humbling.” —Elizabeth Vidal-Duarte

Under her leadership, the team developed a low‑cost IoT‑based alternate wetting and drying (AWD) system that lets farmers remotely monitor and control water levels in rice paddies using smartphones. Kakay oversaw the integration of noncontact laser time‑of‑flight sensors to withstand harsh field conditions, and she championed the use of long-range technology connected to a free community Wi‑Fi network to eliminate Internet service fees.

The results were transformative, Kakay says.

“Our AWD system reduces water consumption by 63 percent and methane emissions by 7 percent annually,” she says. “Turning an academic assignment into a real‑world solution that delivers measurable, sustainable results has been incredibly meaningful.”

Her achievements advanced sustainability for Thailand’s most water‑intensive crop while demonstrating the potential of accessible engineering solutions.

Beyond technical innovation, Kakay cultivated a culture of learning, continuity, and empowerment within her team. She introduced a mentorship framework to support future student cohorts. She and her team produced academic papers, visual media, and presentations to communicate the project’s value to scientific audiences as well as the general public.

“Surattana Kakay is a pivotal figure in turning innovation into reality and delivering tangible benefits to the community,” says IEEE Member Thanasin Bunnam, her faculty advisor and an assistant professor at RMUTT.

Kakay’s leadership journey became a personal milestone, she says: “Leading this project transformed me from a student into a team leader. As a female engineer, it empowered me to advocate for women in engineering and show that gender is no barrier to technical excellence.”

Through her guidance, the AWD project evolved from a classroom assignment into a solution that illustrates IEEE’s mission of advancing technology for humanity.

Faculty Advisor Awards

Navid Shaghaghi, a lecturer and researcher at Santa Clara University, in California, was recognized for his dedication to integrating service learning into engineering education and fostering student innovation that benefits underserved communities in IEEE Region 6 (Western USA).

During his more than six years of engagement with EPICS in IEEE, Shaghaghi, an IEEE senior member, has demonstrated exceptional leadership in advancing sustainable, human‑centered engineering through the long‑running Hydration Automation (HA) project and the HiveSpy initiative. They are part of Santa Clara University’s Frugal Innovation Hub and EPIC Research Laboratory. The HA project is funded by EPICS in IEEE Fischer Mertel Community of Projects.

Since 2019, Shaghaghi has served as principal investigator for the HA project, guiding its evolution from prototype to a robust, field‑tested irrigation automation system that supports small ranches and community farms in California.

The HA project is a low‑cost system that helps reduce water waste by monitoring soil moisture and automating watering. By combining ultrasonic tank sensing, soil sensors, and ongoing technical support, the project improves efficiency, lowers operational costs, and promotes more sustainable urban agriculture.

Under Shaghaghi’s guidance, more than 30 undergraduate and graduate students have gained hands-on experience in IoT development, field deployment, testing, and client collaboration.

His commitment to frugal innovation and human‑centric design has resulted in solutions that are minimalist, affordable, sustainable, portable, and rugged—often challenging conventional approaches to agricultural technology.

“Turning an academic assignment into a real‑world solution that delivers measurable, sustainable results has been incredibly meaningful.” —Surattana Kakay

The HA project has produced new research publications and earned recognition, including a third-place finish by Shaghaghi’s graduate students at this year’s IEEE Rising Stars Project Showcase. During the annual event, students and young professionals present their technical innovations to industry leaders and peers.

The HiveSpy project is a low‑cost, frame‑level IoT monitoring system that helps beekeepers automate labor‑intensive tasks and prevent hive swarming by tracking production yield in real time. By collecting frame‑weight data and generating optimized harvest schedules, the system reduces manual workload while improving the hive’s health and boosting honey output.

Shaghaghi says his mentorship has been shaped by the realities of student turnover, a challenge he embraces with optimism and adaptability.

“The transient nature of student teams is a challenge but one you must embrace, bear‑hug style,” he says. “By energizing your student community and welcoming new contributors, you’ll be amazed by the brilliant solutions they bring.”

His philosophy has allowed him to cultivate a thriving pipeline of student innovators, he says, and he has strengthened his own professional practice as well.

“I’ve been mentoring EPICS in IEEE students since 2019,” he says. “It has taught me resilience and how to operate on a tight budget while still delivering real‑world results.”

Beyond the technical achievements, Shaghaghi’s work reflects a commitment to humanitarian technology and service learning. As the founder and director of the EPIC (Ethical, Pragmatic, and Intelligent Computer) lab, he has built a diverse, interdisciplinary community dedicated to innovation for the benefit of humanity.

For him, he says, the EPICS in IEEE award carries profound meaning: “Receiving this award validates my deepest conviction in humanitarian technology research and strengthens my commitment to service‑learning education.”

His students echo those sentiments. One team member said “Professor Shaghaghi is an engine of progress who keeps forging ahead.”

Through his leadership, Shaghaghi has created an enduring model of mentorship, innovation, and community partnership that is helping to shape the next generation of socially responsible engineers.

Elizabeth Vidal-Duarte is celebrated for her impactful mentorship and leadership in expanding EPICS in IEEE engagement across Peru and IEEE Region 9 (Latin America and Caribbean). Vidal-Duarte, a research professor at San Agustin National University Arequipa, in Peru, is a faculty advisor and technical mentor for two EPICS in IEEE projects. She encouraged students to apply to the EPICS program, helped them identify community needs, and supported them in crafting proposals grounded in service‑learning principles.

Under her leadership, the students developed a functional soft robotic glove used at Clínica San Juan de Dios to help patients improve their fine-motor skills. The clinic’s therapists use the device to measure the range of motion of joints at the beginning and end of each patient’s therapy session to improve their assessments. Compared with traditional manual measurements using a goniometer, the glove significantly reduces evaluation time and enables digitally recorded data, improving clinical efficiency and decision-making.

The second project is an emotion‑recognition system for people with visual impairment. The AI‑powered wearable helps recognize a person’s emotions through real‑time facial‑expression detection and haptic feedback.

The project has resulted in the “Emotion-Aware Assistive System With Wearable Haptic Feedback for Visual Impairment” research paper, which is to be presented at the IEEE International Symposium on Computer-Based Medical Systems, to be held from 3 to 5 June in Limassol, Cyprus.

Vidal-Duarte’s mentorship extends beyond the classroom. She visits rehabilitation centers and clinics to find people with visual impairments to ensure that the technologies she is helping to develop meet their needs.

“EPICS in IEEE has moved me beyond teaching concepts to truly living engineering as a tool for human impact,” Vidal-Duarte says. “Watching students grow as they design solutions that improve lives has been both inspiring and deeply humbling.”

Throughout the development of both projects, Vidal-Duarte provided sustained technical and organizational guidance, helping students define requirements, structure work plans, and overcome challenges in prototyping, testing, and validation.

Reflecting on the broader impact of EPICS, she says the program has given her “more than methodologies and tools—it has given me perspective, purpose, and a global community that constantly challenges me to grow as a mentor and as a human being.”

Her mentorship fostered not only technical excellence but also empathy, ethical awareness, and professional maturity among her students, she says. She guided them in preparing articles for submission to IEEE conferences, interdisciplinary collaboration, and hands-on fieldwork that bridged theory and real‑world constraints.

“Her constant support, her belief in each student’s potential, and her commitment to developing leaders who make a difference define [her] as a faculty advisor,” says Valentina Chabilla, an EPICS in IEEE student team member.

The EPICS recognition reflects her passion for teaching, her dedication to the community, and her impact on projects and students. Her commitment to accessible, sustainable innovation strengthened partnerships between the university and community groups, benefiting underserved populations.

“Receiving this award is both an honor and a responsibility,” she says. “It reminds me of the real impact engineering can have on people’s lives and strengthens my commitment to guiding students in creating meaningful change.”

Her leadership continues to inspire students to view engineering not just as a discipline but also as a powerful force for inclusion, dignity, and social impact.

Advancing the mission

The Excellent Contributor Award recipients exemplify the best of EPICS in IEEE. Through their leadership, they have strengthened the bridge between engineering education and community service, inspiring students to use their skills to create sustainable, real‑world impacts.

As EPICS continues to expand its global reach, the contributions of Kakay, Shaghaghi, and Vidal-Duarte serve as powerful reminders of what is possible when educators, volunteers, and students work together to improve the lives of others through engineering.

A man raises his phone as police move into a crowd. The video is shaky, loud, immediate. Within minutes, it is online. Within hours, it is everywhere. This is how accountability works now. Something happens, someone records it, and that footage can show what really happened, sometimes contradicting official accounts. It can empower citizens and create consequences for officials.

But the footage’s life cycle does not end there.

In recent months, civil liberties groups have warned that adding facial recognition to consumer smart glasses could turn everyday recording into something more troubling: real-time facial identification. It reflects a broader shift already underway, where images and videos captured for one purpose can later be searched, matched, and used for another.

An ouroboros is an ancient Egyptian symbol, a snake or dragon eating its own tail. As I began to see patterns in my broader research on surveillance corporatism and governance lag, I began using the term “surveillance ouroboros” to describe this recursive pattern of observations intended to hold power accountable becoming new input for the same surveillance infrastructure.

Facial recognition changes accountability

During the George Floyd protests in 2020, people filmed police in real time. Phones were pointed at officers, not at each other. The goal was simple: to show what the state was doing. That footage spread quickly and became part of a much larger pool of public data.

At the same time, reporting from outlets including The New York Times and BuzzFeed News showed that law enforcement agencies were using facial-recognition tools, including systems built by Clearview AI. Those systems were built from billions of images scraped from across the internet, including publicly available photos and videos.

The basic approach is now routine: People record the state, or anything else (as in the January 6 attack on the U.S. Capitol), and the state compiles that footage and data into a searchable environment, which may later be used to identify some of the same people who made the footage.

Facial-recognition systems used by law enforcement are increasingly outpacing the legal safeguards.

A 2023 Government Accountability Office review found that federal law enforcement agencies continued to expand their use of facial-recognition systems for criminal investigations despite ongoing concerns around training, privacy protections, civil-liberties safeguards, and oversight. Earlier GAO findings showed that agencies had conducted roughly 60,000 facial-recognition searches before formal training requirements were put in place for personnel using the systems.

The American Civil Liberties Union and other groups have warned that these tools could be used to identify people from images shared online, including protest-related footage. Concerns about facial recognition led some U.S. states and cities, including San Francisco and Boston, to restrict or ban government use of the technology, while federal agencies have continued to face scrutiny over how such systems are tested, deployed, and audited. A 2024 analysis published in Internet Policy Review warned that facial-recognition systems used by law enforcement are increasingly outpacing the legal safeguards meant to govern them, creating growing tensions around data protection, oversight, and proportional use.

The spy network that built itself

Surveillance used to require infrastructure. Cameras had to be installed, and data had to be collected deliberately. That is no longer the case. People carry cameras everywhere. They record constantly and upload in real time. Events are documented from multiple angles without planning or coordination. The cumulative result is a continuous stream of usable data: faces, locations, timestamps, and interactions. The Internet of Things (IoT) also waits all around us, gathering information and releasing it when people least expect it, as Andrew Guthrie Ferguson describes in a recent excerpt of his book Your Data Will Be Used Against You.

RELATED: “Sensorveillance” Turns Ordinary Life Into Evidence

Similar dynamics are emerging globally. A recent analysis in the International Journal of Law and Information Technology examined how facial-recognition systems in China and Japan are expanding faster than the legal frameworks governing them. Reporting by The Guardian described the limited legal protections around the rapid deployment of AI-assisted surveillance infrastructure across parts of Africa.

There used to be a clear distinction between surveillance and accountability. Surveillance meant the powerful watching the people; authorities tended not to share their imagery except under duress or a court order and usually after a long delay. Accountability meant the people watching the powerful and often publishing imagery immediately to head off or counteract official mischief. That distinction no longer holds. The same footage can serve both roles. A recording meant to expose misconduct can later be used to identify someone else entirely.

Surveillance ouroboros is not a future risk. It is already here.

This dynamic persists because people still need to record. In many places, it is one of the only tools available when formal accountability breaks down. When oversight institutions weaken or fail, public documentation becomes a substitute. In that environment, people turn to visibility. But that visibility comes with a cost. The more people that document, the more data that exists. The more data that exists, the easier it is to search, match, and store. Every video feeds the ouroboros. People are not feeding the system because they trust it. They are feeding it because the alternative is silence.

Most of the people in these videos are not the focus. They are in the background, passing by or standing nearby. But that distinction does not matter once the footage enters a system. Today’s facial recognition can identify even a face that passed through the corner of a frame. Someone who did nothing can still become part of a dataset without ever knowing it. As recognition systems improve, older footage becomes more useful—and invasive.

No single decision created this outcome. It emerged gradually through more cameras, better recognition, larger datasets, and easier integration. Each step made sense on its own. Together, they changed what recording means.

Public recording is still necessary. Without it, many forms of abuse would remain hidden. But recording is no longer just exposure. It is also contribution. If you published imagery or video last year, you may already have contributed to a system you have never seen but the ouroboros has.

Surveillance ouroboros is not a future risk. It is already here. Every time someone presses publish, they are doing two things at once. They are exposing power, and they are helping build the system that the powerful will later use to track the less powerful.

This article is crossposted from IEEE Spectrum’s careers newsletter. Sign up now to get insider tips, expert advice, and practical strategies, written in partnership with tech career development company Parsity and delivered to your inbox for free!

Small Startup, Mid-Size Company, or Fortune 100? The Pros and Cons

Early in my career, I walked into a shared office space on my first day as a full stack software developer and sat down between the CTO and the CEO to get onboarded. There were four of us in total. Before the day was over, I received my first assignment.

This was one of the most formative—and most stressful—experiences of my professional life. In the decade since, I have worked at half a dozen companies including Fortune 100 firms, mid-size startups, and companies you’ve probably never heard of. I have also spoken with roughly a thousand developers at various stages of their careers.

Most engineers entering the field are obsessed with landing at Google, Meta, or Amazon. But those roles represent approximately 0.6 percent of software engineering positions. For most of us, the real choice is between a small startup, a mid-size company, and a large enterprise. Each comes with tradeoffs, and your experience will differ from mine. What follows is an honest account of what you might reasonably expect.

The Small Startup

Pros

Your work actually matters. A feature you build might determine whether the company closes its next funding round. You gain exposure to the full spectrum of the business, from deployment pipelines to sales and operations and everything in between. You wear many hats out of necessity. For engineers who want to grow quickly and understand how a product is built end to end, few environments move faster.

Cons

Everything is on fire, always. Work-life balance is difficult to maintain when every release feels critical. Priorities shift without warning and culture tends to reflect the personality of whoever has the most influence in a small room. Startups optimize for speed over craft which means engineers learn to move fast but don’t always learn to build well, and that gap can follow you into your next role.

The Mid-Size Company

Pros

“So this is how a real business works.” There is process, documentation, a quality assurance function, and some form of career structure. The team is large enough to offer a diversity of experience and perspective. Stability is a myth, especially nowadays, but it is considerably more predictable than an early-stage startup.

Cons

“So this is how a real business works?” Processes that enable quality also produce friction. Access controls, approval workflows, and cross-team dependencies slow things down. The career ladder exists but it might stop at senior engineer. Without significant organizational growth, your salary and title can plateau early.

The Large Enterprise

Pros

That badge on your LinkedIn profile just bought you credibility for the next five years. Compensation at this level can be meaningfully higher, particularly when equity is included. The career ladder is long and clearly defined. Engineering practices at mature organizations tend to be more rigorous, and a well-known employer carries market value in future job searches.

Cons

It’s slow. Technology stacks often lag industry trends by several years. Political dynamics shape advancement as much as technical ability does. Skill atrophy is a risk when you spend years on a narrow slice of a legacy system. You are now a small fish in a big pond and it will be harder to get noticed.

The Roadmap I Would Take If I Could Start Over

According to a recent Stack Overflow survey, 47 percent of professional developers work at companies with fewer than 100 employees. This may surprise you because social media is dominated by engineers who work at the most well known companies on the planet.

The path most engineers imagine for themselves and the path most engineers actually walk are two very different things.

If I could do it again, here’s the path I’d take: Start at a small company to build breadth and learn how a business works across functions. This also provides some room to experiment within different roles. Next, move to a mid-size organization with a clear goal of reaching a senior or leadership role. Making a lateral move is easier than trying to get up-leveled at the next company. Finally, target a more mature company where a leadership position opens the door to meaningful equity and long-term growth (aka stocks and bonuses).

Each stop builds something the others cannot. The startup gives you range. The mid-size company gives you a taste of how larger orgs operate. The enterprise gives you leverage, credibility and maybe even some stability.

Your path will not look like mine. At a five person startup, I had no idea what I was in for. Looking back, I would not trade it. Just know what you are signing up for before you sign.

—Brian

Reclaiming Social Engineering for Good

“Social engineering” is a concept that has become associated with phishing, in which scammers manipulate people into disclosing personal information. But shaping human behavior in this way doesn’t have to have such negative effects. Systems engineer Guru Madhavan argues that we need to reclaim the term and govern the practice to defend ourselves from bad actors and benefit from social engineering’s good side.

Read more here.

Get Your Medical Mobile App Verified by IEEE

Smartphone apps are increasingly used to help manage medical conditions, but many of these have not been verified by any regulatory agencies. To help ensure these apps are credible, the IEEE Standards Association recently launched a directory listing apps that have been vetted by experts for technical soundness, ethical design, data security and privacy, and clinical efficacy. The registry will be publically available at no cost, and developers can now apply for approval.

Read more here.

Finding Success in Industry as a Chip Designer

A veteran chip designer reflects on what he learned when moving from academia to industry, where the goal changes from proof of concept to ensuring a design works reliably at scale. Differences in risk tolerance, he discovered, lead to varying approaches in the rapidly growing semiconductor industry.

Read more here.

This article is crossposted from IEEE Spectrum’s careers newsletter. Sign up now to get insider tips, expert advice, and practical strategies, written in partnership with tech career development company Parsity and delivered to your inbox for free!

Job Hopping as an Engineer: The Pros and Cons

I’ve changed jobs more times than I ever imagined I would. In the past 12 years, I’ve worked at seven different organizations. Some of those moves were forced by layoffs. Others were deliberate bets on my own trajectory.

Job hopping, done strategically, is one of the fastest ways to accelerate your compensation and reinvent your professional identity. Engineers who understand when to move and when to stay tend to out-earn and out-rank their peers who simply wait for internal recognition.

Unfortunately, most engineers either job hop too much or not enough, and both mistakes are expensive. Here are the pros and cons of job hopping as an engineer, and when to make a leap.

Pro: It’s the fastest way to grow your salary

Internal raises and external offers operate on completely different logic, and most engineers don’t fully appreciate this until they make their first move.

Within a company, compensation is anchored to your existing salary and capped by organizational pay bands. A strong performance review might get you 5 to 8 percent.

An external offer is a clean slate. The company is bidding for your market value, not adjusting from your current baseline.

My first deliberate job hop doubled my salary in a single year. A later move, at the same job title, pushed my compensation floor to a level that I never would have reached by staying put. Neither outcome was available internally. The math simply does not work in your favor when you stay.

Pro: It lets you reinvent yourself

Every new company is a chance to walk in as a slightly updated version of yourself: the version that learned something from the last place. The version that does not carry the baggage of whatever decision you made two years ago that all your coworkers still remember.

Especially when you’re early in your career, this matters. You get to reframe your experience, take on a different scope, and establish a new reputation from scratch. That kind of reset is difficult to manufacture inside the same organization.

Con: You don’t see the long-term outcome of your work

This is the part nobody talks about, and it took me years to fully appreciate it.

When I joined one company, I built a component library for a website from scratch. Starting projects from scratch is exciting, and the initial implementation held up well for the early use cases. But as the organization scaled, the limitations of my original design became apparent.

I stayed long enough to address them rather than handing that problem to someone else. That experience taught me more about software architecture than any new project ever had.

Engineers who move every 18 months only ever experience the exciting part of building something. They never experience the part where their original decisions stop working. They just repeat the exciting part on a loop, never realizing the debt they are leaving behind.

Con: You cannot job hop your way to a promotion

Above a certain level, things can change significantly.

A new employer can evaluate your past performance through interviews, portfolios, and references. What they cannot do is evaluate your future potential the way a manager who has watched you grow over two or three years can. If you arrive as a senior engineer, you will almost certainly be hired as one.

The promotions that actually changed my career trajectory—from senior to staff engineer, then engineering manager—all happened at one organization over four years. Those transitions required someone to observe my growth over time and make a bet on where I was headed next. That kind of credibility cannot be transferred on a resume.

So when should you actually leave?

The threshold I use is straightforward. If I have produced at least one measurable, clearly definable outcome at an organization, I have a reasonable basis for leaving. Impact, not tenure, is my unit of measure.

I personally think that moving deliberately while early in your career will build a strong compensation baseline.

Then become selective.

Find an environment where real growth is available and stay long enough to build the credibility that job hopping cannot manufacture. Neither constant movement nor blind loyalty is the answer. The question worth asking at every stage is simple: Have I produced something meaningful here yet? If the answer is no, stay. If yes, it might be time to decide what’s next.

—Brian

The USC Professor Who Pioneered Socially Assistive Robotics

What if robots didn’t just help us with physical tasks? USC Professor Maja Matarić helped define the era of socially assistive robotics, designed to provide personalized therapy and care through social interactions. Despite her influence in the field now, the award-winning roboticist didn’t see herself as an engineer at first.

Read more here.

Steve Jobs’ Wilderness Years Shaped His Success as Apple CEO

Steve Jobs is best known as the co-founder and CEO of Apple. But the 12 years he spent away from the company taught him the lessons necessary for his success. A new book tells the forgotten story of Jobs’ “wilderness” years and what he learned while at NeXT Computer. IEEE Spectrum spoke to the book’s author about Apple’s most iconic CEO and the company’s future as it prepares for new leadership under John Ternus.

Read more here.

Learn What It Takes to Become a Cybersecurity Consultant

Cybersecurity consultants have never been more in demand, with data breaches and attacks costing organizations more than US $10 trillion annually to repair. To help you find the skills you need to stand out in the cybersecurity job market, the IEEE Computer Society offers a “What Makes a Great Cybersecurity Consultant” guide. It includes advice from experts, a list of certifications to pursue, and information on key cybersecurity conferences.

Read more here.

This article is crossposted from IEEE Spectrum’s careers newsletter. Sign up now to get insider tips, expert advice, and practical strategies, written in partnership with tech career development company Parsity and delivered to your inbox for free!

The CS Degree Isn’t Dead. The Entry-Level Pipeline Is

There is no shortage of people telling recent engineering graduates that their degree was a mistake and that AI is coming for their jobs before they even land one. I respectfully disagree.

I have been a software engineer for 12 years, done well over 100 interviews on both sides of the table, and run Parsity, an AI engineering program. A few patterns emerge consistently in who actually breaks through in today’s job market. Here’s why I think the job market isn’t as dire as it looks, and what I would do if I were looking for my first tech job.

The Numbers Need Context

The Federal Reserve Bank of New York recently placed unemployment for recent CS graduates in the United States at 6.1 percent, with computer engineering graduates at 7.5 percent. Compared to philosophy majors at 3.2 percent and art history graduates at 3.0 percent, those figures look alarming. They require more context than most headlines provide.

When researchers factor in underemployment (graduates working jobs that don’t require a college degree), then engineers are doing relatively well, coming in below 20 percent, against a 42 percent average across all recent graduates. Many majors reporting lower unemployment are achieving that figure by accepting work entirely unrelated to their field. Scored across unemployment, underemployment, and early-career earnings together, CS and computer engineering still rank among the top fields for overall labor market outcomes.

The degree is not the problem. The hiring pipeline is. Job postings labeled “entry-level software engineer” grew roughly 47 percent between late 2023 and late 2024, while actual hiring into those roles dropped approximately 73 percent in the same window. So-called “ghost jobs,” used to create an illusion of company growth, are everywhere. This makes the front door harder to find, but it exists.

Here Is What To Do About It

Do a broad search of your (real-life) network. Roughly 26 percent of job offers come through referrals. Look at your actual network—classmates, professors, past internship contacts, relatives—and identify people at companies that might be hiring. The goal is a warm introduction to someone who is or knows a decision maker. One introduction carries more weight than a hundred cold applications through a portal.

Find symmetric risk. A junior engineer is a risky hire by definition. A startup carries a matching risk profile, meaning potentially lower compensation, no certainty of longevity, and higher performance expectations. But that shared risk creates mutual interest. The learning curve is steep, the exposure is broad, and the track record transfers directly. For engineers whose longer-term goal is a large organization, a startup is not a detour. It can be how you build the experience those organizations eventually want to see. The first job is for validation and learning. It is not a life sentence.

Manufacture experience rather than waiting for it. Employers want experience but will not hire you to get it. The way through is to create it: a deployed project, an open-source contribution, building something real for a small business or family member. Recruiters are skeptical of toy projects. A deployed application solving a real problem, combined with the ability to talk clearly about the decisions you made and why, still moves the needle.

Gain practical AI engineering skills, not just AI tool fluency. Using Cursor or Copilot is now a baseline expectation. What differentiates candidates is going one level deeper. Most working engineers, including senior ones, have not built a RAG pipeline or designed a multi-agent system. Understanding how to chunk documents, generate embeddings, store and query them from a vector database, and wire it into a production application puts a candidate ahead of a significant portion of the market on a skill in rapidly growing demand. AI and data science roles grew 163 percent in job postings in 2025. The engineers who understand how these systems actually work, not just how to prompt them, are in the shortest supply.

Stop optimizing around conditions you cannot predict. Nobody anticipated the 2021 hiring boom. Nobody predicted this correction. Build durable skills. The demand for engineers who can reason clearly about systems is not going away. Where you start is not where you end.

—Brian

Meta and Microsoft have joined the layoff tsunami. Is AI really to blame?

More major workforce reductions are on the horizon at Big Tech companies: Meta announced it will cut 10 percent of its workforce, or about 8,000 employees, and Microsoft plans to offer buyouts for 7 percent of its U.S. employees in a voluntary retirement program. The cuts are understood by many to be linked to AI. But is AI really to blame? For The Conversation, two academics at the University of Sydney give their two cents.

Read more here.

This Roboticist-Turned-Teacher Built a Life-Size Replica of ENIAC

Tom Burick got his start as a roboticist. But when a financial downturn forced him to close his robotics business, he thought of the effect teachers had on his life and decided to pay it forward. Burick now works as a technology instructor at a school for students with autism, where he recently led a project building a full-scale replica of ENIAC, an historic computer celebrating its 80th anniversary this year.

Read more here.

Proposed Chinese Robot Ban is Latest U.S. Tech Sovereignty Move

Across several industries, the United States has been moving toward limiting the use of sensitive technology made in China. Now, legislation has been introduced to extend the trend to ground robots, including humanoids, dogs, and crawlers. This could benefit some U.S.-based robotics firms—but many of these companies still rely on Chinese-made components. “The U.S. robotics industry is in a pickle,” writes Spectrum tech policy editor Lucas Laursen.

Read more here.

This article is brought to you by AGILINK.

Throughout the exhibition hall at the 2026 IEEE International Conference on Robotics (ICRA), in Vienna, one demonstration seemed to attract a disproportionate amount of attention.

Two robotic hands were making a balloon dog. Slowly and deliberately, the robot twisted a long balloon into loops, bends, and joints without popping it. Visitors stopped, watched, and often returned with colleagues to watch again.

AGILINK’s balloon dog demonstration draws a crowd at ICRA 2026.AGILINK

At first glance, the demonstration appeared almost playful. Among roboticists, however, balloon twisting is widely recognized as an unusually difficult manipulation task.

A balloon is lightweight, highly deformable, slippery, and extremely sensitive to force. Every twist changes its geometry and internal pressure, turning a seemingly simple activity into a continuously changing physical interaction problem.

Humans navigate those changes almost intuitively. While making a balloon animal, people rarely think consciously about force regulation, slip prevention, or contact stability. They simply adjust.

For robots, those adjustments remain remarkably difficult. The challenge is not merely moving fingers to the right positions. The harder part is maintaining stable interaction while the object itself is changing.

Highlights from AGILINK’s ICRA 2026 demonstrations, including visuotactile sensing, in-hand manipulation, balloon-animal shaping, and other contact-rich tasks enabled by the company’s latest OmniHand platform.AGILINK

That distinction helps explain why the balloon dog drew so much attention in Vienna. What appeared to be a dexterity demonstration was, in many ways, a demonstration about contact itself.

As robotic manipulation continues to advance, a growing number of researchers are arriving at a similar conclusion: many of the hardest problems in robotics begin only after contact occurs.

Motion and Contact Intelligence for Robot Manipulation

Balloon twisting combines two challenges that robotics has traditionally struggled to solve simultaneously: long-horizon task execution and contact-rich manipulation.

The first concerns motion.

A balloon dog is not created through a single grasp or twist. It emerges through a carefully ordered sequence of manipulations, each setting the conditions for what follows. A small rotational error introduced early may appear insignificant at first, yet several steps later it can prevent the final structure from forming altogether.

In that sense, balloon twisting is a long-horizon task. Success depends not only on performing individual actions correctly, but also on preserving the future feasibility of the entire manipulation process.

To address this challenge, AGILINK began by collecting demonstrations from professional balloon artists. Human actions were mapped onto robotic hands to establish an initial manipulation policy. But successful demonstrations alone were insufficient.

In practice, some of the most valuable learning occurred when execution began to drift toward failure. Whenever instability emerged, human operators intervened and corrected the manipulation in real time. Those interventions were recorded and incorporated into reinforcement-learning cycles, allowing the system to learn not only how successful demonstrations unfold, but also how experienced operators recover when things start to go wrong.

Through this process, the robot gradually acquired the capabilities required for long-horizon task execution—a collection of abilities that AGILINK groups under the term motion intelligence: the ability to generate actions, coordinate bimanual behaviors, and execute extended manipulation sequences under real-world uncertainty.

OmniHand 3 Ultra-M on display at ICRA 2026.AGILINK

Yet motion alone does not explain why balloon twisting remains difficult. The second challenge is contact.

The robot must continuously regulate force, adjust contact locations, and respond to subtle changes in the object’s state. These decisions are difficult to encode through explicit rules. Even skilled human operators often rely on tactile intuition developed through experience rather than consciously articulated strategies.

Analysis of those interventions revealed that many failures did not originate from incorrect action sequences, but from the breakdown of contact itself.

To better capture those interaction dynamics, AGILINK collected contact-centric intervention data and incorporated those interactions into reinforcement-learning training. Rather than learning only which motions to perform, the system also learned how humans maintain stability when contact conditions begin to deteriorate.

AGILINK describes this capability as contact intelligence: the ability to establish, maintain, and adapt physical interaction as force distribution, friction, deformation, and contact geometry continuously evolve.

The distinction between the two capabilities is subtle but important. Motion intelligence determines what the robot intends to do. Contact intelligence determines whether it can continue doing it. For balloon twisting, both are necessary. One provides the sequence of actions. The other keeps those actions physically viable.

YouTuber KhanFlicks follows OmniHand’s motions while learning to fold a balloon dog at the AGILINK booth.AGILINK

Between a balloon slipping away and a balloon bursting lies a narrow region of stability. Successful manipulation depends on finding that region—and remaining within it throughout the task.

Introducing the OmniHand 3 Ultra-M Dexterous Hand

The balloon dog demonstration showcased a manipulation capability. It also revealed a broader question. How much contact intelligence can be achieved through learning alone? A robot can only regulate what it can perceive. It can only respond as quickly as its hardware allows.

As manipulation tasks become increasingly complex, researchers are finding that progress depends not only on better policies, but also on richer sensing and faster physical response.

That realization formed the backdrop for AGILINK’s second major announcement at ICRA 2026. Alongside the balloon dog demonstration, the company introduced the OmniHand 3 Ultra-M.

OmniHand 3 Ultra-M closely matches the size of an adult human hand.AGILINK

The two exhibits represented different stages of the same technological trajectory. If the balloon dog demonstrated what contact intelligence can already accomplish today, Ultra-M was designed to explore what contact intelligence may require next.

Building Hardware for Contact Intelligence

Roughly the size of an adult human hand, the OmniHand 3 Ultra-M integrates 20 active degrees of freedom within a human-scale form factor.

Its most distinctive feature is a fully direct-drive architecture. By adopting direct-drive actuation throughout the system, the hand is designed to enable faster and more transparent force regulation and higher force-control bandwidth, enabling faster response as contact conditions change. For contact-rich manipulation, responsiveness can be as important as sensing itself.

By adopting direct-drive actuation throughout the system, the OmniHand 3 Ultra-M is designed to enable faster and more transparent force regulation and higher force-control bandwidth, enabling faster response as contact conditions change.

The platform also incorporates tactile sensing across nearly the entire hand. Each fingertip contains a miniature vision-based tactile sensor, while more than 300 three-dimensional tactile sensing points are distributed throughout the palm. Together, they provide information not only about where contact occurs, but how contact is evolving.

The system is designed to estimate pressure distribution, shear forces, local deformation, slip tendencies, and other interaction dynamics that often remain invisible to conventional position-based control systems.

According to AGILINK’s tests, individual sensors achieve force resolution of approximately 0.005 N—roughly equivalent to detecting the weight of a sheet of paper resting on a fingertip. Spatial resolution reaches approximately 0.04 mm, while sensing density approaches 50,000 sensing points per square centimeter.

OmniHand 3 Ultra-M recognizes feather texture through vision-based tactile sensing.AGILINK

For dexterous robots, contact has traditionally been a largely hidden process. Ultra-M is designed to make that process more observable.

Rather than simply detecting that contact has occurred, the system attempts to resolve where interaction is happening, how forces are distributed, whether instability is beginning to emerge, and how manipulation strategies should adapt in response.

The balloon dog offered a glimpse of what contact intelligence can already accomplish. Ultra-M explores a different question: what capabilities may be required to push contact intelligence further?

The Physical World Remains the Hardest Benchmark

The significance of contact intelligence extends far beyond balloon animals. Many tasks that continue to resist automation involve unstable or deformable interaction: cable insertion, garment handling, flexible packaging, delicate assembly, connector mating, tool use, and household manipulation.

These tasks are difficult not because robots cannot reach the correct location, but because maintaining stable interaction after contact begins remains extraordinarily hard.

For decades, robotics achieved many of its successes by reducing uncertainty. Factories were engineered to make robotic motion predictable, repeatable, and highly structured. The physical world behaves differently.

A growing share of robotics research is shifting toward interaction itself—understanding how robots can establish, maintain, and adapt physical contact within environments that remain fundamentally unpredictable.

Objects shift. Materials deform. Friction changes. Contact evolves. Real environments rarely follow scripts. Seen through that lens, the balloon dog was never really about the balloon dog. What attracted attention at ICRA was not simply a visually impressive demonstration, but what it revealed: intelligence in the physical world is ultimately measured through interaction.

As motion generation continues to mature, a growing share of robotics research is shifting toward interaction itself—understanding how robots can establish, maintain, and adapt physical contact within environments that remain fundamentally unpredictable.

For robots moving beyond structured environments and into less predictable real-world settings, managing contact may become as important as motion itself.

New York City was the backdrop of this year’s IEEE Honors Ceremony, held on 24 April.

The event celebrates engineering pioneers who have developed technologies that have changed how people connect and learn about the world. This year’s celebrants included the engineers behind innovations such as text-to-donate technology, AI-powered diagnostic tools, and the graphics processing unit, among many others.

Prior to the Honors Ceremony, IEEE hosted a forum on 23 April for a select group of early-career achievers to exchange ideas and experiences with laureates and awardees, speakers, and IEEE leaders. Attendees from around the world, working in a variety of technical areas, shared their journeys and explored the intersections of technologies, disciplines, and missions.

The event culminated in Friday evening’s black tie Honors Ceremony, where IEEE celebrated medal laureates, including Jensen Huang, who received IEEE’s highest recognition, the IEEE Medal of Honor. Huang is a cofounder of Nvidia and its chief executive.

“IEEE has always been a home to those who see the future before others see it,” Mary Ellen Randall, IEEE president and CEO, said in her welcome speech.

Video highlights and photos from the event are available on the IEEE Awards website.

Exploring mission-driven tech and AI in art

Friday morning began with a conversation between Randall and Marian Croak, the recipient of this year’s IEEE Founders Medal. Croak was honored for “leadership in communication networks, including acceleration of digital equity, responsible artificial intelligence, and the promotion of diversity and inclusion.”

Croak, who serves as vice president of engineering at Google, headquartered in Mountain View, Calif., pioneered Voice over Internet Protocol (VoIP) technologies. When a person speaks into a telephone, VoIP converts their voice into digital signals that are transmitted over the Internet rather than traditional phone lines. Her work enabled audio and video conferencing. She also developed text-to-donate technology to raise money for those affected by Hurricane Katrina, which devastated New Orleans in 2005. The technology enables customers to donate money to a charity via their mobile service provider, which then bills them.

“Empathy has always been a driving force in the engineering that I’ve done,” she said.

She shared advice on how to stay creative: “Get out of the office. Go to an art museum, exercise, or play with children.” Croak said her grandchildren inspire her.

An inside look at microchips

During Friday evening’s Honors Ceremony cocktail hour, attendees explored the history of microchips at the IEEE Global Museum’s Microchips That Shook the World exhibit. The Global Museum, an IEEE History and Heritage program, develops traveling and digital exhibits focused on the history of technology. The museum’s mission is to promote awareness of how technological progress unfolds over generations and how engineers and researchers build on past achievements to benefit humanity.

Drawing from IEEE Spectrum’s Chip Hall of Fame, the Microchips That Shook the World exhibit conveys the roles integrated circuits play in fields such as signal processing, audio engineering, and telecommunications.

Co-curators Stephen Cass, Spectrum’s special projects editor, and Daniel Mitchell, the IEEE senior historian, served as onsite docents for guests. The Commodore 64, one of the artifacts on display, brought up many treasured childhood memories for guests who used the home computer. The exhibit also featured a preview of IEEE’s immersive video project “Inside the Microchip,” which delves beneath the silicon surface of the Nvidia NV20 microchip thanks to forensic photography and sophisticated computer-generated renders. The video, which will be released later this year, aims to teach preuniversity students about the technology.

The daytime program also spotlighted AI’s use in the visual arts. Kathleen Kramer, the 2025 IEEE president, interviewed artist Refik Anadol, who is scheduled to open an AI art museum on 20 June in Los Angeles. Dataland’s exhibits are powered by an open-access model developed by Anadol’s studio.

For the museum’s first exhibition, “Machine Dreams: Rainforest,” the model collected visual data about the natural world from the Smithsonian National Museum of Natural History, London’s Natural History Museum, and the Cornell Lab of Ornithology, with their permission. The information, including up to a half billion images, will form the basis for a variety of AI-produced art, Anadol said.

Anadol said he was inspired to mix AI with art by the movie Blade Runner. He said he believes “machines can become collaborators,” as “data is a form of pigment.”

Data also plays an important role in the work of artist and author Giorgia Lupi. The artist is a partner at design firm Pentagram.

Lupi said she uses data to tell stories, including chronicling her struggles with a chronic illness.

“Data is an abstraction of our reality,” she said.

One of her recent projects, “A Data Love Letter to the Subway,” was shown last year in the Dey Street Passageway in New York City. The video was made using data from the Metropolitan Transportation Authority about each train line, including timetables, ridership, and people’s travel habits. Based on the information Lupi gathered, she documented how commuters traveling on different subway lines encountered one another without realizing it.

By exploring data on this year’s IEEE award recipients, she collaborated with IEEE to create an animated video illustrating the shared pathways and collaborations among the honorees. It debuted at the Honors Ceremony.

Honoring engineering giants

The Honors Ceremony, held at Cipriani 42nd Street, recognized more than 20 laureates and innovators.

More than 92 million selfies are taken worldwide every day, PhotoAiD estimates. A selfie wouldn’t be possible without Eric Fossum’s invention of the CMOS image sensor. Developed at NASA’s Jet Propulsion Laboratory, in Pasadena, Calif., the “camera on a chip” was intended for use in space, but it is now found in smartphones, medical devices, and vehicles. Fossum, an IEEE Life Fellow, received the IEEE Jun-ichi Nishizawa Medal, which recognizes outstanding contributions to materials and device science and technology.

“Engineering is a pursuit of what must be possible. [IEEE is] the spirit, the conscience, of our profession.” —Jensen Huang, founder and CEO of Nvidia

The medal, he said, “is at the top of the IEEE staircase of being recognized by your peers.”

The IEEE Holonyak Medal for Semiconductor Optoelectronic Technologies went to Steven P. DenBaars, a professor of materials and electrical and computer engineering at the University of California, Santa Barbara. DenBaars was honored for his work in semiconductors, which laid the foundation for high-resolution LED and laser displays, modern solid-state lighting, and more.

“This work has always been a team effort...I’m excited and curious about the role gallium nitride micro LEDs will play in optical communications,” he said in his acceptance speech.

The ceremony ended with the Medal of Honor presentation to Huang, who received a standing ovation. He was recognized for his “leadership in the development of graphics processing units and their application to scientific computing and artificial intelligence.”

The IEEE honorary member donated his cash prize to fund scholarships for new graduates. The Jen-Hsun and Lori Huang Foundation matched his gift, and the additional donation is destined for IEEE TryEngineering, which provides teachers with a library of lesson plans and offers educational summer camps.

“Engineering is a pursuit of what must be possible. [IEEE is] the spirit, the conscience, of our profession,” Huang said.

The Institute is celebrating its 50th anniversary this year. Launched in 1976, the publication was designed to keep members informed about IEEE and what its constituents were doing, as well as to report on the organization’s initiatives, technical standards, products, and services.

That directive expanded over the years to include our reporting on key historical technical achievements recognized as IEEE Milestones and support for young professionals with career-guidance articles and information about educational resources.

The Institute has gone through many iterations in the past 50 years. What began as a monthly four-page insert in the print edition of IEEE Spectrum became a separate newspaper published six times a year and mailed along with Spectrum in 1977, and then a monthly publication the following year.

Today we publish all of The Institute’s articles online, with a curated selection appearing in our 16-page quarterly printed in the March, June, September, and December Spectrum issues.

To provide members with a quick summary of the latest online news, in 2003 a bimonthly newsletter, The Institute Alert, began appearing in your inbox. You also can stay up to date by following our Facebook, Instagram, and LinkedIn pages.

Although much has changed, an original subsection from 1976—“IEEE People”—has been maintained for the past five decades. We continue to celebrate IEEE members from around the world through our profiles, which are among our most popular articles.

Andrew Ng has serious street cred in artificial intelligence. He pioneered the use of graphics processing units (GPUs) to train deep learning models in the late 2000s with his students at Stanford University, cofounded Google Brain in 2011, and then served for three years as chief scientist for Baidu, where he helped build the Chinese tech giant’s AI group. So when he says he has identified the next big shift in artificial intelligence, people listen. And that’s what he told IEEE Spectrum in an exclusive Q&A.

Ng’s current efforts are focused on his company Landing AI, which built a platform called LandingLens to help manufacturers improve visual inspection with computer vision. He has also become something of an evangelist for what he calls the data-centric AI movement, which he says can yield “small data” solutions to big issues in AI, including model efficiency, accuracy, and bias.

Andrew Ng on...

• What’s next for really big models
• The career advice he didn’t listen to
• Defining the data-centric AI movement
• Synthetic data
• Why Landing AI asks its customers to do the work

The great advances in deep learning over the past decade or so have been powered by ever-bigger models crunching ever-bigger amounts of data. Some people argue that that’s an unsustainable trajectory. Do you agree that it can’t go on that way?

Andrew Ng: This is a big question. We’ve seen foundation models in NLP [natural language processing]. I’m excited about NLP models getting even bigger, and also about the potential of building foundation models in computer vision. I think there’s lots of signal to still be exploited in video: We have not been able to build foundation models yet for video because of compute bandwidth and the cost of processing video, as opposed to tokenized text. So I think that this engine of scaling up deep learning algorithms, which has been running for something like 15 years now, still has steam in it. Having said that, it only applies to certain problems, and there’s a set of other problems that need small data solutions.

When you say you want a foundation model for computer vision, what do you mean by that?

Ng: This is a term coined by Percy Liang and some of my friends at Stanford to refer to very large models, trained on very large data sets, that can be tuned for specific applications. For example, GPT-3 is an example of a foundation model [for NLP]. Foundation models offer a lot of promise as a new paradigm in developing machine learning applications, but also challenges in terms of making sure that they’re reasonably fair and free from bias, especially if many of us will be building on top of them.

What needs to happen for someone to build a foundation model for video?

Ng: I think there is a scalability problem. The compute power needed to process the large volume of images for video is significant, and I think that’s why foundation models have arisen first in NLP. Many researchers are working on this, and I think we’re seeing early signs of such models being developed in computer vision. But I’m confident that if a semiconductor maker gave us 10 times more processor power, we could easily find 10 times more video to build such models for vision.

Having said that, a lot of what’s happened over the past decade is that deep learning has happened in consumer-facing companies that have large user bases, sometimes billions of users, and therefore very large data sets. While that paradigm of machine learning has driven a lot of economic value in consumer software, I find that that recipe of scale doesn’t work for other industries.

Back to top

It’s funny to hear you say that, because your early work was at a consumer-facing company with millions of users.

Ng: Over a decade ago, when I proposed starting the Google Brain project to use Google’s compute infrastructure to build very large neural networks, it was a controversial step. One very senior person pulled me aside and warned me that starting Google Brain would be bad for my career. I think he felt that the action couldn’t just be in scaling up, and that I should instead focus on architecture innovation.

“In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data. Having 50 thoughtfully engineered examples can be sufficient to explain to the neural network what you want it to learn.”
—Andrew Ng, CEO & Founder, Landing AI

I remember when my students and I published the first NeurIPS workshop paper advocating using CUDA, a platform for processing on GPUs, for deep learning—a different senior person in AI sat me down and said, “CUDA is really complicated to program. As a programming paradigm, this seems like too much work.” I did manage to convince him; the other person I did not convince.

I expect they’re both convinced now.

Ng: I think so, yes.

Over the past year as I’ve been speaking to people about the data-centric AI movement, I’ve been getting flashbacks to when I was speaking to people about deep learning and scalability 10 or 15 years ago. In the past year, I’ve been getting the same mix of “there’s nothing new here” and “this seems like the wrong direction.”

Back to top

How do you define data-centric AI, and why do you consider it a movement?

Ng: Data-centric AI is the discipline of systematically engineering the data needed to successfully build an AI system. For an AI system, you have to implement some algorithm, say a neural network, in code and then train it on your data set. The dominant paradigm over the last decade was to download the data set while you focus on improving the code. Thanks to that paradigm, over the last decade deep learning networks have improved significantly, to the point where for a lot of applications the code—the neural network architecture—is basically a solved problem. So for many practical applications, it’s now more productive to hold the neural network architecture fixed, and instead find ways to improve the data.

When I started speaking about this, there were many practitioners who, completely appropriately, raised their hands and said, “Yes, we’ve been doing this for 20 years.” This is the time to take the things that some individuals have been doing intuitively and make it a systematic engineering discipline.

The data-centric AI movement is much bigger than one company or group of researchers. My collaborators and I organized a data-centric AI workshop at NeurIPS, and I was really delighted at the number of authors and presenters that showed up.

You often talk about companies or institutions that have only a small amount of data to work with. How can data-centric AI help them?

Ng: You hear a lot about vision systems built with millions of images—I once built a face recognition system using 350 million images. Architectures built for hundreds of millions of images don’t work with only 50 images. But it turns out, if you have 50 really good examples, you can build something valuable, like a defect-inspection system. In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data. Having 50 thoughtfully engineered examples can be sufficient to explain to the neural network what you want it to learn.

When you talk about training a model with just 50 images, does that really mean you’re taking an existing model that was trained on a very large data set and fine-tuning it? Or do you mean a brand new model that’s designed to learn only from that small data set?

Ng: Let me describe what Landing AI does. When doing visual inspection for manufacturers, we often use our own flavor of RetinaNet. It is a pretrained model. Having said that, the pretraining is a small piece of the puzzle. What’s a bigger piece of the puzzle is providing tools that enable the manufacturer to pick the right set of images [to use for fine-tuning] and label them in a consistent way. There’s a very practical problem we’ve seen spanning vision, NLP, and speech, where even human annotators don’t agree on the appropriate label. For big data applications, the common response has been: If the data is noisy, let’s just get a lot of data and the algorithm will average over it. But if you can develop tools that flag where the data’s inconsistent and give you a very targeted way to improve the consistency of the data, that turns out to be a more efficient way to get a high-performing system.

“Collecting more data often helps, but if you try to collect more data for everything, that can be a very expensive activity.”
—Andrew Ng

For example, if you have 10,000 images where 30 images are of one class, and those 30 images are labeled inconsistently, one of the things we do is build tools to draw your attention to the subset of data that’s inconsistent. So you can very quickly relabel those images to be more consistent, and this leads to improvement in performance.

Could this focus on high-quality data help with bias in data sets? If you’re able to curate the data more before training?

Ng: Very much so. Many researchers have pointed out that biased data is one factor among many leading to biased systems. There have been many thoughtful efforts to engineer the data. At the NeurIPS workshop, Olga Russakovsky gave a really nice talk on this. At the main NeurIPS conference, I also really enjoyed Mary Gray’s presentation, which touched on how data-centric AI is one piece of the solution, but not the entire solution. New tools like Datasheets for Datasets also seem like an important piece of the puzzle.

One of the powerful tools that data-centric AI gives us is the ability to engineer a subset of the data. Imagine training a machine-learning system and finding that its performance is okay for most of the data set, but its performance is biased for just a subset of the data. If you try to change the whole neural network architecture to improve the performance on just that subset, it’s quite difficult. But if you can engineer a subset of the data you can address the problem in a much more targeted way.

When you talk about engineering the data, what do you mean exactly?

Ng: In AI, data cleaning is important, but the way the data has been cleaned has often been in very manual ways. In computer vision, someone may visualize images through a Jupyter notebook and maybe spot the problem, and maybe fix it. But I’m excited about tools that allow you to have a very large data set, tools that draw your attention quickly and efficiently to the subset of data where, say, the labels are noisy. Or to quickly bring your attention to the one class among 100 classes where it would benefit you to collect more data. Collecting more data often helps, but if you try to collect more data for everything, that can be a very expensive activity.

For example, I once figured out that a speech-recognition system was performing poorly when there was car noise in the background. Knowing that allowed me to collect more data with car noise in the background, rather than trying to collect more data for everything, which would have been expensive and slow.

Back to top

What about using synthetic data, is that often a good solution?

Ng: I think synthetic data is an important tool in the tool chest of data-centric AI. At the NeurIPS workshop, Anima Anandkumar gave a great talk that touched on synthetic data. I think there are important uses of synthetic data that go beyond just being a preprocessing step for increasing the data set for a learning algorithm. I’d love to see more tools to let developers use synthetic data generation as part of the closed loop of iterative machine learning development.

Do you mean that synthetic data would allow you to try the model on more data sets?

Ng: Not really. Here’s an example. Let’s say you’re trying to detect defects in a smartphone casing. There are many different types of defects on smartphones. It could be a scratch, a dent, pit marks, discoloration of the material, other types of blemishes. If you train the model and then find through error analysis that it’s doing well overall but it’s performing poorly on pit marks, then synthetic data generation allows you to address the problem in a more targeted way. You could generate more data just for the pit-mark category.

“In the consumer software Internet, we could train a handful of machine-learning models to serve a billion users. In manufacturing, you might have 10,000 manufacturers building 10,000 custom AI models.”
—Andrew Ng

Synthetic data generation is a very powerful tool, but there are many simpler tools that I will often try first. Such as data augmentation, improving labeling consistency, or just asking a factory to collect more data.

Back to top

To make these issues more concrete, can you walk me through an example? When a company approaches Landing AI and says it has a problem with visual inspection, how do you onboard them and work toward deployment?

Ng: When a customer approaches us we usually have a conversation about their inspection problem and look at a few images to verify that the problem is feasible with computer vision. Assuming it is, we ask them to upload the data to the LandingLens platform. We often advise them on the methodology of data-centric AI and help them label the data.

One of the foci of Landing AI is to empower manufacturing companies to do the machine learning work themselves. A lot of our work is making sure the software is fast and easy to use. Through the iterative process of machine learning development, we advise customers on things like how to train models on the platform, when and how to improve the labeling of data so the performance of the model improves. Our training and software supports them all the way through deploying the trained model to an edge device in the factory.

How do you deal with changing needs? If products change or lighting conditions change in the factory, can the model keep up?

Ng: It varies by manufacturer. There is data drift in many contexts. But there are some manufacturers that have been running the same manufacturing line for 20 years now with few changes, so they don’t expect changes in the next five years. Those stable environments make things easier. For other manufacturers, we provide tools to flag when there’s a significant data-drift issue. I find it really important to empower manufacturing customers to correct data, retrain, and update the model. Because if something changes and it’s 3 a.m. in the United States, I want them to be able to adapt their learning algorithm right away to maintain operations.

In the consumer software Internet, we could train a handful of machine-learning models to serve a billion users. In manufacturing, you might have 10,000 manufacturers building 10,000 custom AI models. The challenge is, how do you do that without Landing AI having to hire 10,000 machine learning specialists?

So you’re saying that to make it scale, you have to empower customers to do a lot of the training and other work.

Ng: Yes, exactly! This is an industry-wide problem in AI, not just in manufacturing. Look at health care. Every hospital has its own slightly different format for electronic health records. How can every hospital train its own custom AI model? Expecting every hospital’s IT personnel to invent new neural-network architectures is unrealistic. The only way out of this dilemma is to build tools that empower the customers to build their own models by giving them tools to engineer the data and express their domain knowledge. That’s what Landing AI is executing in computer vision, and the field of AI needs other teams to execute this in other domains.

Is there anything else you think it’s important for people to understand about the work you’re doing or the data-centric AI movement?

Ng: In the last decade, the biggest shift in AI was a shift to deep learning. I think it’s quite possible that in this decade the biggest shift will be to data-centric AI. With the maturity of today’s neural network architectures, I think for a lot of the practical applications the bottleneck will be whether we can efficiently get the data we need to develop systems that work well. The data-centric AI movement has tremendous energy and momentum across the whole community. I hope more researchers and developers will jump in and work on it.

Back to top

This article appears in the April 2022 print issue as “Andrew Ng, AI Minimalist.”

The end of Moore’s Law is looming. Engineers and designers can do only so much to miniaturize transistors and pack as many of them as possible into chips. So they’re turning to other approaches to chip design, incorporating technologies like AI into the process.

Samsung, for instance, is adding AI to its memory chips to enable processing in memory, thereby saving energy and speeding up machine learning. Speaking of speed, Google’s TPU V4 AI chip has doubled its processing power compared with that of its previous version.

But AI holds still more promise and potential for the semiconductor industry. To better understand how AI is set to revolutionize chip design, we spoke with Heather Gorr, senior product manager for MathWorks’ MATLAB platform.

How is AI currently being used to design the next generation of chips?

Heather Gorr: AI is such an important technology because it’s involved in most parts of the cycle, including the design and manufacturing process. There’s a lot of important applications here, even in the general process engineering where we want to optimize things. I think defect detection is a big one at all phases of the process, especially in manufacturing. But even thinking ahead in the design process, [AI now plays a significant role] when you’re designing the light and the sensors and all the different components. There’s a lot of anomaly detection and fault mitigation that you really want to consider.

Heather GorrMathWorks

Then, thinking about the logistical modeling that you see in any industry, there is always planned downtime that you want to mitigate; but you also end up having unplanned downtime. So, looking back at that historical data of when you’ve had those moments where maybe it took a bit longer than expected to manufacture something, you can take a look at all of that data and use AI to try to identify the proximate cause or to see something that might jump out even in the processing and design phases. We think of AI oftentimes as a predictive tool, or as a robot doing something, but a lot of times you get a lot of insight from the data through AI.

What are the benefits of using AI for chip design?

Gorr: Historically, we’ve seen a lot of physics-based modeling, which is a very intensive process. We want to do a reduced order model, where instead of solving such a computationally expensive and extensive model, we can do something a little cheaper. You could create a surrogate model, so to speak, of that physics-based model, use the data, and then do your parameter sweeps, your optimizations, your Monte Carlo simulations using the surrogate model. That takes a lot less time computationally than solving the physics-based equations directly. So, we’re seeing that benefit in many ways, including the efficiency and economy that are the results of iterating quickly on the experiments and the simulations that will really help in the design.

So it’s like having a digital twin in a sense?

Gorr: Exactly. That’s pretty much what people are doing, where you have the physical system model and the experimental data. Then, in conjunction, you have this other model that you could tweak and tune and try different parameters and experiments that let sweep through all of those different situations and come up with a better design in the end.

So, it’s going to be more efficient and, as you said, cheaper?

Gorr: Yeah, definitely. Especially in the experimentation and design phases, where you’re trying different things. That’s obviously going to yield dramatic cost savings if you’re actually manufacturing and producing [the chips]. You want to simulate, test, experiment as much as possible without making something using the actual process engineering.

We’ve talked about the benefits. How about the drawbacks?

Gorr: The [AI-based experimental models] tend to not be as accurate as physics-based models. Of course, that’s why you do many simulations and parameter sweeps. But that’s also the benefit of having that digital twin, where you can keep that in mind—it’s not going to be as accurate as that precise model that we’ve developed over the years.

Both chip design and manufacturing are system intensive; you have to consider every little part. And that can be really challenging. It’s a case where you might have models to predict something and different parts of it, but you still need to bring it all together.

One of the other things to think about too is that you need the data to build the models. You have to incorporate data from all sorts of different sensors and different sorts of teams, and so that heightens the challenge.

How can engineers use AI to better prepare and extract insights from hardware or sensor data?

Gorr: We always think about using AI to predict something or do some robot task, but you can use AI to come up with patterns and pick out things you might not have noticed before on your own. People will use AI when they have high-frequency data coming from many different sensors, and a lot of times it’s useful to explore the frequency domain and things like data synchronization or resampling. Those can be really challenging if you’re not sure where to start.

One of the things I would say is, use the tools that are available. There’s a vast community of people working on these things, and you can find lots of examples [of applications and techniques] on GitHub or MATLAB Central, where people have shared nice examples, even little apps they’ve created. I think many of us are buried in data and just not sure what to do with it, so definitely take advantage of what’s already out there in the community. You can explore and see what makes sense to you, and bring in that balance of domain knowledge and the insight you get from the tools and AI.

What should engineers and designers consider when using AI for chip design?

Gorr: Think through what problems you’re trying to solve or what insights you might hope to find, and try to be clear about that. Consider all of the different components, and document and test each of those different parts. Consider all of the people involved, and explain and hand off in a way that is sensible for the whole team.

How do you think AI will affect chip designers’ jobs?

Gorr: It’s going to free up a lot of human capital for more advanced tasks. We can use AI to reduce waste, to optimize the materials, to optimize the design, but then you still have that human involved whenever it comes to decision-making. I think it’s a great example of people and technology working hand in hand. It’s also an industry where all people involved—even on the manufacturing floor—need to have some level of understanding of what’s happening, so this is a great industry for advancing AI because of how we test things and how we think about them before we put them on the chip.

How do you envision the future of AI and chip design?

Gorr: It’s very much dependent on that human element—involving people in the process and having that interpretable model. We can do many things with the mathematical minutiae of modeling, but it comes down to how people are using it, how everybody in the process is understanding and applying it. Communication and involvement of people of all skill levels in the process are going to be really important. We’re going to see less of those superprecise predictions and more transparency of information, sharing, and that digital twin—not only using AI but also using our human knowledge and all of the work that many people have done over the years.

Quantum computing is a devilishly complex technology, with many technical hurdles impacting its development. Of these challenges two critical issues stand out: miniaturization and qubit quality.

IBM has adopted the superconducting qubit road map of reaching a 1,121-qubit processor by 2023, leading to the expectation that 1,000 qubits with today’s qubit form factor is feasible. However, current approaches will require very large chips (50 millimeters on a side, or larger) at the scale of small wafers, or the use of chiplets on multichip modules. While this approach will work, the aim is to attain a better path toward scalability.

Now researchers at MIT have been able to both reduce the size of the qubits and done so in a way that reduces the interference that occurs between neighboring qubits. The MIT researchers have increased the number of superconducting qubits that can be added onto a device by a factor of 100.

“We are addressing both qubit miniaturization and quality,” said William Oliver, the director for the Center for Quantum Engineering at MIT. “Unlike conventional transistor scaling, where only the number really matters, for qubits, large numbers are not sufficient, they must also be high-performance. Sacrificing performance for qubit number is not a useful trade in quantum computing. They must go hand in hand.”

The key to this big increase in qubit density and reduction of interference comes down to the use of two-dimensional materials, in particular the 2D insulator hexagonal boron nitride (hBN). The MIT researchers demonstrated that a few atomic monolayers of hBN can be stacked to form the insulator in the capacitors of a superconducting qubit.

Just like other capacitors, the capacitors in these superconducting circuits take the form of a sandwich in which an insulator material is sandwiched between two metal plates. The big difference for these capacitors is that the superconducting circuits can operate only at extremely low temperatures—less than 0.02 degrees above absolute zero (-273.15 °C).

Superconducting qubits are measured at temperatures as low as 20 millikelvin in a dilution refrigerator.Nathan Fiske/MIT

In that environment, insulating materials that are available for the job, such as PE-CVD silicon oxide or silicon nitride, have quite a few defects that are too lossy for quantum computing applications. To get around these material shortcomings, most superconducting circuits use what are called coplanar capacitors. In these capacitors, the plates are positioned laterally to one another, rather than on top of one another.

As a result, the intrinsic silicon substrate below the plates and to a smaller degree the vacuum above the plates serve as the capacitor dielectric. Intrinsic silicon is chemically pure and therefore has few defects, and the large size dilutes the electric field at the plate interfaces, all of which leads to a low-loss capacitor. The lateral size of each plate in this open-face design ends up being quite large (typically 100 by 100 micrometers) in order to achieve the required capacitance.

In an effort to move away from the large lateral configuration, the MIT researchers embarked on a search for an insulator that has very few defects and is compatible with superconducting capacitor plates.

“We chose to study hBN because it is the most widely used insulator in 2D material research due to its cleanliness and chemical inertness,” said colead author Joel Wang, a research scientist in the Engineering Quantum Systems group of the MIT Research Laboratory for Electronics.

On either side of the hBN, the MIT researchers used the 2D superconducting material, niobium diselenide. One of the trickiest aspects of fabricating the capacitors was working with the niobium diselenide, which oxidizes in seconds when exposed to air, according to Wang. This necessitates that the assembly of the capacitor occur in a glove box filled with argon gas.

While this would seemingly complicate the scaling up of the production of these capacitors, Wang doesn’t regard this as a limiting factor.

“What determines the quality factor of the capacitor are the two interfaces between the two materials,” said Wang. “Once the sandwich is made, the two interfaces are “sealed” and we don’t see any noticeable degradation over time when exposed to the atmosphere.”

This lack of degradation is because around 90 percent of the electric field is contained within the sandwich structure, so the oxidation of the outer surface of the niobium diselenide does not play a significant role anymore. This ultimately makes the capacitor footprint much smaller, and it accounts for the reduction in cross talk between the neighboring qubits.

“The main challenge for scaling up the fabrication will be the wafer-scale growth of hBN and 2D superconductors like [niobium diselenide], and how one can do wafer-scale stacking of these films,” added Wang.

Wang believes that this research has shown 2D hBN to be a good insulator candidate for superconducting qubits. He says that the groundwork the MIT team has done will serve as a road map for using other hybrid 2D materials to build superconducting circuits.