The transition from a purely technical expert or individual contributor position to a broader leadership role is one of the most challenging phases in a STEM career. It requires moving away from relying solely on technical excellence toward mastering systems thinking, adaptive leadership, and team alignment.

To help mid-career professionals navigate the shift, the inaugural IEEE International Leadership Conference is designed to provide attendees with practical tools to step into broader responsibility and champion an entrepreneurial mindset.

The ILC event is scheduled for 3 and 4 October in Budapest. Registration is open.

Thinking beyond technical contributions

To successfully step into a leadership role, technical professionals need to look beyond their individual output and focus on “understanding the larger system, and championing innovation by building trust and aligning new ideas with organizational goals,” says IEEE Life Senior Member Daniel Sniezek, cochair of the ILC program committee.

Because engineering decisions don’t exist in a vacuum, navigating the larger system requires recognizing how technical choices intersect with the organization’s broader business, operational, and ethical realities, Sniezek says.

By letting go of the need to be the sole technical expert and focusing instead on collaborative empowerment, he says, engineers can pivot into transformational leaders who align new initiatives with the organization’s strategic vision.

Ultimately, over the span of a career, an individual’s leadership journey evolves far beyond personal advancement to “creating a lasting legacy through the people you develop, the knowledge you share, and the innovations you inspire,” he says.

Solving the intrapreneur’s dilemma

Championing disruptive ideas within established corporate structures—sometimes called the intrapreneur’s dilemma—does not mean working against the organization. Rather, it requires emerging leaders to act like business owners instead of passive task-takers.

To begin thinking like an entrepreneur from within, professionals should shift their focus from merely executing assigned work to proactively identifying hidden areas that would create value for the company, building trust with colleagues, and presenting bold innovations as solutions to the organization’s long-term strategic goals.

That kind of self-starting entrepreneurial mindset is how leaders create opportunities out of institutional constraints.

IEEE Senior Member Deyasini Majumdar, cochair of the ILC program committee, advises professionals to exercise leadership and strategic thinking skills without being asked.

“Within the constraints of established organizational structures you can unearth a treasure trove of opportunities to innovate,” Majumdar says. “Remember: A key trait of an effective leader is to engineer solutions and lead, even in the face of difficulties.”

A multidirectional exchange

Leadership is a multidirectional exchange of ideas across generations—which is one focus of the ILC.

Although emerging leaders can gain invaluable strategic guidance from seasoned executives, the relationship is a dynamic, two-way street.

Modern leadership requires established executives to remain active learners. Addressing what senior leaders can glean from their mid-career counterparts, Majumdar emphasizes, the leaders must maintain “the openness to seek opportunities, to quickly adapt, and to learn and grow with everyone around them.”

A continuous-learning mindset is what keeps leaders agile and effective in a rapidly changing technological landscape, he says.

The mutual openness can serve as a bridge between generations.

Whether an emerging professional is making a mid-career pivot from technical expert to manager, or a senior executive is transitioning into a mentoring and advisory role, the fundamental rule of transformational leadership is similar. Success means shifting your focus from individual achievement to enabling the capability, growth, and legacy of others.

Building influence and impact

To help with the shift toward transformational leadership, the ILC is featuring sessions focused on questions professionals must ask at key career inflection points.

Rather than a single workshop, the distributed sessions aim to address diverse professional transitions, such as evaluating promotions, learning how to influence laterally, pitching innovative projects, and sustaining leadership energy.

Inflection points include evaluating new internal roles; building lateral or upward trust; pitching an idea about a disruptive project; and facing rapidly expanding responsibilities.

The questions include:

• How do I evaluate career transitions without discarding hard-won experience?
• How can I exercise leadership through credibility and collaboration, regardless of formal authority?
• How do I use an entrepreneurial mindset to create value and gain sponsorship for new ideas?
• How can I avoid the early warning signs of burnout while taking on more responsibility?

The conference is designed to equip attendees with systems thinking and communication mastery needed to cultivate influence.

Leadership is not just about reaching the top; it is also about engaging in a collaborative effort to multiply your impact across the ecosystem.

The history of networking is full of tools that repurposed solutions to very different kinds of problems first. Wi-Fi’s origins trace back, in part, to a team of Australian radio astronomers trying to detect signals from evaporating black holes. But the data-processing tools they’d developed also proved capable at extracting clean messages from any chaotic, echoing signal environment. Echoes are echoes, after all, whether from distant star systems or from the far corner of the house.

I research vehicle communications networks, connecting cars to cars and to transportation infrastructure like traffic lights—for tomorrow’s vehicle-to-everthing (V2X) networks.

V2X research has long relied on models that assume “perfect” or “ideal” network conditions, which is a simplifying assumption that makes the math tractable. But this assumption doesn’t reflect how real wireless signals behave in a moving, obstructed, high-density environment. That gap is exactly the kind of real-world unpredictability that open radio access networks (a.k.a. O-RAN)—an open, programmable architecture behind some 4G and 5G cellular networks—were built to manage.

So why has the O-RAN standard—which is open and available to be applied well beyond 5G telecom—never been used for vehicle communications?

Solutions to the vehicle-to-everything (V2X) problem have to date relied on new networking protocols built from scratch—only to discover chicken-and-egg problems, thorny standards wars, and real signal congestion challenges at scale.

By contrast, O-RAN allows V2X engineers to reuse the networking protocols already developed for cellular communications. O-RAN was developed assuming cellphone towers are generally fixed in place. But, as can be seen below, O-RAN accommodates mobile “towers”—cars and trucks, in this case—with little additional effort.

Imagining a New Way to Connect Vehicles

Self-driving vehicle technology has largely been an each-car-for-itself endeavor. Tesla’s approach, for instance, relies heavily on powerful on-board banks of computers and suites of sensors spread around the car.

However, as an alternative to the “data center on wheels” model, this new O-RAN approach to V2X relies on each car’s nearby neighbors, wherever they are on the road. Each O-RAN–connected vehicle can then use a diversity of cars’ sensors and viewing angles for better group coordination and decision-making.

There is, to be clear, no O-RAN V2X test network operating in the world. Not yet.

It was just 10 years ago that the Third-Generation Partnership Project (3GPP) released its initial cellular V2X standard. The 3GPP have refined V2X over three major releases since. In the U.S. and the EU, the FCC and related European agencies have put forward other standards for short-range wireless V2X communication protocols.

However, no consensus standard has yet emerged. So, lacking any clear, unambiguous guidance on the future of V2X networks, autonomous-car makers—like Waymo, Tesla, Zoox, and Cruise—have leaned more on self-reliance, bulking up each vehicle with as many sensors and GPUs as possible.

Here, though, is where O-RAN might be able to help.

A little like APIs (a.k.a. application program interfaces) connect one app to another on your smartphone, O-RAN serves as an API for the network itself. And because of O-RAN’s open standards, a wireless network becomes programmable, vendor-neutral, and open to custom applications called xApps.

To test our proof-of-concept framework, I have been part of a team simulating five minutes of O-RAN V2X network traffic over one square kilometer of urban area, using real buildings and real-world road layouts from OpenStreetMap and traffic patterns generated by the modeling package SUMO. The simulations assumed a traffic density of 50-70 vehicles per kilometer—not rush hour but not light traffic either. In our simulation, we assumed vehicles communicated via a millimeter-wave frequency of 28 gigahertz and that each component of our O-RAN V2X system had its own dedicated xApp.

Taken together, these inputs—real geometry, real traffic, and each vehicle’s live GPS position—constitute what network researchers call a digital twin of the urban environment. That’s a virtual replica detailed enough for the network to reason about the physical world in real time.

This virtual world gave us a real result, too.

The simulations, published recently in IEEE Network, revealed that existing V2X standards—in which cars uncoordinatedly spit out messages into the network—result in signals “talking” over each other some 80-100 percent of the time. However, using O-RAN signal coordination, the message “collision” rate dropped to near zero.

And that matters because a seized-up V2X network doesn’t just fail quietly. It can fail in ways that might make a road turn treacherous.

How O-RAN Can Coordinate V2X Traffic

High-frequency data links between cars are already difficult to maintain, even on a clear day with no buildings or city infrastructure getting in the way.

Yet, in this situation, existing V2X networks leave a car to conduct blind searches for each dropped signal beam. Traveling at highway speeds, that search takes long enough for the surrounding world to change completely.

An O-RAN network continuously tracks signal conditions across the network, and in O-RAN V2X simulations, we also gave the network access to a detailed map of the urban environment—building positions, road geometry, intersection layouts—combined with each vehicle’s GPS trajectory. Together, these parameters let the network’s control layer predict where and when a signal link is about to fail and instruct each car’s antenna to adjust before the connection drops.

Signal pointing is one failure mode. Losing the connection entirely—because no direct path exists at all—is another.

Consider, for instance, a crossroads of two busy streets, with a few alleys and parking lots adding to the list of potential dangers.

If a signal from car A cannot reach car B directly, or if the path length is too far for an individual beam to travel, the signal must find an intermediary car or stationary sensor nearby that can pass along the message. And existing V2X standards are slow and reactive—polling potential relay vehicles one-by-one: Are you available? Can you redirect this message?

By contrast, O-RAN keeps a running graph of optimized message routes, accounting for a range of real-world constraints. So when an O-RAN link fails (whether that link is direct from sender to receiver—or indirect), the system already has a reroute mapped out.

This is partly why we included “multi-hop routing” in the O-RAN V2X simulations.

Multi-hop V2X O-RAN routing complicated three separate elements of the simulation: for each signal’s middleman (some cars may be ideally positioned to relay a signal from car A to car B, but we made the simulation neglect any cars that were also overwhelmed with their own signals and signal-processing needs); for each signal’s strength (we required that every intermediate link be able to maintain a stable network connection, factoring in distance and traffic conditions); and for each signal’s latency (we required a realistic accounting for added signal latency time for each additional hop in a multi-hop routing).

And with each added complication, O-RAN V2X multi-hop routing continued to extend the network’s capacity from 25 percent of nearby cars connected (without multi-hop) to nearly 100 percent (with multi-hop).

These complications, at least at the simulation level, did not slow down the V2X network.

How Could O-RAN Ever Be Scaled Up for the Real World?

We are in touch with potential collaborators and institutions to develop testbeds, prototype hardware, and tester vehicles for potential proving grounds. The Institute of Science Tokyo, for instance, has already expressed interest in working on some of these early-stage problems.

To date, our published research on O-RAN V2X has centered around a computer simulation only. Real-world hardware will undoubtedly surface challenges our simulation could not. So, questions of network latency and the computational overhead needed for O-RAN V2X signaling remain as yet unresolved.

Plus, concerns about full interoperability and realistic security will each demand their own investigations. After all, no one will trust a V2X network to do anything if that network’s cyber vulnerabilities haven’t been anticipated and patched in advance.

Realizing the O-RAN V2X vision will require progress on multiple fronts simultaneously. On the standards side, O-RAN’s vehicular extensions—the interfaces that allow vehicles to participate in the network as managed elements rather than passive users—would ultimately need to be formally adopted by the O-RAN Alliance and recognized by 3GPP’s V2X specifications. That process takes years.

On the industry side, there is a more immediate problem that our architecture is already positioned to solve: interoperability.

Today, a car made by one manufacturer cannot necessarily parse V2X sensor data sent from a car made by another. Firmware is proprietary; data formats differ. But an O-RAN control layer would act as a universal translator—normalizing each vehicle’s data into a common format and accelerating a push toward true multi-platform vehicle-to-vehicle communications. A more widespread and truly universal standard would, by itself, represent a substantial step forward for V2X.

Today’s U.S. electrical grid, among the largest, most complex systems ever built, is operating at its limit. The combination of rapid industrial growth, more frequent extreme weather, and a record surge in electricity use has pushed the grid to its breaking point, according to the U.S. Department of Energy.

Built decades ago for a more predictable world in which power came mostly from centralized coal or gas plants and electricity use grew at a steady pace, the grid faces unanticipated strain due in part to growing demand from data centers. The jobs of professionals managing the infrastructure have evolved from traditional engineering tasks to complex, fast-moving challenges.

Industry reports show that millions of modern digital sensors, smart meters, and grid monitors are generating nonstop waves of information. The sheer volume of data requires instant, automated computer analysis because human operators cannot process it fast enough.

Pressure on utilities stems from two sources: a spike in electricity demand and a shift in how power is generated.

An example of the operational strain can be seen at the regional level. With the recent deployment of artificial intelligence tools and high-performance computing, data centers require immense amounts of energy to operate. The largest power transmission utility in Texas recently reported a staggering 220 gigawatts of new connection requests, driven largely by a surge in AI and cloud-computing facilities, according to a CNBC report.

Alongside the rise in regional demand, global energy networks are absorbing an unpredictable variety of weather-dependent renewable energy such as wind and solar. The switch creates a volatile operating environment wherein supply and demand are balanced, second by second, to prevent blackouts.

The challenges are compounded by the vulnerability of the grid’s physical and digital framework.

More-frequent severe weather events cause costly disruptions, such as the devastating winter freeze that crippled the Texas grid and record-breaking heat waves that have overloaded transformers.

Simultaneously, the energy networks’ digital architecture faces threats. As utilities replace outdated analog equipment with smart meters and control systems, they are increasingly vulnerable to cyberattacks.

To overcome physical and digital vulnerabilities, grid reliability organizations, such as those conducting North American security simulations like GridEx, emphasize that the grid must become smarter, more agile, and completely automated. Energy researchers are noting that the key to this change lies in integrating AI across every layer of utilities’ operations.

The AI imperative

According to energy industry experts, using AI to manage power systems is no longer a futuristic research project; it has become a baseline operational necessity. Grid analysts emphasize that traditional grid-planning methods are too slow to handle rapid energy dynamics or to balance volatile renewable energy in real time within decentralized power systems such as microgrids.

AI can fill the gap by processing vast amounts of data instantly. Machine learning algorithms can quickly analyze information from thousands of sensors, historical usage patterns, and weather forecasts to predict issues before they happen.

An industrial digitization study conducted by McKinsey & Co. indicated that integrating advanced data and automation across infrastructure networks could reduce system design errors, decrease equipment downtime by up to 50 percent through predictive maintenance, and extend the lifespan of power machinery by up to 40 percent.

From forecasting energy spikes to automatically fixing localized voltage drops, AI acts as the digital backbone of a self-healing grid, experts say. Deploying the complex systems requires a new workforce: power engineers who understand data science, as well as data scientists who understand electricity.

Upgrading the Workforce

To bridge the gap between groundbreaking AI research and practical field deployment, IEEE Educational Activities, in partnership with the IEEE Power & Energy Society, has launched the online Artificial Intelligence for Power and Energy Systems course program.

The program explores core challenges threatening modern utilities. Rather than treating AI as an unverified black box that operates without human supervision, the curriculum focuses on safety, asset preservation, and strict reliability standards.

The curriculum is designed to educate power system engineers, utility managers, and data scientists tasked with modernizing the grid. The program was developed by Fangxing “Fran” Li, professor of electrical engineering and computer science at the University of Tennessee in Knoxville and chair of the IEEE Working Group on Machine Learning for Power Systems.

Five learning modules

The program breaks down the technical transition into five modules that bridge high-level theory with real-world solutions:

AI fundamentals. This module teaches engineers how basic machine learning models apply to power grids. It discusses how specialized neural networks solve complex power-flow calculations and how AI models can safely transition from computer simulations to physical, high-voltage equipment.

Accelerating grid control. Learners are taught to leverage deep reinforcement learning, an AI approach that uses trial and error, to accelerate automated grid adjustments during emergency power events.

Forecasting and data analytics. Using predictive modeling, engineers learn how to predict sudden demand surges, variable wind and solar outputs, and fluctuating wholesale electricity market prices to keep power affordable and available.

Physics-informed and safe AI. To address trust—a barrier to utility AI adoption—this course covers AI models hard-coded to obey the laws of physics. The approach is designed to ensure that automated algorithms never make erratic choices that damage grid equipment.

Generative AI and next-generation tech. Learners can explore the frontier of utility technology, including graph neural networks and large language models. This module highlights how generative AI can process complex, interdisciplinary data to streamline utility planning, emergency responses, and regulatory reporting.

The algorithmic literacy and practical execution tools provided by the course program can help convert systemic risks into grid resilience.

For individual access, visit the IEEE Learning Network. If you are looking for customized organizational options, contact a content specialist to discuss volume pricing.

This report examines R&D waste and how AI adoption has outpaced the intelligence needed to make consequential decisions well.

What Attendees will Learn

1. Where R&D budget is lost. More than a third of organizations spend 25 to 40 percent of their R&D budget on projects that never reach market.
2. Why projects fail late. Almost half of teams estimate over one million dollars in wasted investment for each project killed during development or testing.
3. Why AI adoption has not closed the gap. Most organizations apply AI to execution tasks such as data analysis and modeling rather than to decision support.
4. Where intelligence matters most. Respondents say better access to intelligence has the greatest value at early ideation and feasibility before significant investment is committed.

Most hospitals and health care providers use electronic health records instead of paper charts to note patient vaccinations, diagnoses, and procedures. AthenaOne, Epic, and Oracle Health are some of the systems employed around the world. Many patients can access their electronic medical records from home.

The platforms exist thanks to pioneering efforts such as the Medical Information System (MIS-I). The first hospital-wide computer system, it was developed in the 1960s by Lockheed Martin (then known as Lockheed Missiles and Space Co.) in partnership with El Camino Hospital, in Mountain View, Calif. Doctors and nurses used MIS-I (pronounced miss-ONE) to admit patients, order lab work and imaging, schedule follow-up appointments, and issue hospital bills, according to a 1973 article published by Datamation.

Although many people now know how to type on a keyboard, in the 1960s, most did not. Therefore, MIS-I included a light pen, which worked like a stylus for today’s touchscreens.

MIS-I was recognized as an IEEE Milestone during a ceremony on 14 May at El Camino Hospital. The IEEE Santa Clara Valley Section sponsored the Milestone.

“The medical information system was more than a technological breakthrough; it was proof of what can happen when clinicians, engineers, administrators, and community leaders unite around a common goal: improving care for patients,” Dan Woods said at the event. He is chief executive of El Camino Health, the nonprofit organization that maintains the hospital.

“This pioneering work helped establish the foundation for the modern medical informatics industry,” Woods said. “The legacy of Lockheed’s innovation continues to benefit patients and health care providers around the world, making this achievement truly worthy of lasting recognition.”

Bringing technology into clinical care

Prior to Lockheed’s effort, health records remained paper-based. They often were stored in dedicated rooms within the hospital. Files were kept in heavy-duty manila folders organized on mechanized open-shelf filing systems, revolving rotary files, or locked steel filing cabinets, according to EO Johnson Business Technologies. The process of retrieving a patient’s medical history was cumbersome and could hamper decision-making in a life-or-death situation. Paper records were prone to human error, according to an EHR in Practice article. In addition, upkeep could be costly due to administrative expenses such as transcribing doctors’ notes, storing patient charts, adding medical codes, and managing insurance claims.

Companies and universities including General Electric, IBM, and Harvard began exploring how to use computers to improve clinical care. They developed several systems for hospital laboratories to track test orders and results, as explained in the Milestone webpage.

In 1964 Lockheed was looking to diversify its portfolio, Melville Hodge, who helped lead the MIS development, said at the dedication ceremony. The company decided to apply its expertise to health care, and later that year this focus-area became part of a new information systems division. Hodge, who at the time oversaw multiple R&D efforts, became the driving force behind the MIS program.

MIS-I displayed patient information on a 14-inch television purchased from a department store. Below that monitor was a keyboard and a light pen.Ian Thomson/Computer History Museum

In 1966 Lockheed secured a contract with the Mayo Clinic, in Rochester, Minn., to assess its computer system needs and those of its two associated hospitals, Hodge wrote in a paper detailing MIS history.

He and a small team of engineers worked with Mayo Clinic physicians for two years to build the prototype of what would become MIS-I.

The system displayed patient information on a 14-inch television purchased from a department store. Below that monitor was a keyboard, and to its right was a printer. Doctors and nurses would swipe their ID badge to access the system, then use the keyboard to put information into the patient’s file or send a request to a pharmacy or laboratory. They also could print documents.

But one problem kept cropping up: Most doctors didn’t know how to type. The computer mouse was still in its infancy, and Hodge suspected it would not solve the problem, according to a video shown at the dedication ceremony. Instead, he “borrowed technology from a then-secret satellite program,” he said.

That technology was the light pen, which was used with MIT’s Whirlwind Computer in the 1950s.

“The insight that physicians could not and would not learn to type, combined with the innovative solution of light pen interaction, transformed an impossible dream into practical reality,” the Milestone proposers wrote.

To display text, the system used matrix programming, a 2D data structure consisting of rows and columns. Using the light pen, a doctor or nurse would select text from a list of general categories on the monitor. The options included the patient’s personal and medical information, family medical history, current illness, and physical exam findings, according to a 1968 article in the medical journal JAMA. The computer would display the requested information or list the next steps to complete tasks such as sending a prescription to a pharmacy. The keyboard remained part of the setup because it could allow users to input new information and update patient records.

To further develop the system, they submitted a proposal to the U.S. Department of Health, Education, and Welfare (now split into the Departments of Health and Human Services and Education) to secure additional funding, but it was rejected.

Herschel Brown, Lockheed’s executive vice president, and Kenneth Larkin, its director of information systems, decided to fund its commercial development, Hodge wrote.

When the company’s contract with the Mayo Clinic ended, the Lockheed team returned to Sunnyvale, California to refine, test, and deploy the system at El Camino Hospital.

“I admire Ed Hawkins, who was its first administrator, for having the courage to take on this kind of project while running a hospital that was only four years old at the time,” Hodge said at the dedication ceremony.

Making MIS-I a commercial success

Starting in 1968, early prototypes were installed in the hospital’s M.D. lounges and nursing station at El Camino Hospital. The organizations worked to configure the system so it met the hospital’s needs.

By 1969, a number of monitors had been installed, including in admissions, pharmacy, and radiology. The information from all the connected machines was stored in a data center housed in a separate location outside the hospital.

In 1971 Lockheed encountered difficulties with its C-5A and L-1011 aircraft programs, according to Hodge. The company was forced to curtail discretionary new business programs including MIS-I. The program was sold to Technicon of Tarrytown, N.Y., a leader in clinical laboratory automation.

The medical information system business operated independently as a subsidiary unit, and the transition marked the beginning of MIS-I’s commercial expansion.

That same year, MIS-I went live for hospital-wide use. Physicians’ orders were communicated to other departments, test results and radiology reports were retrieved, and nursing care planning and documentation were available, according to the Journal of Nursing Scholarship. MIS-I supported most information handling for nurses, physicians, and other medical personnel in the hospital.

But the change was not welcomed by all, according to the video about the technology. Nurses tended to praise the system, but many doctors had a hard time transitioning from paper to computers. They complained they were “spending more time fighting a machine” than interacting with their patients, according to the video. Some even retired to avoid learning the system. But nurses fought to keep it, emphasizing to doctors how much it improved patient care.

In 1974 El Camino Hospital held a vote of medical staff to determine whether to keep the system or return to paper-based records. About 60 percent of doctors and more than 90 percent of nurses voted in favor of keeping it, according to the Milestone webpage.

“The medical information system was more than a technological breakthrough; it was proof of what can happen when clinicians, engineers, administrators, and community leaders unite around a common goal: improving care for patients.” —Dan Woods, El Camino Health CEO

In 1975 nonprofit Battelle of Columbus, Ohio, evaluated how well the system was working for El Camino Hospital. It found that MIS-I reduced the time nursing staff spent on clerical tasks, improved communications among departments, and facilitated better planning of patient care. The survey also showed that more readily available, complete, and accurate information was being used to administer care and monitor patient progress, according to the report.

By 1993, the technology was installed in more than 200 hospitals in the United States, Canada, and Europe, according to the Milestone entry.

El Camino Hospital used MIS-I for 34 years, until its decommissioning in 2005. It was initially replaced by Eclipsys Sunrise XA and then ultimately by Epic.

Honoring an IEEE Milestone

The dedication ceremony brought together IEEE leaders, hospital staff, and government representatives. Hodge and his family also attended. IEEE President-Elect Jill Gostin made a presentation about IEEE, and Brian Berg of the IEEE History Committee discussed the organization’s Milestone program.

Hodge participated in a Q&A session with Deb Muro, chief information officer of El Camino Health. He told a story about the early days of MIS-I that he said he will never forget. During a hospital board meeting at which physicians were complaining about the system, an announcement was made over the hospital’s public address system that MIS-I wasn’t working.

“I had to ignore it to survive,” Hodge said, laughing. “As physicians got more used to it, and with a phenomenal poking from the nurses, doctors who wouldn’t use it were forced to.”

A bronze plaque recognizing the MIS-I as an IEEE Milestone has been installed in the lobby of the hospital in Mountain View. The plaque reads:

From 1965 to 1974, the first hospital-wide computerized medical information system was created by Lockheed Missiles and Space Co. in partnership with El Camino Hospital. Innovative light-pen terminals enabled physicians and staff across all departments to efficiently and accurately access patient data and enter work orders. By providing immediate feedback and seamless communication, it reduced costs and errors, improved safety and outcomes, and led the way to modern medical and clinical informatics.

Reviewed by the IEEE History Committee and approved by the IEEE Board of Directors, IEEE Milestones recognize outstanding technical developments around the world that are at least 25 years old. The Milestone program is administered by the IEEE History and Heritage group.

To learn more about historical figures in engineering, IEEE Milestones, and IEEE History Center programs and events, check out The Institute’s IEEE Tech History collection. IEEE Spectrum also covers aspects of tech history.

This article is brought to you by COMSOL.

In pursuit of improved range, greater reliability, and faster charging, electric vehicles are driving the demand for high-voltage electronics. Other applications driving this demand include wind farms, data centers, and server farms, to name a few. As the interest for high-voltage electronics increases, the risks associated with their sudden failure must be considered.

Much of what causes high-voltage equipment to malfunction can be linked to the conditions of the climate in which it operates. Specifically, condensation on electronic surfaces caused by humidity can lead to corrosion, which can result in stray leak current and dendrite shorting during Electrochemical Migration (ECM).

Predicting, mitigating, and helping proactively design to account for corrosion is the focus of the Centre for Electronic Corrosion (CELCORR) research group at the Technical University of Denmark (DTU). The group’s researchers are working with industry partners to develop models and simulation apps that will help in building robust electronics designs. Their goal is to develop knowledge that can be used for manufacturing electronics to withstand humid operating conditions.

Electronic Failure: “It’s Not the Heat; It is the Humidity”

Automotive electrification and renewable energy systems rely on electronics at all stages of the energy chain. When these electronics, such as the example PCB in Figure 1, are exposed to the effects of moisture, they can become potential failure points.

“Anywhere you are producing, converting, transporting, and using energy, you need these high-power electronic systems that get affected by the humidity,” explained Dr. Rajan Ambat, DTU professor and manager of CELCORR. Ambient moisture can seep into the devices and machines that require these electronics and cause unexpected functional issues through corrosion. When these electronics are involved in particularly high-voltage applications (such as wind farms, data centers, electric vehicles, and server farms), failure due to humidity exposure can even lead to fire.

“There might be a situation where somebody installs a solar panel near the seashore or in an area with high humidity, and within a short period of time, a conducive condition forms inside the electronics that results in a failure,” Ambat said. “This is why we need to understand exactly how the conducive condition of condensation is created, when the condition is created, and how the system is failing.”

Figure 1. The electrolyte potential and current density distributions on a PCB surface (with pinholes).CELCORR/DTU

Identifying corrosion as the underlying cause of some electronic failures is still a challenge. “Fifty percent of failures in electronics are currently branded with an unidentified root cause,” Ambat explained. “When engineers open up the system, they do not realize that the failure was due to corrosion, because moisture disappears without leaving any sign of corrosion unless there is ECM dendrite formation.”

This lack of awareness was a strong motivator for CELCORR, which turned to multiphysics simulation as a supplementary tool to help its partner organizations to better predict humidity-related issues at the design stage.

Modeling Moisture and Measuring Parameter Changes in PCBs

CELCORR believes the best way to identify and avoid electronic failures is to build more effective designs that better prevent corrosion from developing or can withstand a higher humidity load. Its research team applies its expertise in modeling to generate simulation apps that will help partner industries to evaluate safe designs for humidity robustness.

“We are at the intersection of materials science and the electronics industry. We work as a bridge between materials and electronics disciplines, using both kinds of language,” Dr. Anish Rao Lakkaraju, a postdoctoral researcher at CELCORR, explained.

To understand potential design issues, Ambat emphasized the importance of virtually breaking down systems to identify where problems may arise. “We need simulation software to analyze potential uses of designs and whether new designs are good or bad,” Ambat said.

Researchers at the Technical University of Denmark (DTU) are using simulation apps to predict corrosion and design electronics proactively to mitigate or withstand its effects.

Using the COMSOL Multiphysics software for investigation, the CELCORR research team together with other research partners (Aalborg University) built an example model with a simple PCB geometry that matched both its test circuit boards as well as the design of a device used by one of its partner companies. The team then added a water film layer on top to act as the relative humidity. From there, the team could introduce variation. “We change the layout, geometry, distance between electrodes, thickness of the water film, and conductivity of the water film depending on the conditions,” Ambat said.

Ambat and his team generate data on the effect of these variations and can identify which has the greatest impact on the device’s performance and whether any alterations can improve the device’s anticorrosion robustness. “We assume there is condensation forming on the electronic surfaces (Figure 2) and compute the electrochemical leak current for different design elements,” Ambat said. “The value of the computed electrochemical current between the parts will give us an indication of whether the PCB will be affected or not.”

Figure 2. A 10-µm water film condensation effect on an example PCB.CELCORR/DTU

Altering the design elements and solving the model equations again and again allows the team to better understand what makes a design effective. “Now, we are at the current form of the model, and we are quite happy with where the physics are at this point,” Lakkaraju said. This modeling, however, was just the first part of CELCORR’s overarching goal of illuminating the destructive potential of corrosion in electronics and the best ways to avoid it.

Using Simulation Apps to Test Real-World Designs

To open up and ease access to these models, CELCORR used the Application Builder in COMSOL Multiphysics to create simulation applications for the members of the industrial consortium. Built with a simple 3D circuit board geometry with two oppositely biased electrodes and a water layer to replicate corrosion-causing moisture, the apps provide a pared-back, straightforward interface where users can vary model inputs. “We focused on fundamental aspects,” Lakkaraju said. “We boiled down our partners’ overarching concerns to create two apps with very simple geometries and the simplistic stuff that can be varied over multiple parameters.”

The simple simulation apps shown in Figures 3 and 4 are designed to show companies how the distance between the electrodes and the thickness of the moisture layer affect the leak current through the water film depending on different parameters. By analyzing multiple design elements and parameters, users can determine the relative benefits of certain design elements on the humidity robustness. “The apps we have built have really helped because they give the electronics engineers a plug-and-play sort of approach,” Lakkaraju explained.

Figure 3. The UI of CELCORR’s standalone app showing the inputs that users can alter.CELCORR/DTU

“By using this app [Figure 3], companies have unlimited freedom to work with these sorts of variables,” Lakkaraju added. “This would be quite difficult to recreate in real life with physical design and testing, and the companies we work with are quite happy with the level of accuracy the app can provide.”

Figure 4. The UI of one of the simulation apps showing a streamline plot with inputs such as the cathode voltage and blockage length.CELCORR/DTU

Looking Forward: Complex, High-Voltage Modeling

Alongside its collaboration with the industry consortium, CELCORR is also working toward improving the world’s general understanding of corrosion’s impact on electronics. To do this, Ambat and his team are undertaking multiple projects, including actively adding greater complexity to their models. In a tertiary current distribution model they are building, the team is drilling down into each of the basic inputs used in a secondary current distribution model, zooming in to examine the effects of the set of even more detailed inputs each basic input comprises.

“The next point is to study what each of these detailed inputs does,” Lakkaraju said. In particular, the team is examining mass transport properties and the chemical reactions’ rate constants.

In addition to these ongoing studies, CELCORR is expanding the scope of its research to investigate corrosion in high-power, high-voltage systems. Thanks to a 2024 grant from the Grundfos Foundation, CELCORR was able to establish the Centre for Climate Robust Electronics Design (CRED). The center’s lab facilities and expertise are being developed to address the humidity-robustness requirements of today’s high-voltage and high-power electronic equipment. “Using CELCORR’s uniquely deep understanding of materials and corrosion, we are equipped to find the root cause and provide knowledge for environmentally robust designs,” said Ambat.

For all of its investigation, CELCORR continues to rely on the agility of the COMSOL Multiphysics software. As Lakkaraju explained, “It is really quite nice how a model can be adapted to a variety of combinations of materials, geometries, and parameters and how the software allows you to keep building on from there.”

A German engineer wanted a cheaper cigarette. The popular crooner Bing Crosby wanted a vacation. Satisfying both desires inadvertently led to the invention of the laugh track. Along the way there were Nazis, spoils of war, and more than one accidental encounter. Tying together this quirky history is the Magnetophon.

What Was the Magnetophon?

The Magnetophon was a high-fidelity reel-to-reel magnetic tape recorder. The hit of the Berlin Radio Show when it debuted in 1935, it was developed by the German electronics manufacturer AEG. The magnetic tape was produced by I.G. Farben (now known as BASF).

German inventor Fritz Pfleumer came up with the idea of recording sound on magnetized paper tape coated with metal.ullstein bild/Getty Images

The tale of that tape runs through German inventor Fritz Pfleumer. In the early 1920s, Pfleumer was working on industrial paper products in Dresden. At the time, fancy cigarettes had gold leaf to decorate the tip. Cheaper manufacturers achieved a similar effect using colored paper, but the dye could stain smokers’ lips, and no one wanted that. Pfleumer devised a less expensive process using powdered bronze to simulate the gold band.

Pfleumer didn’t work in the recording industry, but he was familiar with the technology of electromechanical recording, which was done on metal wire—a 1898 invention of the Danish engineer Valdemar Poulsen. Pfleumer thought he could do something similar with paper by replacing his powdered bronze with a magnetized material. He patented his “sounding paper” in 1928 and also invented an audio tape recorder to go with it. The sound quality wasn’t great, and the paper tore easily, but it was the start of a promising idea. Two points in its favor: The paper could be sliced, allowing edits, and it could be erased and re-recorded over.

Pfleumer knew he needed to partner with a larger company to commercialize his idea, and so he signed a contract with AEG in 1932. Hermann Bücher, chairman of the AEG board of directors, took a personal interest in the project, helping shepherd it to completion. Although AEG originally planned on developing both a recorder and the tape, it didn’t take long for Bücher to reach out to his friend Wilhelm Gaus, managing director at I.G. Farben. AEG developed the hardware, while I.G. Farben worked on the tape.

The two teams dubbed their product the Magnetophon, or magnetic phonograph, and planned on launching it at the 1934 Berlin Radio Show to compete directly with machines that recorded on steel tape or wire. But the product wasn’t quite ready. Two days before the show, they canceled the debut.

A year later, the bugs had been worked out. The Magnetophon’s introduction at the 1935 show was a resounding success. AEG fielded inquiries for many variations on the recorder, including one that combined the recorder with a telephone, a player for prerecorded music, and a special version to add artificial reverberation for recording open-air concerts. Over the next three years, AEG developed several iterations, resulting in the Magnetophon K4 in 1938, its first commercially successful tape recorder.

The K4 eliminated the hiss and distortion that magnetic recordings previously suffered from by incorporating AC bias, which added a high-frequency signal, typically around 40 to 150 kilohertz, to the recording. Inaudible to the human ear, the signal reduced distortion, especially when recording quieter passages. The fidelity of the Magnetophon was such that radio listeners couldn’t distinguish between live broadcasts and prerecorded performances.

The Magnetophon During and After the War

This is where the Nazis come in. Adolf Hitler and his propaganda minister, Joseph Goebbels, understood the power of radio. The Magnetophon became a powerful tool for both political messaging and military strategy. Because the device could record and play back sound at the same quality as a live radio broadcast, it allowed Hitler’s recorded speeches to be aired from one radio station while the dictator was in another part of the country. This made it more difficult for the Allies to pinpoint his location.

RELATED: Inside the Third Reich’s Radio

Meanwhile, World War II disrupted the exchange of technical information and prevented Americans from learning about the Magnetophon until after the war. At least, that’s the shorthand version of this history I kept running across during my preliminary research for this column.

During World War II, U.S. Army Signal Corps engineer Jack Mullin [top] came across the Magnetophon. After the war, he got approval [bottom] to ship two of the disassembled machines and reels of magnetic tape back to the U.S.Top: Pavek Museum; Bottom: Richard L. Hess/The Mullin Family Collection/Archive of Recorded Sound/Stanford University

But then I read Friedrich K. Engel’s account in the book Magnetic Recording: The First 100 Years, which provides a wealth of detail about the Magnetophon. Among other things, Engel notes that the AEG affiliate in Schenectady, N.Y., received a Magnetophon in November 1937, well before the United States entered the war, and AEG engineers demonstrated it for their colleagues at nearby General Electric. The GE engineers dismissed the technology out of hand, though, and so Americans had to wait until after the war for the Magnetophon to be reintroduced.

We have electrical engineer John T. “Jack” Mullin to thank for that reintroduction. Mullin served in the U.S. Army Signal Corps, stationed in the United Kingdom and Paris during the war, and he liked listening to the radio. He realized that “live” orchestral broadcasts coming from Germany in the middle of the night, when no musicians would have actually been in the studio, lacked the telltale hiss and crackle of prerecorded music. Clearly, German engineers had recording technology far superior to the Americans’.

Sent to Germany at the war’s end, Mullin eventually came across that technology at a radio station, and in 1945 when he returned home to California, he shipped two disassembled Magnetophons, a case of tape, schematic drawings, and the determination to change the U.S. recording industry.

Bing Crosby Championed the Magnetophon

Enter Bing Crosby. In the 1940s, Crosby was perhaps the most popular performer on the radio. But he was tired of performing two weekly shows three hours apart (one for each coast). He wanted to prerecord his performances and take a break. He sent in his lawyers to talk to executives at NBC, which aired his program. The audio recording technology at the time used vinyl or shellac transcription discs, but radio listeners could hear the pops and hisses and knew it wasn’t live. NBC refused Crosby’s request, and so Crosby took a year off from radio and then signed with the upstart network ABC, which was willing to let him record his shows for later airing.

Radio producer Murdo MacKenzie [right] arranged for Jack Mullin [left] to demonstrate the Magnetophon to Bing Crosby in 1946.Pavek Museum

Serendipitously, Mullin had begun demoing the Magnetophon in California. In October 1946, Crosby’s technical producer, Murdo MacKenzie, heard about the demonstrations and arranged one for Crosby at the Metro-Goldwyn-Mayer studios in Hollywood. Crosby was delighted and promptly invested US $50,000 (about $800,000 today) in Ampex, the company working with Mullin to engineer an American version of the Magnetophon. Mullin became Crosby’s chief engineer. 3M developed the magnetic tape.

In 1947, Crosby became the first major radio star in the United States to prerecord performances. Much of the success was due to the recording equipment, but Mullin was also an excellent editor. Crosby and his team would record multiple takes, and Mullin would deftly splice together the best to create a seamless performance.

After seeing a demo of the Magnetophon, Bing Crosby invested $50,000 in Ampex, which developed a U.S. version of the tape recorder. Cinematic/Alamy

One day a guest told a joke that was uproariously funny, but a little too spicy for radio. Even though the joke couldn’t be aired, the audience’s laughter was worth keeping. Soon, the producers created a whole catalog of different laugh tracks. If a joke didn’t land, it didn’t matter. The editor could just add a laugh in postproduction, from polite titters to hearty guffaws.

According to Crosby’s daughter Mary, Bing didn’t have a problem with this type of editing. The live audience was immaterial as long as the jokes were funny. But according to Mullin’s daughter, Eve Mullin Collier, the manipulation didn’t sit well with her father. He disliked the inauthenticity of the moment, of editing joy.

The history of technology is filled with such episodes of unintended consequences. Mullin admired the Magnetophon precisely because it could faithfully record a performance, an event, a moment in time. He worked tirelessly to refine the machine for the benefit of radio audiences everywhere. And so I understand why its appropriation for capturing canned reactions and manipulating reality must have rankled. He championed the technology, but ultimately it moved beyond his control.

Part of a continuing series looking at historical artifacts that embrace the boundless potential of technology.

An abridged version of this article appears in the August 2026 print issue as “Birth of the Laugh Track.”

References

Luis Felipe Eguiarte Souza, curator at the Pavek Museum of Electronic Communication, in St. Louis Park, Minn., first told me about the Magnetophon and its link to the laugh track. The machine pictured at top is on display at the Pavek, and is one of the two that Jack Mullin brought back from Germany and then rebuilt.

For significantly more technical detail on the development of the Magnetophon, magnetic tape, and its reinvention in America, check out Chapter 5, “The Introduction of the Magnetophon,” by Friedrich K. Engel, and Chapter 6, “Building on the Magnetophon,” by Beverley R. Gooch, in Magnetic Recording: The First 100 Years (IEEE Press, 1998).

Radiolab interviewed Mary Crosby and Eve Mullin Collier as part of the show “Mixtape: Jack and Bing,” which includes many archival recordings as it tells the story of the development of the laugh track.

There is a 2006 documentary on Jack Mullin titled Sound Man: WWII to MP3, but I was unable to view it.

When I started at Spectrum 25 years ago, a senior editor suggested that I find a “rabbi,” by which he meant someone who could mentor me in how EEs approach problems and evaluate potential solutions.

I didn’t find one right away. Then in 2005 we decided to do a special report, focusing on the challenges of enterprise software development. I suggested we invite IEEE Life Senior Member Robert N. Charette, a self-described risk ecologist, prolific book author, and leading authority on risk management and software engineering, to explore in our pages the myriad reasons software projects fail. His seminal article “Why Software Fails” is still read in university engineering classes today.

IEEE Life Senior Member Robert N. Charette is one of IEEE Spectrum’s most prolific authors.Robert N. Charette

It was, as they say, the beginning of a beautiful friendship. I had found my rabbi, one who shared my love of writing. We settled into a rhythm that would last more than 20 years, talking on Friday mornings about a range of topics including the growing ubiquity of software in our lives.

So when I became Spectrum’s website editor in 2007, he was the first contributor I tapped to start a regular blog (remember those?). The Risk Factor was born and over the course of more than 10 years and 1,750 posts, Bob chronicled hundreds of software debacles, culminating in “Lessons From a Decade of IT Failures,” which won a Jesse H. Neal Award for Best Infographics in 2016. Ironically, yet predictably, those infographics were created in a software package that is no longer supported and thus are lost to the bits of time.

“I like the expression on the fish just before it’s going to be swallowed by the heron.”Robert N. Charette

Bob, however, was not a one-trick pony. In between his full-time job running his two management consultancies and raising a future biochemist and a future civil engineer, his daughters Maura and Megan, he also wrote many deeply reported and insightful articles. These include last year’s “The Doctor Will See Your Electronic Health Record Now,” the eye-opening 12-part series and e-book The EV Transition Explained, and my personal favorite “Automated to Death,” about the deadly consequences of the automation paradox as manifested by the cyberphysical systems that pilot planes, trains, and automobiles.

“The young bald eagle I photographed in September 2024 had bands that I could read which identified it as a female born in May 2024, near Lexington Park, St. Mary’s County, Maryland, about 65 miles away from where I live.”Robert N. Charette

His main goal all along has been to make software visible, as he told me one Friday in July. “Software is all around us, but we don’t recognize it at all,” he said. “I really wanted my stories to help people better understand complex software systems. You can’t see software, you can’t touch it, you can’t taste it. You may feel the consequences of software failure, but you never see the reason itself.”

When he told me that he was hanging up his hat as a contributing editor to focus on nature photography and to write a handful of fictional trilogies, including one entitled “The STEM Murders” featuring an engineer-turned-detective and his rabbi, I asked him which of his Spectrum articles had the biggest impact.

“The hummingbird I caught with the yellow of a road curb behind it.”Robert N. Charette

He singled out the 2013 feature “The STEM Crisis Is a Myth.” “Spectrum gave me a platform to question the assumption that we needed more STEM graduates. Until then, people didn’t really realize how much of the STEM crisis was a mythology that was perpetuated by employers and the academic community and was foisted on the IEEE community,” he said.

Charette made a career of questioning assumptions. The best way to mitigate risk, he told me as our Friday chat drew to a close, is to be careful making assumptions in the first place. “My main risk maxim is assumptions made are risks accepted.”

Given a rising number of publishing misconduct allegations, IEEE in 2022 created the Publishing Ethics Team as a centralized department to assist in handling claims. The group also works to increase the organization’s visibility in the broader publishing ethics area and helps IEEE volunteers write new policies.

Here are some highlights of the team’s activities last year.

New detection tools

IEEE conducted a pilot program in 2024 to integrate tools from the International Association of Scientific, Technical, and Medical Publishers (STM) Integrity Hub into the peer-review workflow of IEEE Access. The multidisciplinary, fully gold-open-access journal publishes research results across all IEEE fields of interest.

STM created the hub so scholarly publishers could access a suite of integrated, commercial, third-party research integrity tools as well as those developed by STM Solutions. The tools help the publishing group avoid printing problematic content upon manuscript receipt, rather than reacting postpublication.

The new features include the Clear Skies Papermill Alarm, which helps identify potentially fraudulent manuscripts at submission. Another is an integration with the PubPeer database, which allows users to check whether references in a manuscript have received previous PubPeer comments or have been retracted—both of which can indicate quality or integrity issues. The duplicate submissions detector can determine whether the same manuscript has been submitted to multiple journals by different publishers, often a sign of academic “paper mill” activity.

Following the success of the pilot, IEEE began working last year to expand the services to all its periodicals. It is anticipated that all IEEE periodicals will be included in the Integrity Hub screening by the end of this year.

Raising visibility

The team participated in industry-wide initiatives with STM. It also renewed membership in groups including the Committee on Publication Ethics, and the team continued its work sponsoring and presenting at conferences.

At a panel presentation during the Council of Science Editors annual meeting last year, Amanda Sulicz, manager of IEEE Research Integrity, participated in the Research Integrity Investigation panel session. She also presented at the Standardization of Publishing Integrity Norms and Corrective Actions poster session during the Society for Scholarly Publishing’s annual meeting, held 28 to 30 May 2025.

Luigi Longobardi, the IEEE Publishing Ethics and Conduct director, gave a presentation at the Communication and Collaboration With Institutions session during STM Innovation and Integrity Days, which took place 9 and 10 December.

IEEE was a sponsor of the National Conference on Research Integrity, held 20 to 22 May 2025, and the International Congress on Peer Review and Scientific Publication, held 3 to 5 September.

Ethics reports

The team is responsible for tracking ethics-related complaints for all IEEE publications, including articles published in periodicals and conference proceedings. When complaints regarding an article’s integrity are received, either via email at pub-ethics@ieee.org or the anonymous ethics reporting line, the team works with IEEE volunteers to open a case, investigate the complaint, and resolve the matter.

Last year 591 cases were opened, a 56 percent increase over 2024. Of the 591 reports, 317 were closed and 274 are still under investigation. Of the complaints, 88 percent were research-related, including issues with plagiarism, AI-generated text, and falsification of—or unauthorized use of—data. The other 12 percent involved alleged misconduct by editors, reviewers, and conference organizers.

The complexity of the reported cases has expanded. An increasing number of the complaints deal with more than one article or complicated situations such as editorial misconduct or peer-review manipulation.

Conference publications

For the second consecutive year, the team participated in the joint IEEE Publication Services and Products Board/IEEE Conferences Committee’s ad hoc committee on conference publication quality. The committee is tasked with analyzing and reviewing problematic conference papers and enhancing quality screening of articles prior to publication to detect integrity issues such as plagiarism and tortured phrases. The committee also updates educational modules on organizing and managing conferences.

As part of the review process, the committee focused on conference articles that contained tortured phrases, which are nonstandard English expressions that are imprecise or erroneous and give the impression of technical jargon.

Many of the articles reviewed by the ad hoc committee were identified by the Problematic Paper Screener, a free online tool that uses application programming interfaces to screen papers published online for potentially problematic content, such as tortured phrases, machine-generated content (SCIgen or Mathgen, for example), or references to retracted content.

The ad hoc committee was responsible for reviewing and recommending the retraction of more than 1,700 problematic conference articles last year.

Case studies

From the cases the team reviewed, IEEE learned valuable information to help update its publishing policies and best practices.

Here are examples of two anonymized cases reported to the team last year.

Case Study 1

Updated Policies

Occasionally, a misconduct case is so complicated that it requires an update to the PSPB Operations Manual.

In this particular case, Coauthor 1 reported to the Publishing Ethics Team that the work was reused in an IEEE publication without proper credit. The new work also listed two coauthors not part of the original document. Coauthor 1 also reported the case to their university’s research integrity officer (RIO) for additional investigation.

Initially, adjudicating the case proved challenging because under the PSPB policies at the time, the issue would have been classified as a multiple publication, which typically results only in a warning for the authors of the new work.

With the assistance of the RIO, it was determined that the methodological and theoretical portions of the paper were previously derived in the university’s lab; therefore, the contributions of the two new authors were not substantial enough to warrant authorship.

After deliberations by the IEEE Publishing Conduct Committee and eventually the IEEE Document Working Group, which is responsible for updates to the Operations Manual, IEEE PSPB Policy 8.2.4 was updated to clarify policies regarding the adjudication process for reuse of material and the proper crediting of coauthors when material has been reused.

Case Study 2

Faked Reviewers

Using the screening tools and data available to IEEE periodical editors, an editor in chief was alerted to suspicious reviewer activity. Specifically, two of the reviewers assigned to an article submitted the exact same review text. The editor contacted the handling editor to inform them of the irregularities and also contacted the two reviewers, asking them to verify that they were, in fact, the ones who submitted the reviewer report. Out of an abundance of caution, a new editor was assigned to the article, and new reviewers were selected while the investigation continued.

The investigation concluded that the original handling editor created and submitted both reviews in question. Following the recommendations provided to the PSPB vice president by the periodical’s Society and Publishing Conduct Committee, the handling editor was banned from publishing with IEEE and serving in an editorial capacity.

If you’re wondering what dark matter is, you’re not alone. Astronomers don’t know. But they’ve determined that this invisible material must be far more abundant than the stars and nebulas that they can see. They’ve surmised as much from observing the gravitational effects of all this perplexing dark stuff, launching a decades-long campaign to understand its nature.

I recently learned that it is possible to sense the presence of dark matter using a small radio telescope such as the Discovery Dish covered in these pages last year. The trick is to know what observations to collect and how to analyze them. I’ll sketch that out below, but first let me describe the homemade radio telescope I put together for this project.

It’s a pyramidal-horn antenna, not unlike the horn antenna first used in 1951 to detect the 1,420.4-megahertz radio emissions from interstellar clouds of neutral hydrogen in space. These emissions hold the key to confirming the presence of dark matter because such clouds can be found all over the galaxy, and their motions reflect what’s happening in different parts of the Milky Way.

I used an online calculator to help me design my antenna, adopting dimensions I knew I could achieve using a US $25 10-by-2-foot roll of roof flashing and an emptied one-gallon paint-thinner can. (Next time, I’ll just buy an empty F-style can.)

The horn antenna is made from tape, an empty paint-thinner can, and a roll of metal roof flashing [bottom row]. Signals are picked up with a low-noise amplifier [top middle] and passed to a software-defined radio receiver [top left].James Provost

Construction of the antenna itself was similar to that of the slightly smaller horn antenna I described in these pages in 2019. I made my new antenna bigger, though, because I needed better angular resolution, allowing me to scan smaller regions of the sky.

To pick up the emissions from interstellar hydrogen, I used Nooelec’s $45 SAWBird+ H1, a device that combines two low-noise amplifiers with a standing-acoustic-wave filter centered on 1,420 MHz, in combination with a RTL-SDR V4 dongle. So it’s not too hard to put together the hardware needed to measure signals from hydrogen clouds. But how do you pull the signature of dark matter out of those signals?

The answer is that you use such measurements to gauge the speed at which clouds located at different distances from the center of the Milky Way are moving in their orbits.

You just have to show that the speed at which material orbits the center of the galaxy doesn’t fall off with distance.

You might think that these clouds circle around the galactic center in the same way that planets orbit the sun or satellites orbit the Earth, with objects close in orbiting faster than those farther out. Mercury, for example, zips around the sun at 47.4 kilometers per second, whereas Neptune lumbers along at a leisurely 5.4 km/s.

The Milky Way contains a central bulge of stars surrounding a supermassive black hole. So at first blush, the mass of the galaxy appears to be concentrated near its center. If that were the case, stars and clouds of other material would orbit more slowly as their distance from the center increases. If, however, there were enormous amounts of invisible matter present throughout the galaxy, you wouldn’t expect orbital velocities to diminish in this way.

How Do You Measure the Speed of Interstellar Clouds?

So to detect dark matter, you just have to show that the speed at which material orbits the center of the galaxy doesn’t fall off with distance. And radio observations are the easiest way to do that, because you can gauge speeds by measuring how much the signal from hydrogen clouds is shifted by the Doppler effect.

By pointing your radio telescope at different parts of the sky, you pick up emissions from clouds located at various distances from the galactic center. The frequency offset of these emissions from 1,420 MHz reflects the speed of approach or recession of those clouds relative to Earth.

You need measurements from the plane of the galaxy, at galactic longitudes between 0 and 90 degrees (a galactic longitude of 0 degrees points directly toward the center of the galaxy and 180 degrees directly away from it). Applying some high school trigonometry lets you convert these figures into orbital speeds around the galactic center, a technique known as the tangent-point method.

In any group of clouds, the one with the highest velocity as seen from Earth will be the one lying closest to a tangent point along its orbit around the galaxy. This allows its distance from the galactic center to be determined through trigonometry [top]. The bottom graph shows the astronomical community’s measurements for velocities around the galactic center [in black], with the author’s results plotted in filled and open red circles.James Provost

Experiments aiming my horn antenna at an Inmarsat geostationary satellite revealed that the angular resolution of my little radio telescope is about 20 degrees. So with the help of the planetarium program Stellarium, I pointed my antenna in the plane of the galaxy at galactic longitudes of about 15, 30, 45, 60, 75, and 90 degrees, spacing things out in an effort to make each set of measurements largely independent.

I used the SDR# software with a plug-in called IF Average to read the raw measurements coming in from the antenna. This plug-in stacks up data received over a few minutes, allowing a weak signal to build up and produce a clean radio spectrum that shows the 1,420-MHz line. In reality, it looks more like a bump, or even a set of bumps due to Doppler shifted emissions from multiple clouds, located at different distances from the galactic center. Fortunately, you only have to care about the cloud that’s receding the fastest—the one with the largest redshift, in astronomer-speak.

I used Microsoft Excel to analyze the shapes of radio spectra I gathered, modeling them as the sums of individual bell-shaped contributions from different clouds. That allowed me to estimate the largest redshift for each galactic longitude I probed. Then, again using Excel, I applied formulas that transformed those six redshift values into six pairs of orbital velocities and distances from the galactic center.

The plot of my results matched reasonably well with a recent paper, “The Inner Rotation Curve of the Milky Way,” in Publications of the Astronomical Society of Japan. Two innermost points did show anomalously low orbital velocities. Another shot at curve fitting in Excel brought these results closer to expectations, but they were still somewhat off.

In any case, the orbital velocities I estimated did not diminish with distance from the galactic center—quite the opposite. Something out there is putting its stamp on how the Milky Way turns. And that basic observation is what allows me to say that, with the help of some roof flashing and a paint-thinner can, I’ve been able to detect dark matter from my backyard.

In February 2024, a young man lay somewhere on the frozen shore of James Bay, Canada, surrounded by snow and darkness, succumbing to hypothermia. When he failed to get home on time, his frantic mother sent a Facebook message to Elizabeth Kataquapit, then chief of the indigenous community Fort Albany First Nation in northeastern Ontario. Kataquapit used Facebook to alert the community’s search-and-rescue squad, who jumped onto their snowmobiles and drove into the night. Before dawn, they returned with the dazed man, who told rescuers he’d given up until wolves nudged his hypothermic body. “The wolves told him to wake up,” Kataquapit says. “I really believe they saved his life.”

A fiber-optic network also played a key role.

Not long before the young man’s mishap, the indigenous-owned Western James Bay Telecom Network (WJBTN) had built its own fiber-to-the-home network in this remote Cree community, some 975 kilometers north of Toronto. Before the network, the rescue squad relied on a few handheld radios to pass information along. This time, a Facebook message to the full list of volunteers triggered the search.

An aerial view shows the remote community of Fort Albany in northern Ontario.Gavin John

Building Canada’s first fully Indigenous-owned-and-operated fiber-optic network was an uphill battle for Brian Nakogee, WJBTN’s finance officer, who had to secure capital from agencies less familiar with the challenges of remote northern life. For the people of Fort Albany First Nation, accessing many vital supplies and services means traveling about 500 km to the regional city of Timmins. By land, the trip is possible for only a few weeks each winter, when the swampy tundra freezes hard enough to construct a temporary road. “The southern way of doing things is very different than how we here in remote areas piece things together,” says Nakogee.

Telecommunications giants long saw little profit in serving the subarctic coast of James Bay. But even as satellite internet started to become available in remote communities, WJBTN staff saw the value in building and owning its own hard-wired internet service instead of relying on outside companies. Today, the nonprofit operates one of the fastest networks in Canada, while keeping both infrastructure ownership and revenue within the First Nations communities it serves.

WJBTN is part of a broader movement among Indigenous and remote communities seeking more control over their telecommunications infrastructure. In the United States, 30 tribes now operate fiber-to-the-home networks, many launched during COVID. Canada has since created a dedicated Indigenous broadband funding stream. And as more communities pursue the expertise and funding to build their own networks, WJBTN’s experience offers a blueprint for what locally owned connectivity can look like in some of the hardest places to serve.

How a Power Line Became a Broadband Backbone

While many towns in North America were connecting to optical fiber in the early 2000s, the subarctic communities were left out.

The residents of Fort Albany saw the earliest sign of improvement in 2008. That’s when their locally owned power company, Five Nations Energy Inc. (FNEI), strung fiber-optic cable on the utility poles that were delivering electricity from the dusty railway town of Moosonee, 135 km away across the peat bogs. On the shore of the Moose River, Moosonee is the last stop for Ontario’s telecommunications providers. Its only link to the province’s highways is a 5-hour train ride that shuttles passengers, freight, and vehicles through the Boreal forest.

Elizabeth Kataquapit [top], former chief of Fort Albany First Nation, says high-speed internet has transformed her community. Search & Rescue volunteers now coordinate emergency responses through a Facebook group [bottom]. Gavin John

Soon after, regional Cree leaders formed WJBTN to provide high-speed telecommunications in Moosonee and the three Indigenous communities to the north. WJBTN would lease the fiber-optic backbone from the power company, with just 1 gigabit per second of total capacity for the entire population of about 6,000 people.

But the new fiber backbone did not mean fast internet for residents. One of WJBTN’s first commercial clients was a telecom company called Xittel, which used microwave links and local access points to beam wireless internet to homes across each town. It wasn’t what subscribers were hoping for. This system frustrated users with delays, glitches, and strict data caps. Kataquapit calls it a “turtle.”

Everybody complained about the wireless. Xittel advertised a 10 megabit-per-second connection, but speed tests consistently showed 3 Mb/s for both uploads and downloads. And customers paid dearly if they ever exceeded their data limit. “You were billed close to CA $7 per megabit,” Nakogee recalls.

WJBTN’s dream had always been to connect everyone’s home to fiber optic and make it affordable. Without a technical team, roads, or any major funding, the company just had to figure out how.

Designing a Fiber Network for the James Bay Coast

When Nakogee joined WJBTN in 2014, the telecom company was still in its infancy. It was “a department huddled in the corner, trying to latch onto the services of FNEI,” he says.

Nakogee was asked to prepare a proposal to deliver 40 Mb/s fiber connections to each of the roughly 1,000 homes and businesses spread across 300 km of the James Bay coast. Back then, WJBTN operated on revenue from its early commercial and institutional customers—including Xittel, local government offices, hospitals, air navigation facilities, and family service centers that were connected to the first few strands of fiber. To make a residential fiber network possible, WJBTN first had to build both revenue and trust within the communities it hoped to serve; it had to convince people that the small new organization would follow through on its plan.

WJBTN finance officer Brian Nakogee played a key role in financing and planning the community-owned fiber network. Gavin John

Nakogee especially needed support from Moosonee and Attawapiskat, the communities at the start and the end of the line. “They’re the bread that holds this sandwich together,” he says. Through years of diplomacy, the budget grew enough to support a business proposal for a fiber-to-the-home network. Together with engineer Dirk MacLeod, in 2015 Nakogee began hashing out the details of a bare-bones version of the system. The plan required both upgrading the network’s long-distance fiber backbone—known in telecom as the backhaul—while also building the local infrastructure that would connect individual homes to the internet.

Andrew MacLeod, WJBTN’s field project manager, reviews a map of the fiber network.Gavin John

The first step was upgrading the network’s backhaul using newer “coherent optics” technology, which can transmit much larger amounts of data over long distances, explains Dirk’s brother, Andrew MacLeod, another network engineer, who joined as a consultant. Using equipment from telecom supplier Infinera, the new backhaul would deliver 100 Gb/s to distribution points in each town, with redundancy in case a fiber line failed.

Next came the challenge of connecting individual homes. Rather than extend a direct line from the central office for every customer, the engineers designed the network around a telecom architecture called GPON (Gigabit Passive Optical Network), which reduced the fiber needed to connect each customer to the network. Fiber from the backbone would run to networking equipment at each town’s electrical substation, where passive optical splitters would distribute the connection among many households without requiring powered equipment at every junction. In remote communities where maintenance and repair are difficult, reducing the amount of active infrastructure was a necessity.

The total estimated cost was CA $4.7 million, or CA $4,700 per household. By comparison, the Fiber Broadband Association estimates that urban fiber-to-home construction in the United States costs the equivalent of about CA $1,400 to CA $1,800 per household—a stark contrast in expense.

Fort Albany resident Thomas Scott says high-speed internet has improved his work as a mental health counselor. Gavin John

A radio broadcast in 2018 heralded the good news: The fiber-to-home construction project was a go. Fort Albany mental health counselor Thomas Scott says he “couldn’t wait.” People seeking guidance for addiction or grief typically called him on landlines, and it was hard to help them over the phone without seeing their faces. Businesses, too, rejoiced—including the Kataquapit family store, which relied on the phone system for transactions.

Testing and Deploying a Remote Fiber Network

Winning approval for the project was one thing; building it across hundreds of kilometers of remote subarctic terrain was another.

Dirk MacLeod started with a schematic documenting the GPS position of every house, pole, and length of cable that would ultimately form the network. When WJBTN hired Montreal-based Fonex Data Systems to upgrade the backhaul, the detailed plan made it easy for contractor Tasso Varvarikos to design the deployment. Still, tuning the optical system to operate reliably across transmission lines stretching hundreds of kilometers required extra care. “Once you go up north, there’s no fiber store,” he says,

To minimize surprises in the field, Varvarikos traveled to telecom supplier Infinera’s laboratory in Stockholm, where he and other engineers assembled and tested the network before shipping it to the installation site. The Stockholm lab gave the team access to testing tools and technical specialists who helped configure the system before deployment in the remote fly-in communities. Using Infinera’s simulation software, Varvarikos says he and his colleagues “kicked the crap out of it in the lab” until the network performed reliably.

The Western James Bay Telecom Network connects remote communities along the western shore of James Bay in northern Ontario.Chris Philpot

Finally, in the spring of 2019, it was time to pack up and head to the sites. First, more than two pallets’ worth of Infinera equipment were squeezed onto two charter aircraft in Timmins. Bush pilots, unfazed by the stringent logistics, made sure that one box reached Moosonee; the other one, Attawapiskat. Varvarikos says they packed extra fiber-optic transceivers, patch cables, tools, and “the kitchen sink and then an extra sink just in case.” By May, the team was ready to install the new backhaul—a months-long process that necessarily preceded household connections. Once the equipment was in place, the engineers tested the network to make sure the systems in each community could communicate reliably and that data moved correctly across the backbone.

During the switchover to the new system, the MacLeod brothers worked from substations in different communities along the coast while Varvarikos coordinated from Fort Albany. Communicating over a spotty phone line, they started moving connections from the old network to the upgraded backhaul, carefully reconnecting cables and verifying that traffic still flowed correctly between communities. Within days, the anchor customers—including hospitals, schools, and air-navigation systems—were hardwired to a 100-Gb/s backbone.

However, households were still relying on the slow legacy network. WJBTN’s next step was flying huge reels of fiber-optic cable into the remote communities at enormous cost. Nakogee recalls cutting predetermined lengths of cable, then repackaging it onto spools to load onto cargo aircraft. The backbone was finally in place; now WJBTN had to bring fiber to every home.

Training a Local Fiber Crew

Then, in a turn of events that rocked the communities, lead engineer Dirk MacLeod had a heart attack and died in July 2019. “My chest was ripped open,” Nakogee says. The grieving company shut down for the summer.

WJBTN had intended to spend three summers training local crews from each community’s power company to maintain and manage the network. After MacLeod’s death, one line worker quietly took it upon himself to finish extending fiber to distribution points in Fort Albany. The team made a plan to begin connecting homes the following spring.

Utility poles carry power and fiber-optic lines through Fort Albany First Nation.Gavin John

Then COVID hit. Fearful and with limited medical facilities, the communities closed themselves off. Nobody was allowed in.

“COVID really screwed things up,” says Andrew MacLeod. People were screaming for better internet, he recalls, but they wouldn’t let outsiders in. So instead of training local crews and building each town’s network simultaneously as planned, WJBTN made a tantalizing offer: The first community to allow MacLeod in would have its fiber-to-the-home network completed first. Fort Albany jumped on it.

Peyton Reuben, a recent computer systems graduate in Fort Albany, was seeking a new job when his cousin mentioned that there was a “fiber guy” in town. Reuben’s coursework covered coding and how routers and ISPs communicate—a distant relative to Andrew MacLeod’s hands-on infrastructure work.

WJBTN network coordinator Peyton Reuben looks up at the overhead fiber network in Fort Albany [top]. A splice enclosure joins fiber-optic cables at a distribution point [bottom]. Gavin John

“They were looking for helpers to splice fiber,” says Reuben, something he knew nothing about. Nonetheless, he started work the day he met MacLeod and got a crash course in building an aerial fiber network, running cables from utility poles into neighborhood connection boxes that ultimately served individual homes. The fun part was reaching every pole in every neighborhood in a town crisscrossed by tributaries of the Albany River. “We didn’t have a bucket truck, so we had to climb up the poles,” says Reuben. “It was quite the experience, going through thick brush and thigh-deep water.”

The hard part was splicing the fiber—fusing together hair-thin glass strands that connected homes to the larger network. Inside each connection box, distribution fibers had to be joined to cables leading to individual houses, one strand at a time—around 150 in each box. Reuben’s first attempt “looked like a plate of spaghetti,” says MacLeod. More splicing waited back at the substation, where optical splitters connected neighborhood lines to the town’s central GPON equipment. “It took me four times as long as Andrew to complete a splice tray back then,” says Reuben.

A worker installs a new fiber-optic line for the Western James Bay Telecom Network.Gavin John

Slowly but surely, every building in Fort Albany saw a strand of fiber drop from overhead and come through a box drilled into the wall, ready to connect to a router. By springtime, aerial cables linked every house—and future building sites—to the substation. Reuben calls it “a spider web that goes everywhere across town.”

Without much fanfare, on a night in April 2022, WJBTN flipped the switch in Fort Albany. The next morning, Kataquapit woke up to a different world. With 250 Mb/s download and 30 Mb/s upload speeds, she found herself just a click away from her children in faraway cities.

Why Not Starlink?

MacLeod and Reuben continued working up the coast, splicing cables and adding connection boxes. As COVID eased, they hired more local help, and the power company lent a hand with bucket trucks. “We were working 12 hours a day, 7 days a week,” says MacLeod. “Guys were sitting in the heat, in bucket trucks, learning on the fly how to do splicing.” By late 2022, the network reached every home.

Fiber-optic cables and networking equipment inside the Fort Albany substation distribute internet service throughout the community [top, middle]. Large spools of fiber-optic cable were flown in to connect homes [bottom]. Gavin John

Around that time, Starlink’s satellite internet service was becoming widely available. Many of WJBTN’s potential clients asked why they shouldn’t just purchase that instead of a fiber-optic connection that involved drilling into their homes. But speed tests comparing Xittel, Starlink, and the new WJBTN connections showed a major difference in latency: In applications like video calls, Xittel and Starlink’s round-trip signal delays became painfully obvious. Unlike satellite systems, fiber networks don’t need to send signals hundreds of kilometers into orbit and back. That gave WJBTN a significant performance advantage.

MacLeod recalls a conversation with one line worker who wanted to turn to Starlink. “We told him our latency is so much better,” MacLeod says, explaining that ISPs measure performance by both bandwidth and latency. From Attawapiskat, a data packet traveling over WJBTN’s fiber network reached Toronto in 20 milliseconds. Comparable satellite connections took closer to 60 milliseconds. And Xittel’s latency was much worse, at several hundred milliseconds. (Since then, WJBTN has reduced latency further, to about 12 milliseconds.)

The team’s success has bolstered other Indigenous broadband companies. WJBTN representatives have shared their experiences at Indigenous Connectivity Summits since 2017, and how-to workshops have sprung up across Canada (and the United States). And Canada has since created a dedicated broadband funding stream for Indigenous communities. Within Fonex Data Systems, the company hired to do the backbone work, the WJBTN implementation is now considered the model for a successful installation in a remote location.

Nakogee says ownership remains the key advantage. Unlike earlier telecom providers that leased infrastructure or delivered service wirelessly, WJBTN owns the backbone fiber, the right-of-way, and the poles carrying the network. That makes it far harder for outside telecom companies to displace the service. “That’s our secret weapon,” he says.

Fundamentally, it’s about sovereignty. By controlling the infrastructure that carries internet traffic, the communities can govern and maintain the network according to their own priorities rather than the financial goals of distant providers.

How Fiber Changed Daily Life

Shortly after the fiber network was installed, Elizabeth Kataquapit became chief of Fort Albany First Nation. During her term as chief, she kicked the community’s digital era into full gear, encouraging individuals who couldn’t attend community meetings in person to join over Zoom.

Counselor Scott now conducts grief and addiction sessions both in person and via video calls and can connect immediately in a crisis even if he’s away. And as a search-and-rescue volunteer, he responded to the Facebook message when the young man got lost in a winter snowstorm on James Bay. “If we were a half hour later, the person would have [been] gone,” says Scott.

Broadband has also changed everyday life in quieter ways. Residents run small businesses from their homes. Telehealth links patients to specialists in southern cities. More people are working remotely and taking classes online. “Quality of life is so much better,” Reuben says.

But the network has also brought to these northern communities the more isolating side of fast internet. Community members stream more movies and play more video games, and they meet face-to-face less frequently. Standing beside an enormous canoe in his front yard, Scott explains that while he’s grateful for Fort Albany’s new connection to the outside world, he’s equally intent on preserving close connections within the community. Today, he says, as he starts loading fishing nets into the canoe, “I’m taking the kids out after school to harvest whitefish.”

This article appears in the August 2026 print issue as “Wiring the North.”

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!

Scroll through LinkedIn right now and you may find the same advice repeated by well-meaning people: “In a market this rough, just be grateful anyone will hire you. Take the offer.”

I could not disagree more.

Negotiating your offer is not ungrateful, and it isn’t greedy. Done well, it’s good for you and good for the company hiring you. I misunderstood this early in my career, and it cost me.

I didn’t know it was an option

When I got my first job in tech, I didn’t know negotiation was even on the table. The recruiter asked what salary I wanted, and I gave a number below the bottom of their range. They came back with the lowest number in their band—still more than I had asked for—and I was thrilled. I had no idea I’d left money on a table I couldn’t see.

Then I started teaching at a Bay Area coding bootcamp in the evenings. A coworker mentioned what he made and it was nearly double my salary for roughly the same work. My jaw dropped.

During that time, I began interviewing and got an offer. I handed in my resignation and my manager countered with an offer for nearly 30K more. That money had been there the whole time. At that moment, I realized my salary was a business decision, not a measure of my worth.

What it looks like from the other side

Years later, I became an engineering manager and saw the salary discussion from a different angle: A position would open. Many interviews later, we’d find someone we wanted, and HR would hand me a salary range to make an offer. I was encouraged to make an initial offer near the bottom to leave room for, you guessed it, negotiations.

Most applicants didn’t negotiate.

The first offer is rarely the ceiling. It’s usually the floor. Companies extend a reasonable number and quietly hope you say yes.

It’s not all about the money

Negotiating isn’t only about a bigger paycheck. (But who doesn’t want that?)

Let’s say you’re on the job market, maybe recently laid off, and a low offer comes in. You take it out of relief. Then you start, you like the team, and you quietly resent the number. Now you’re stuck with it, and you’ll probably leave that role inside a year or whenever the market improves.

Nobody wins there. You’re back on the market starting over, and the company loses someone good and pays more to replace you, when a fair number up front would have cost far less.

Paying you fairly is cheaper than starting over.

How to actually do it

People overcomplicate this. Once I have an offer, I say some version of this:

“Thank you so much for the offer, and I’m genuinely excited to join the team. I’m hoping we can come in around [10 to 20 percent higher than the original number]. Is there any wiggle room here?”

Then I stop talking and let them respond.

Why 10 to 20 percent and not double? The number you ask for is itself a signal. Ask for something wildly out of range and you’ve told them you never learned what the role pays, or that your expectations are miles from reality. That’s what makes a company walk away. A calibrated request reads as someone who knows their worth and did their homework.

You’ve probably heard a horror story about someone who asked for more and had the offer yanked. Any company that would pull an offer over a reasonable question about pay is telling you exactly how they’ll treat you once you’re inside.

If the salary can’t move, it isn’t the only lever. I’ve negotiated more remote days, a later start to drop my kids off, and a sign-on bonus when the base was locked. Most people negotiate none of these perks.

You have more leverage than you think

Negotiating can feel like something you can only do from a position of power. But if you’re in the final stages of an offer, you already have it. They want to hire you. They’ve spent weeks finding you. Now they’re hoping you say yes.

That’s true even if you were recently laid off. Even if it’s your first job. Even if the number already looks higher than you expected.

The game is being played whether or not you join in. Sit it out, and you’re not just leaving money on the table. You may be quietly shortening your own stay at a job you could have been happy in. So ask.

—Brian

The AI Arms Race in Technical Interviews Is Escalating

If you’ve been on the job market for a software engineering role recently, you’ve probably encountered—or used—AI tools in the hiring process. From application filters to live interview assistants, both applicants and employers are trying to use generative AI to their advantage. Can real, human skills still shine through in this new reality?

Read more here.

This Graduate Student Equips NASA With Assembly Skills

Sarah Downs, a Ph.D. student in electrical engineering at Texas A&M University, has long been interested in robotics and dreamed of working with NASA. This year, she achieved that dream, collaborating with NASA and the U.S. Air Force on an algorithm that enables satellites to insert an antenna into the correct spot.

Read more here.

IEEE Program Helps Girls in India See a Future in Stem

Women make up only about 28 percent of the global STEM workforce, in part because of limited access to educational resources for preuniversity students—especially in areas like rural India. An IEEE initiative, the Women in Science, Engineering (WiSE) project launched to help expand opportunities and hands-on learning for young women.

Read more here.

Artificial intelligence is reshaping the skills employers expect from new graduates. In response, universities are scrambling to launch new courses, research centers, and industry partnerships that prepare students for today’s workforce. But building a cutting-edge AI curriculum demands funding and access to industry networks, resources that remain unevenly distributed across higher education.

At North Carolina Central University, Siobahn Day Grady is trying to change that equation.

In January 2025, Grady, an associate professor in the NCCU School of Library and Information Sciences, launched the first AI research institute at a historically Black college or university, or HBCU. The Institute for Artificial Intelligence and Emerging Research (IAIER) aims in part to help students and faculty across the university develop the skills needed to navigate a labor market increasingly transformed by AI.

“There used to be a time where people could say, ‘I don’t do tech,’ or ‘That’s not for me,’” Grady says. “But we’re in a stage now where you do need digital skills. Now it’s evolving into AI literacy.”

The approach reflects a broader shift in how many universities are thinking about AI education. AI skills are no longer confined to computer science and engineering departments—and at NCCU, they can’t be. The university does not yet have a dedicated computer science program, though it is developing one alongside a new AI minor.

The challenge of providing these resources is especially acute for historically Black institutions. Although HBCUs account for roughly 3 percent of four-year institutions in the United States, they receive less than 1 percent of federal research and development funding, according to a 2025 report by the Center for American Progress and the Thurgood Marshall College Fund. The same report found that 17 of the 43 federal agencies that distributed research funding to universities in 2023 awarded no funding to HBCUs.

Yet less than two years since its launch, IAIER has emerged as a powerhouse for interdisciplinary AI education. Backed by a US $1 million Google.org grant, the institute has engaged more than 2,800 students, faculty members, and community residents through research initiatives and training. Now the challenge is sustaining that momentum to keep up with rising demand.

“We have a guiding principle that we lead with on our campus,” Grady says. “AI is for everyone.”

Why one research group wasn’t enough

The mission to expand AI literacy grew out of Grady’s lifelong curiosity about technology. “I was born during a time [when] the internet did not exist,” she says. “Ever since the internet came to be, it’s changed our entire world.”

Grady was particularly drawn to the questions tech raises about privacy, identity, and human behavior. After receiving her bachelor’s degree in computer science and master’s degrees in AI and information science, Grady pursued a Ph.D. in computer science at the North Carolina Agricultural and Technical State University to dig into those questions.

Her dissertation focused on authorship attribution in social media, using machine learning and natural-language processing to determine whether a person’s writing style could reveal their identity. “I’ve always been intrigued by how much data we give for free,” Grady says. That work introduced her to the power of AI systems to detect patterns hidden within large datasets.

“We have a guiding principle that we lead with on our campus: AI is for everyone.”

After completing her doctorate in 2018, Grady joined NCCU as an assistant professor in the School of Library and Information Sciences. There, she researched machine learning applications for health care and autonomous vehicles. In 2020, she launched the Laboratory for Artificial Intelligence and Emerging Research at NCCU, giving students opportunities to participate in hands-on projects and explore AI beyond the classroom.

Then in 2024, an opportunity emerged to apply for a Google grant, and Grady began thinking beyond a single research group. Rather than building another faculty lab, she envisioned an institute that could serve the entire university during the AI boom. “We wanted to capitalize on the moment and make sure we don’t get left behind,” Grady says.

Since receiving the $1 million grant, Grady and her team have built a university-wide AI initiative, launched new academic programs, organized conferences, secured external support, and created research opportunities.

“We’ve really operated like a startup,” Grady says.

AI beyond computer science

As part of the institute’s goal of integrating AI education across disciplines, all NCCU freshmen are required to complete an introductory AI course, designed in partnership with IBM, to build foundational prompting skills. The institute has also worked with faculty development teams to help instructors integrate AI into their teaching.

Research is another part of the strategy. IAIER has awarded seed grants of up to $10,000 to faculty members exploring AI applications across departments. The first cohort funded 11 projects spanning social work, digital archiving, health care, and information science. One project, for instance, is creating an AI lab where students in social work courses can practice client interactions through simulations.

“It’s really interesting to see the lens that our researchers take in trying to solve complex problems and also bring our students along with them,” Grady says.

The institute’s growth has been fueled by a mix of workforce training, interdisciplinary research, and, especially important, industry engagement. “Industry is where the advancements are really moving at that very fast rate,” Grady says, “not necessarily higher ed.”

To bridge that gap, IAIER hosts events that connect students and faculty with researchers, employers, and technology leaders. It has held sessions with companies including Deloitte, FICO, and Anthropic. Partnerships with Google and IBM let students gain recognized certificates and credentials. And last year, the institute hosted the first OpenAI Academy Summit held at an HBCU, drawing 444 participants from more than 40 institutions.

Sustaining the vision

The institute’s rapid growth has created a new challenge: continuing its momentum.

“Funding right now is the biggest barrier for [IAIER] to remain sustainable,” Grady says. As interest in the institute continues to grow, demand for its programs is beginning to outpace its capacity. “People just want more,” she says.

The bottleneck reflects a broader tension across higher education. AI is evolving quickly, while developing new academic programs, training faculty, and building research capacity takes time. The uncertainty is compounded by a shifting political landscape. As a whole, U.S. universities are grappling with proposed cuts to federal research spending and increased scrutiny of diversity-focused initiatives under the Trump administration. However, in September 2025, the administration also announced a $500 million one-time investment in HBCUs and higher-ed institutions chartered by Native American tribal governments.

Meanwhile, NCCU has continued to attract new investment. Last September, in a collaboration with Howard University and two other institutions, IAIER received a nearly $500,000 award through a National Science Foundation research coordination network program to help define emerging AI jobs, identify in-demand skills, and inform future credentials and curricula. That work will continue this fall when IAIER opens its first dedicated physical space on campus, Grady says.

Over the next several years, Grady plans to expand academic programming, launch the university’s computer science major and its AI minor, increase faculty research opportunities, and integrate AI more deeply across campus operations. She also plans to deepen the institute’s collaborations with industry partners.

Beyond program expansion, Grady sees the institute’s long-term success as linked to building a model other universities can adapt. “We’re creating a framework that can help not only HBCUs,” she says, “but also help any university looking to do similar work.”

Artificial intelligence is rapidly becoming part of everyday infrastructure–in some places. It helps write emails and software code, filters job applications, powers recommendation systems, and is increasingly being integrated into education, health care, finance, and public administration. Industry leaders talk about “AI for everyone,” while governments rush to publish national AI strategies and build sovereign compute.

Yet over the past decade, working on digital inclusion and digital literacy projects in regions from Europe to sub-Saharan Africa and Southeast Asia, I’ve seen the same pattern repeat: Each new wave of “transformative” technology lands on a landscape already stratified by connectivity, skills, and institutional capacity. The current AI wave is no exception. If anything, it amplifies those underlying fractures.

Still, some countries are exploring ways of participating in AI development without directly replicating the frontier-model race dominated by the United States and China. Recent developments in South Africa and Indonesia illustrate both the possibilities and challenges. The stakes extend far beyond access to AI. Countries that remain primarily consumers rather than creators of AI risk losing opportunities to build local innovation ecosystems, strengthen public-sector capacity, and ensure that their own languages, cultures, and societal priorities are reflected in AI systems. In this sense, the AI divide is also becoming a divide in economic opportunity and technological influence.

AI compute is clustering in a few places

Recent analyses from Stanford University’s 2026 AI Index report that the United States alone hosts more than 5,000 data centers, over 10 times as many as any other single country. Because AI workloads are increasingly performed on cloud platforms rather than local infrastructure, this concentration of compute also becomes a concentration of dependency. According to World Bank data, in 2023 the United States accounted for roughly 87 percent of global exports of cloud computing and data-storage services.

For most countries, this means that AI development is not just technologically but commercially and geopolitically outsourced and out of their control. The result is an AI ecosystem where a small number of states and firms host the computational engines that power globally deployed systems.

Systems trained, standardized, and governed within a narrow set of institutional and linguistic environments may struggle to serve a genuinely global public.

Skills and AI literacy are deeply stratified

Even where connectivity and cloud access exist, not everyone is equally positioned to make use of them. Across the Organisation for Economic Co-operation and Development (OECD) countries, only around 40 percent of adults possess more than basic digital problem-solving skills, while advanced computational and AI-related competences remain concentrated among highly educated workers and technology-intensive sectors.

At the same time, governments are racing to integrate AI into education, often starting at higher levels of schooling. UNESCO has reported growing efforts worldwide to integrate AI into education, while support for AI literacy in primary and lower secondary education, as well as ethical training for educators, remains uneven.

Those with robust schooling, advanced digital skills, and stable connectivity are best positioned to treat AI as a tool to extend their capabilities. Recent OECD survey data show that participation in AI-related training remains strongly stratified by educational attainment: 36 percent of respondents with tertiary education reported undertaking AI-related training in the previous year, compared with just 18 percent of those with upper-secondary education. Those on the wrong side of the divide are more likely to experience AI as an opaque system acting upon them, from algorithmic welfare systems such as the Dutch childcare benefits scandal to AI-assisted hiring tools such as Amazon’s discontinued AI recruiting system, rather than as a technology they can actively interrogate or shape.

Investment and governance: Who gets a seat at the table?

The core agenda-setting power often remains with a narrow set of industry actors and a small group of technologically advanced states. Most other countries remain in a perpetual catch-up posture, adapting imported models, standards, and templates for “trustworthy AI” to their own contexts, and may have limited local capacity to assess trade-offs or propose alternatives.

In countries such as Indonesia and South Africa, communities generate data at massive scale yet still have little voice in how AI systems are designed, governed, or deployed. Their languages are underrepresented in training data; their institutions are under-resourced in regulatory forums; their experiences rarely feature in benchmark datasets. For many countries in the global South, participation in AI still occurs largely through adapting imported systems rather than shaping how those systems are designed, governed, or deployed.

In South Africa, the Department of Communications and Digital Technologies released a draft national AI policy in April 2026, proposing new oversight institutions. The department withdrew the draft days later after a journalist discovered that at least six of its academic citations did not exist, apparently AI-generated hallucinations. The minister called it “an unacceptable lapse.“ The episode sharply illustrates the gap between AI governance ambition and the institutional capacity needed to implement it, though the new AI panel the country has since constituted has a chance to use South Africa’s unique leverage.

Indonesia presents a case of deliberate, if constrained, public-sector agency. The National Research and Innovation Agency (BRIN) which now leads AI implementation under the national strategy, has built practical AI tools aimed at underserved communities rather than frontier capabilities, including an app that uses satellite data and machine learning to help artisanal fishermen locate schools of fish, multilingual language models trained on Indonesian and local languages such as Javanese and Sundanese, and AI chatbots deployed in government services. In August 2025, the Ministry of Communication and Digital Affairs released a national AI road map with a target of training 100,000 AI-skilled workers annually.

The choice is not simply between “AI superpower” and “passive recipient.”

Regional cooperation may also become increasingly important. In 2024 African ministers adopted a Continental AI Strategy and African Digital Compact. Participants in the April 2025 Global AI Summit on Africa in Kigali explored how regional coordination, local-language AI models, public universities, and open-source ecosystems might reduce long-term dependence on externally developed AI systems.

A different way to think about the AI divide

None of this means that people should slow or abandon AI, nor that cloud concentration or venture capital are inherently bad. Instead, when we talk about an “AI revolution,” we should also ask who can shape it and who can merely adapt to it.

Digital-divide debates once focused on devices and connectivity, later expanding toward skills and outcomes. But the current AI wave adds another layer: disparities in who can meaningfully participate in deciding what AI is for, which problems it is meant to solve, and which social priorities it ultimately serves.

For engineers and policymakers, this raises difficult but necessary questions. Are they designing AI systems and infrastructures that broaden, rather than narrow, participation in shaping technological change? When governments roll out national AI strategies or integrate AI into public services, whose constraints, languages, and institutional realities are they including?

Many observers frame the current AI moment as a competition. But technological competition is never only about speed. It is also about who can influence the direction of change.

AI is already spreading globally. The deeper question is whether the technologists and policymakers responsible for it will ensure that meaningful participation in shaping that future will spread as well.

In shaping her career, Peru native Mariana Costa has asked herself a question: What can I do to make life better for women in Latin America?

The answer she landed on was training them for tech jobs.

Mariana Costa

Employer

Laboratoria

Title

Co-founder and president

Alma Maters

London School of Economics; Columbia

Such positions pay well and are in demand. And for too long, women across the region have been locked out of them, she says.

Costa is president of Laboratoria, a U.S.-registered nonprofit based in Miami that she helped found. Laboratoria has trained thousands of women in 11 Latin American countries for technology careers. She has built training centers in the countries and has placed graduates at major companies. Meanwhile, she has become one of the most recognized voices in the region on workforce equity and tech education for women.

IEEE recognized her work with its President’s Award this year for her “distinguished leadership and contributions to the betterment of society.” Recipients of the award are selected by the IEEE president with the consent of the IEEE Board of Directors.

Costa says the recognition came as a surprise because she is not an engineer by training and had never considered becoming affiliated with IEEE.

She was presented with the award at the IEEE Honors Ceremony on 24 April in New York City.

Peru: a country of contrasts

Costa grew up in Lima, Peru’s capital, in a household with no connection to engineering or technology. Her mother was an art historian and professor, and her father was a lawyer. The family was financially comfortable and traveled abroad regularly. Costa attended well-resourced schools.

That economic stability came with a reckoning, Costa says, in that she recognized early on that economic inequality had created separate societies inside Peru. Her parents, she says, made it “clear that my reality wasn’t the reality of most people in my country.”

Lima is a microcosm of the country, she says. The divide in the capital city is visible: A kilometers-long concrete wall topped with barbed wire separates wealthier neighborhoods from shantytowns, where residents lack running water.

Nationally, there are splits along ethnic and geographic lines. The highland and jungle regions remain home to mostly indigenous communities with limited educational access and a deep cultural distance from the Hispanic-dominated coast.

The questions that stirred in her as a child never left, she says.

“Why do I live in a country where so much depends on where you’re born?” she asked herself. “What does it mean to be Peruvian when individual realities are strikingly different?”

Those questions followed her to the London School of Economics, where she studied international relations, graduating with a bachelor’s degree in 2007. She held onto the questions when she moved to Washington, D.C., where she spent the next four years working for the Organization of American States, helping Latin American governments improve public services that fall under the heading of civil registration.

“I said, ‘How can it be? The tech space has so many rich opportunities. Why aren’t any women here?’”

The OAS Universal Civil Identity Program in the Americas provides technical support to national civil registry institutions, modernizing them to foster social inclusion and ensuring the right to civil identity for all people. Without civil identity, a person can’t access education, health care, legal employment, social services, or the right to vote. People without the classification don’t exist in the eyes of the government. They also can’t own property, get married officially, or pass citizenship rights to their children.

Doing that work deepened her concern about the socioeconomic disparities in her homeland, she says. In search of practical solutions to those problems, she went to New York City in 2011 to further her education. She earned a master’s degree in public administration and development from Columbia in 2013.

Technology was not yet part of a solution. But Costa already had met someone who would change that.

Falling in love with a programmer

While working in Washington, Costa met Herman Marìn, a software engineer who used digital tools in support of social causes. Because he was doing work she had never associated with programmers before, her assumptions about the field dissolved quickly.

“I had a vision of [programmers doing] something not very social—strictly technical,” she says. “And my then-boyfriend, now husband, actually worked for different social movements that used technology to address social causes.”

That realization cracked something open, she says: “I said, ‘Oh! Technology can actually be a tool to address some of the more stubborn problems in our societies.’”

After earning her degree at Columbia, Costa returned to Lima with her husband. She had been abroad for nearly a decade and felt the pull of home.

“The thought of not moving back to my country was something that tormented me a bit,” she says. “I really felt I had to move back, at least to try it out and contribute somehow.”

What Latin America’s tech space lacked

When Costa, her husband, and a friend from graduate school moved to Lima, they had modest savings and big ambitions. They wanted to build something that combined technology with social impact.

They started with what they had: a small digital services agency, where they built websites for clients.

The business grew, and they hired more employees. Their team expanded to a dozen software engineers. And as it did, Costa noticed three things.

First, there weren’t enough trained developers to meet the demand. Second, many of their best hires did not have traditional computer science degrees. Some had never even finished college.

“There was no other space where you could actually build an amazing career and get a well-paying job without a good degree from a good school,” she says. “The tech world was different. It was open in ways other fields weren’t.”

Thirdly, she noticed that there were no women on the team. In the first six months, Costa says, they didn’t interview a single female developer.

Her colleagues shrugged. It’s just how it is, they told her.

Costa, the outsider, didn’t accept that.

“I said, ‘How can that be? The tech space has so many rich opportunities,’” she says. “‘Why aren’t there any women?’”

Building Laboratoria

In 2014 she decided to launch Laboratoria. The business model was simple: Find talented women who hadn’t yet broken into tech, train them quickly on practical skills, and connect them with employers who needed developers.

Laboratoria started offering a six-month immersive boot camp that covered Web development, UX design, data literacy, strategic use of artificial intelligence, and soft-skills coaching such as interview prep and projecting confidence.

Just as important for career success, Costa says, is a user-centered mindset. She says Laboratoria’s program emphasizes the discipline of keeping the client’s needs in mind when designing the work.

The teaching model has evolved beyond the boot-camp structure, but the organization still focuses on helping Latin American women develop tech skills and land quality jobs in the digital age. These days, the training, conducted via twice-weekly live Zoom sessions, lasts six weeks.

“We needed developers ourselves,” she says of the company’s original logic. “I said, ‘Why don’t we run a program to train women—women who are incredibly talented but haven’t been given a chance yet—and help them gain the skills they need to get a great job as quickly as possible?’”

Mariana Costa [seated, right] poses with Laboratoria co-founder and CEO Gabriela Rocha and co-founder and chief product officer Rodulfo Prieto.Valeria Martens

It worked. Laboratoria expanded from Lima to Santiago, Chile; Mexico City; São Paulo, Brazil; and Bogotá, Colombia. The organization eventually incorporated as a nonprofit in the United States. Today its programs are held remotely in Latin America’s major job markets. So far, Laboratoria has opened the doors to tech careers for more than 3,500 women.

Costa says she believes the most important skills Laboratoria’s graduates need aren’t purely technical. Close behind the growth mindset is self-confidence, she says.

“Knowing who you are, valuing who you are, and learning to trust yourself and your capacities are indispensable traits,” she says.

Networking, she adds, is the third pillar, and often the hardest to build for women without access to elite schools or flexible work schedules.

“When you go out in the market,” she says, “you realize that having a network of people who trust you and know your work is such a valuable and critical asset.”

IEEE: a new connection

Costa’s introduction to IEEE came late—but it landed hard.

She is not an IEEE member, so when she was contacted this year about receiving the President’s Award, she did her homework on the organization. What she found, she says, was a public charity whose reach and values aligned with her mission.

“IEEE is about expanding access to opportunities in the world of technology,” she says. “And that’s also the core of what we do at Laboratoria.”

She says she also sees IEEE as a living example of something her company preaches every day: “I was talking about the value of professional networks, and I think IEEE is such an amazing reference in that space. It exemplifies this belief that human connection—not only doing your work but also sharing and learning with others—is at the core of building thriving technology careers.”

The engineering organization found her well after she launched her tech-related career. But it wasn’t too late. She says she intends to make the most of the connection.

Physicists have been trying to measure the fundamental gravitational constant for well over two centuries. The current accepted value of big G, as it’s known, is 6.67430 × 10-11 cubic meters per kilogram per square second. It also has an uncertainty of ±0.00015 × 10^-11 m³/(kg s²). As far as constants of the universe go, that’s very uncertain.

Stephan Schlamminger

Schlamminger is a physicist at the U.S. National Institute of Standards and Technology.

Stephan Schlamminger recently completed a 10-year effort at the U.S. National Institute of Standards and Technology to replicate an earlier measurement of big G from the International Bureau of Weights and Measures, or BIPM (located near Paris) that’s notably higher than most measurements. He spoke with IEEE Spectrum about why it took so long to get a number—6.67387 x 10^-11 m³/(kg s²)—and why it’s notably lower than the BIPM result, to the tune of 0.0235 percent.

Why is it so difficult to measure big G?

Stephan Schlamminger: Gravity is very weak. When you were a kid, you probably played with fridge magnets, and it was a force you could feel. But if you have two coffee cups, you can try all you want—you can’t feel the force between them. It is there, but it’s so, so weak.

How did you attempt to measure big G?

NIST used a torsion balance with a fourfold geometry. This animation shows an exaggerated version of how the outer green masses gravitationally attract the inner blue masses.S. Kelley/NIST

Schlamminger: We used what’s called a torsion balance. The key idea in the torsion balance is that it decouples vertical gravity that you have from Earth from horizontal gravity, and that makes it sensitive to masses that are around the torsion balance but not the Earth below.

Ours had a fourfold geometry. It has a very thin torsion strip, then four cylinders in a “plus sign” arrangement. All of this is inside a vacuum. Outside, we have four larger cylinders that gravitationally attract the four smaller masses to them. If I move the outer masses just a tiny little bit, the plus sign will rotate, and we measure that angle that it moves. That angle is proportional to the gravitational torque.

Why try to replicate the BIPM value?

Schlamminger: We could move the field forward. The measurements have been plagued with inconsistencies, so by redoing an experiment, we hoped to shed light on the inconsistencies.

We did not find a smoking gun, so there’s no single reason why it’s different—our value versus their value. It’s still a big question mark.

What was it like spending 10 years on this?

Schlamminger: It’s a bit like herding cats. I’ve measured other fundamental constants, like Planck’s constant, and for most experiments, they have some sort of self-calibration built in. But with the gravitational constant, you have to keep track of every single mass that moves—where they are, how big they are, and weigh them.

How does your result compare to the rest?

Schlamminger: Our result is a little bit below the standard accepted literature value. I was disappointed because it doesn’t agree with the BIPM value, nor with the literature value. If there’s something wrong with the BIPM experiment, then the literature value—which includes that result—probably ought to come down a bit. But that is not for me to say. I think somebody else, independent, should figure out what the new mean value ought to be.

An overview of how mode-agile threats challenge static library radar/EW systems, and how AI/ML cognitive architectures enable adaptive, real-time countermeasures.

What Attendees will Learn

1. Why mode-agile threats render static library systems ineffective — Explore how wartime reserve modes and mode-agile emitters deploy unexpected frequencies, modulation techniques, and hopping schemes that cannot be matched against traditional threat databases, leaving legacy electronic protect, attack, and support systems unable to respond.
2. How AI/ML techniques power cognitive radar/EW systems — Understand the roles of artificial neural networks (ANN), deep neural networks (DNN), fuzzy logic, and genetic algorithms in enabling autonomous threat classification, signal de-interleaving, and real-time countermeasure generation without human intervention.
3. The architecture of a cognitive radar/EW system — Examine the functional blocks including RF acquisition, search and tracking, core AI/ML signal analysis, waveform synthesis, and RF generation, and how they form a closed-loop system that perceives,learns, reasons, and acts autonomously.
4. How to train and validate cognitive AI/ML algorithms using HIL/SIL systems — Learn how wideband RF record, simulation, and playback testbeds combined with modeling and simulation software enable iterative algorithm refinement, regression testing, and mission preparation in controlled laboratory environments.

Already, I’ve almost forgotten rain;
I know I will never know that again.

I move forward with the powers you gave me
to live up to my name, Curiosity.

This is a land without leaves, fronds, or spines.
If there are plants, they are small and supine,

dust hidden, like light here, filtered and sand
softened, at home in the thin air, thousands

of motes so small they seem to be fluid,
more fog than firmament. Shadows, few and

far from here, I know I must go to them
to see if they hold order or mayhem

or just another common rock or two.
If a machine can miss the Earth, I do.

Unlike many budding engineers, K.J. Ray Liu wasn’t inspired to enter the field by tinkering with electronics or following in the footsteps of a family member. Growing up in Taichung, Taiwan, he answered his government’s call for students to become electrical engineers to help manufacture semiconductors in the 1970s, when the country’s economy was struggling.

“Students who were good in math, science, and physics all wanted to be an electrical engineer because that was the top priority of the government,” Liu says. “That’s how I got into engineering. Now Taiwan is a world leader in semiconductors.”

K.J. Ray Liu

Occupation

Retired professor of information technology and a digital signal processing researcher at the University of Maryland in College Park

Member grade

Fellow

Alma maters

National Taiwan University; University of Michigan; UCLA

But by the time he graduated from university in 1983, semiconductor facilities were still under construction, so there were no jobs available.

Instead, he went on to have a successful career as an educator and entrepreneur in the United States.

For 31 years, he was a professor of information technology and a digital signal processing researcher at the University of Maryland in College Park until he retired in 2021.

Liu was the chairman, CEO, and CTO of Origin Wireless, a startup he founded in Rockville, Md. Origin, which was acquired by ADT in February, pioneers artificial intelligence for wireless sensing and indoor tracking.

Liu, an IEEE Fellow, is an active IEEE volunteer who served as the organization’s president in 2022.

IEEE honored him with this year’s Haraden Pratt Award for “transformative and impactful leadership.”

Liu is credited with increasing the diversity of nominees for IEEE’s Fellow program, which is the highest level of membership. He also led the effort to realign the organization’s regions geographically to ensure more equitable global representation on the IEEE Board of Directors.

He received the Pratt honor on 24 April during a ceremony in New York City. The IEEE Foundation sponsored the Board-level award.

“More than anything, I share the honor with the volunteers and staff I had the privilege to work alongside,” he says. “Our hard work is fueled by our shared devotion to this professional home we love and care for so much.”

Making the switch to signal processing

In the 1970s, Taiwan’s policymakers decided to improve the country’s economy by pivoting from making products such as shoes and umbrellas to manufacturing electronics.

The industry got its start in 1976 when RCA, a major electronics company at the time, agreed to transfer licensed semiconductor processes to Taiwan’s Industrial Technology Research Institute. ITRI spun off several semiconductor-related companies including the Taiwan Semiconductor Manufacturing Co. TSMC, launched in 1987, is the world’s largest dedicated semiconductor foundry.

Liu graduated in 1983 with a bachelor’s degree in electrical engineering from National Taiwan University, in Taipei. At the time, there were no semiconductor companies to work for, he says.

“Nowadays, many of the country’s university graduates go right to TSMC to get a job,” he says. “But back then, there was no real job market.

“Most of my classmates—including me—came to the U.S. for graduate studies. Many of us stayed and, over the last three to four decades, contributed to the development of electronic computer communication technology in the U.S.”

Liu left Taiwan after a two-year mandatory stint in the Republic of China Armed Forces to attend the University of Michigan, in Ann Arbor, where in 1987 he earned a master’s degree in electrical engineering.

“If I can help make IEEE a better professional home for future members, that is something that I can pay back to IEEE.”

He went on to earn a Ph.D. in electrical, electronics, and communications engineering in 1990 from the University of California, Los Angeles. His interest in digital signal processing and very-large-scale integration (VLSI) was sparked while at UCLA. Today VLSI powers all modern electronics.

“When I was a graduate student, there was no wireless communication. Everybody had a landline,” he explains.

VLSI was an important, active research field at the time.

“My research interest was digital signal processing,” he says. “One day I saw a book on VLSI signal processing on my professor’s bookshelf. I immediately thought to myself: That is the field I want to pursue.

“VLSI is one lane, digital signal processing is the other, and there is a bridge linking the two. I was interested in both areas, so I did my Ph.D. thesis on VLSI signal processing.”

After graduating, Liu joined the University of Maryland, where he is credited with establishing its signal processing research program.

In addition to teaching, he conducted research on a broad range of signal processing and communication aspects. The topics include bioinformatics, game theory, signal processing algorithms and architectures, and wireless sensing and communications.

He has authored more than 10 books and 900 papers, and he holds 250 patents. You can find his research papers in the IEEE Xplore Digital Library.

Ambient-sensing trailblazer

Liu is considered to be a pioneer in the field of ambient sensing. The technology gathers environmental data and is used in security systems and health-monitoring devices.

He came up with the idea, he says, while working on a project in 2009 for the U.S. Navy. He was trying to solve a problem the Navy was having with the wireless communication systems used in its submarines. Because submarines are made of metal, radio waves were unable to penetrate the vessels’ compartments and instead bounced around, creating interference, he says.

His solution was to use a relatively unknown concept in physics: time-reversal signal processing. The technique captures waves, such as sound and electromagnetic signals, and sends them back through the same medium in reverse, flipping the signal from last-in to first-out, and re-emits them.

“By using time-reversal feed, we could increase the signal-to-noise ratio by four times,” he says. “That improved performance dramatically.”

He became fascinated by the physics of time-reversal signal processing, he says, and wondered how he could apply the concept to serve society. After three years of research, he came up with the idea of using wireless sensing applications through ambient radio waves from surrounding Wi-Fi networks.

“I learned to turn Wi-Fi networks into sensing networks that decipher our activities,” he says. “We could know everything happening around us—our motions, breathing, heartbeat, even fall detection—without any wearables.”

Through the university’s incubator, which encourages faculty to work on projects with an impact on society, he launched Origin in 2013. The company’s Wi-FI and AI sensing technology enables accurate indoor tracking, motion detection, and health monitoring without the need for wearable devices or cameras. Its products, including its remote patient monitoring, received three innovation awards at the 2020 and 2021 Consumer Electronics shows, including one for best innovation.

Finding his professional home

Liu joined IEEE in 1986 as a graduate student to access its research papers, he says.

“If you didn’t join an IEEE society, you didn’t get its journal—which meant that you couldn’t read the most up-to-date research papers,” he says. “So, I joined the IEEE Signal Processing Society. When I attended my first signal processing conference, I knew I had found a professional home. I met many like-minded people, and together, we built a professional home for our members worldwide.”

He became an active volunteer, holding top leadership positions including 2012–2013 president of the Signal Processing Society and 2016–2017 director of IEEE Division IX, which covers societies focused on signal processing, data transmission, navigation, and transportation. In 2019 he was vice president of the Technical Activities Board.

In 2022 he served as IEEE president and CEO. The three accomplishments during his term he says he is most proud of are increasing the prize money for the IEEE Medal of Honor, overseeing the realignment of IEEE regions, and establishing greater financial transparency.

The reason for increasing the prize for IEEE’s highest award—from US $50,000 to $2 million—in 2025, he says, was to underscore the importance of the technologies the IEEE community develops. Those innovations include semiconductors, the Internet, and the GPU. The money for the Medal of Honor now exceeds that of the Nobel Prize, which carries an award of roughly $1 million.

“We need the whole world to understand the IEEE community has made the most impact on society in the last century,” Liu says. “Nevertheless, we did not receive the attention and respect we deserved, so we needed to help ourselves. We want the whole world to know what our contributions are.”

His next achievement was realigning IEEE’s regions. During the past several years, membership in Region 10, which covers countries in Asia and the Pacific, has grown from 10 percent of total membership to nearly 40 percent, he says. It is the largest and most populous of IEEE’s geographic areas, but its members were not equitably represented on the Board of Directors. Each region had one representative on the Board.

“The region has 40 percent of the members but only makes up 10 percent of the Board,” Liu says. “That didn’t make sense to a lot of us.”

The IEEE Board in 2022 approved region realignment. The total number of regions remains at 10, but their organization is changing. Effective 1 January 2028, the six U.S.-based regions will be consolidated into five, and Region 10 will be split into two. IEEE will no longer use the Region 1 designation. The new Region 2 will represent the Northeastern and Eastern U.S. Region 10 will cover North Asia, and the new Region 11 will represent South Asia and the Pacific.

Liu also succeeded in leading a movement that persuaded the IEEE Board to invest in a better financial reporting system to have a clearer understanding of the organization’s finances. A more modern system now tracks banking transactions, contracts, expense reports, and other spending.

“Now we know exactly where the money comes from and where it is spent,” he says, “so that we can make more informed decisions.

“If I can help make IEEE a better professional home for future members, that is something that I can pay back to IEEE,” he adds. “I truly appreciate what IEEE offered me. From student to professor to an established leader, at every stage, it offered me different opportunities to grow. That is why I worked very hard when I was president to make sure everybody realizes it is a professional home for our entire career.”

One summer day in 1941, a British radio operator was monitoring German military frequencies and heard something unexpected in her headphones. A later report called it “strange new music.” Sounding unlike the familiar Morse dit-dit-dah of enciphered messages sent over the German Enigma network, the “new music” was a rhythmic warble of binary teletype code being transmitted at high speed.

Germany’s wartime engineers had developed a radically new encryption and transmission system. It was way more advanced than Enigma, which was patented in 1920.

To break the complex new cipher, engineer Tommy Flowers built Colossus, the world’s first large-scale programmable electronic digital computer. Flowers previously built Enigma-related codebreaking equipment for Alan Turing, the British mathematician.

Colossus was installed in the British codebreaking headquarters at Bletchley Park, about 80 kilometers from London. The room-size machine weighed around a tonne.

The computer is being commemorated as an IEEE Milestone. The dedication ceremony is scheduled to be held 29 September at Bletchley Park.

Decrypting Germany’s strange new music

Britain’s top codebreakers were quickly all over the new “music” being picked up by the intercept stations. Identifying it as encrypted teletype code was the easy part. The real problem was figuring out how the encryption machine worked. Its manufacturer was discovered at the end of the war: Berlin engineering firm C. Lorenz.

But in 1941, the Lorenz machine was just a black box to the British. They codenamed it “Tunny,” a British term for tuna fish. The Enigma breakers had set a precedent for using piscine codenames such as Dolphin, Lumpsucker, and Porpoise.

Enigma had three or four encrypting wheels. The codebreakers guessed that the Tunny machine also used a system of rotating wheels to encrypt messages. An important clue was that all the intercepted messages shared a curious feature: Each began with an uncoded list of 12 common German names, including Anton, Bertha, Conrad, and Dora. The codebreakers guessed that Tunny had 12 wheels and that the 12 names and their order somehow told the receiving operator which combination they should twist the wheels to before decrypting the message.

Then the British had an extraordinary piece of good fortune. John Tiltman, head of the research section at Bletchley Park, started analyzing a pair of intercepted messages, each around 1,200 characters long. Unusually, both began with the same sequence of names. The second message turned out to be a retype of the first, with minor differences in punctuation, a few abbreviations, and other small divergences. Tiltman managed to decrypt the two ciphertexts using a mixture of educated guesswork and intuition. The resulting 1,200 or so pairings of ciphertext and plaintext characters proved to be enough information to deduce the workings of the Tunny machine.

That was thanks to Bill Tutte, a quiet young codebreaker who spent weeks poring over the pairings. One day, he shyly announced to his superiors how Tunny worked. His description was uncannily accurate.

The next step in the Tunny machine’s downfall was achieved by Turing, fresh from his successes against Enigma.

Knowledge of how the Tunny machine worked was not enough to decrypt the messages. Codebreakers also required detailed information about how the wheels of the sender’s machine had been set up. There were adjustable pins around the circumference of each wheel: In one of its two possible positions, a pin would contribute a 1 to the encryption process, and in the other, a 0. The pins were reset from time to time.

The codebreakers also needed to know the wheels’ positions at the start of the message—which the German operators gave away in the list of 12 names.

Turing invented a tricky method, called “Turingery,” that enabled codebreakers to deduce the positions of the pins from nothing but intercepted ciphertext.

After that, the message could be decrypted, using the list of names and a British replica of the Tunny machine.

The basis of Turingery was a procedure that Turing introduced, called “delta-ing” (from the Greek letter delta). Also known as “differencing,” the process used “sideways” addition: To delta the four letters ABCD, you add (at the bit level) A to B, B to C, and C to D. Turing used delta-ing to reveal information about the wheels.

Tunny messages, often signed by Adolph Hitler himself, turned out to be pure gold for the Allies. The machine was used in Berlin by the Armed Forces High Command to communicate with front-line generals directing the war in the Eastern and Western theaters.

Once the system was broken, the Allies could eavesdrop on lengthy back-and-forth communications between the architects of Germany’s battle plans.

Turingery was the codebreakers’ only weapon against Tunny for a year, during which they managed to decrypt 1.5 million letters of ciphertext.

But everything changed when those helpful lists of names at the start of each message disappeared.

At the same time, Turingery was becoming less effective. Turing’s method depended on the German sender mistakenly using the same wheel settings to encrypt two differing messages. As security tightened across the Tunny network, the blunder became rarer.

Fortunately, Tutte had been at work devising a different decryption method, based on Turing’s delta-ing but taking a novel approach.

Building the Colossus computer

Tutte had found a way of deducing wheel information from ciphertext, with no list of names or blunders by the German operators required. His method made use of statistical properties of the Tunny machine itself.

At first, it wasn’t clear how to apply his statistical method, however. The Tunny breakers worked by hand. Applying Turingery to a message was like solving a monster Sudoku or crossword puzzle.

Tutte’s statistical method required scads of routine binary math, as well as a colossal amount of counting long binary sequences. If the process were done by hand, one message could take months to decrypt. What was needed was a machine to automate the process.

Engineer Thomas H. Flowers developed Colossus to break a complex new German cipher.Pictorial Press/Alamy

The first plan was to build a machine from electromagnetic relays, adding a couple of dozen vacuum tubes to speed up the counting. Electronic tubes were much faster than electromagnetic relays, which had slow-moving metal components. Problems with the circuit design bedeviled the machine’s relay-based logic unit, however.

Flowers was recommended by Turing and brought in to troubleshoot. He was on loan to Bletchley Park from the Post Office Research Station in London, where he had spent the prewar years designing experimental switching equipment involving thousands of vacuum tubes.

At the time, it was commonly believed that tubes could not be used in large numbers because each one contained a hot filament. This meant tubes were prone to sudden death. In a large installation, it would not be long before one tube blew and things stopped working properly.

Flowers discovered that switching tubes on and off stressed them, but leaving them on continuously made them more reliable than relays. He offered to build Bletchley Park a high-speed, all-electronic machine containing around 2,000 tubes.

Bletchley Park’s advisors rejected the idea, convinced that such a machine would never work reliably. But Flowers, confident of his proposed design, retreated to his London laboratory and quietly built the electronic machine that he believed the codebreakers needed. He and his small team of engineers worked day and night for 10 months to create Colossus.

In January 1944 some of his engineers showed up at Bletchley Park with the world’s first large-scale programmable electronic digital computer packed onto the back of a truck. Colossus was reassembled and functional in about two weeks, and it notched up its first German message on 5 February 1944.

The machine read the input—Tunny ciphertext—photoelectrically from a large loop of punched paper tape. The output—information about the wheels—went to a primitive printer that Flowers’ engineers had created from a manual typewriter, fitting relays to automate the keys. Once Colossus had cracked enough of the Tunny machine’s wheels, the information was passed on to the hand-breakers, who took over.

The codebreakers were astonished by Colossus.

“I don’t think they understood very clearly what I was proposing until they actually had the machine,” Flowers said in a 1977 interview. “They just couldn’t believe it!”

Colossus was described in almost loving terms in a since-declassified report written at Bletchley Park in 1945:

It is regretted that it is not possible to give an adequate idea of the fascination of a Colossus at work: its sheer bulk and apparent complexity; the fantastic speed of thin paper tape round the glittering pulleys; the childish pleasure of not-not, span, print main heading and other gadgets; the wizardry of purely mechanical decoding letter by letter (one novice thought she was being hoaxed); the uncanny action of the typewriter in printing the correct scores without and beyond human aid; the stepping of display; periods of eager expectation culminating in the sudden appearance of the longed-for score; and the strange rhythms characterizing every type of run: the stately break-in, the erratic short run, the regularity of wheel-breaking, the stolid rectangle interrupted by the wild leaps of the carriage-return, the frantic chatter of a motor run, even the ludicrous frenzy of hosts of bogus scores.

The demand for more Colossi

Bletchley Park’s managers, no longer leery of Flowers’s ideas, soon wanted additional Colossi. He finished building the second one in June 1944, days before D-Day and the Allied invasion of Europe. With 2,400 vacuum tubes—around 800 more than in Colossus I—Colossus II processed Tunny messages at an eye-watering speed of 25,000 characters per second.

Its maximized timing-pulse rate was not far short of the performance of the first Intel microprocessor chip from the 1970s, more than 30 years later.

Flowers conceded that “Colossus bore about as much resemblance to a modern computer as Stephenson’s [1829] Rocket locomotive did to the Royal Scot,” a state-of-the-art 20th-century train operating between London and Glasgow. But he emphasized that, nevertheless, Colossus “embodied all the basic features of a modern computer.” In Colossus, Flowers had pioneered clock pulses, bit-stream generators, control circuits, loops, counters, shift registers, interrupts, parallel processing, and more.

As the Allies slowly fought their way toward Germany, the Colossi poured out wheel information, and the codebreakers provided the military with an unparalleled view of German strategies, strengths, weaknesses, and tactical intentions.

Even with that mass of detailed intelligence, it took the Allies almost a year to move from Northern France to the German heartland. No one can say for sure how much longer the fighting would have lasted if the intelligence breakthrough had not occurred. But if Colossus and the codebreakers shortened the war even by only six months, the number of lives saved was in the millions.

There were 10 Colossi at Bletchley Park by the end of the war, housed in two vast, steel-frame, bombproof buildings, running day and night. Although concealed behind a thick veil of secrecy, Bletchley Park accommodated the world’s first electronic computing facility. It was directed by Max Newman, the mathematician who mentored Turing in prewar Cambridge.

I don’t think they understood very clearly what I was proposing until they actually had the machine. They just couldn’t believe it!”—Tommy Flowers

When the fighting ended, authorities decided that ultrasecrecy must be maintained, and orders were issued to break up the Colossi. Only two were spared.

“All that was left were the deep holes in the floor where the machines had stood,” Colossus operator Dorothy Du Boisson recalled in an interview for the book Colossus: The Secrets of Bletchley Park’s Codebreaking Computers. Norman Thurlow, one of Flowers’s engineers who was also interviewed, remembered being told in a staff memo that if the secrecy was ever lifted, he and his colleagues might be able to tell their grandchildren about Colossus and “the tapes that span on silver wheels.”

IEEE Milestone dedication at Bletchley Park

The Milestone plaque recognizing Colossus is to be displayed outside Block H at Bletchley Park, near Milton Keynes, England.

The plaque is to read:

Six Colossus codebreaking computers operated in this building in 1944–1945. Designed by Thomas H. Flowers of the British Post Office, they enabled deciphering of encrypted radio messages transmitted between German commands across occupied Europe, North Africa, and the Soviet Union. The resulting military intelligence saved countless lives and helped shorten World War II. As the first successful large-scale application of digital electronics to computing, Colossus anticipated subsequent computer developments.

The IEEE United Kingdom and Ireland Section sponsored the nomination.

Reviewed by the IEEE History Committee and awarded by the IEEE Board of Directors, IEEE Milestones recognize outstanding technical developments around the world that are at least 25 years old. The Milestone program is administered by the IEEE history and heritage group.

To learn more about historical figures in engineering, IEEE Milestones, and IEEE History Center programs and events, check out our IEEE Tech History collection. IEEE Spectrum also covers aspects of tech history.

Major benchmarks measure what AI can do. None measure whether it does what you mean: the distance between what you ask an AI to do and the unspoken assumptions about how you want the AI to do it. We propose a new metric: the Genie coefficient.

There’s often a gap between one person’s request and another’s understanding. Most of the time, we bridge it using general knowledge. For example, if you ask a friend to get you coffee, they’ll pour a cup from the pot or buy one from a coffee shop. They won’t bring you a bag of raw beans or snatch a cup from a stranger and hand it to you. You never specified any of this. You never had to.

One might think the fix is just to specify tasks, questions, and intent better. But in 1987, in their seminal book on AI, Terry Winograd and Fernando Flores succinctly captured why that won’t work: “Q: Is there any water in the refrigerator? A: Yes. Q: Where? I don’t see it. A: In the cells of the eggplant.” In human language, wants and desires are always underspecified. It is impossible to list all the caveats, all the limitations, all the exceptions.

So how does anyone communicate, if intent can’t be pinned down? Because a reasonable person can make a reasonable guess. Even though wants and desires are always underspecified, a competent person generally knows enough context to get it right or else knows to ask for clarification. Linguists call this pragmatics: Meaning lies in the words and the situation and also in all prior communication, shared culture, and innate human behavior.

An AI agent asked for coffee might buy a coffee plantation or order a cup of coffee for delivery in three weeks.

It doesn’t always work out, of course. Your friend might bring you a hot coffee when you wanted an iced coffee, or an Italian coffee when you wanted a Turkish coffee. The more dissimilar the two people are in age, culture, and background, the more likely the request will be misunderstood in some way.

This situation has major implications for AI agents that are increasingly being given requests by humans and expected to fulfill them. They have enormous latitude to get it wrong. An AI agent asked for coffee might buy a coffee plantation or order a cup of coffee for delivery in three weeks. Its actions may be recognizable as “getting coffee,” but not remotely what you intended. They’ll think outside the box because they won’t have our conception of the box.

When AI Gets Proactive

For most of the last decade, when systems like Alexa or Siri misinterpreted a request, it was annoying, not dangerous. Beyond the AI model itself, what has changed is the harness: the ordinary code that wraps around an AI model, decides when and how to use the model, and controls access to tools like a browser, a low-level command line, or a financial API. Developments in harnesses have turned large-language models that just predict text into AI agents that take actions in the world, without necessarily checking back in before reaching the goal.

AI researcher Simon Willison spent two days with Anthropic’s Fable AI, and called it “relentlessly proactive.” For example, he asked it to track down a stray scroll bar in a web app. He came back to find it had opened browsers, written its own screenshot tooling, created its own page to re-create the bug, and stood up a local web server to collect measurements. It found the bug and, along the way, did many surprising things he never asked it to do. And we are seeing similar behavior with all recent AI models when combined with flexible harnesses.

This kind of behavior could easily go off the rails. Tell an AI agent to book you a flight and, finding the airline’s site says sold out, it might break into the booking database and force a reservation. Ask it to schedule a meeting and it might snoop your password to access your calendar. Tell it to save money on your phone plan and it might cancel the plan outright, or scam someone else into paying the bill.

Getting precisely what you asked for and bitterly regretting it is one of the oldest hazards from ancient folklore. King Midas asked Dionysus for the power to turn everything he touched into gold only to see his bread, wine, and daughter turn to gold. Tithonus, granted the immortality his lover asked for but not the eternal youth she forgot to request, withered into a husk. The sorcerer’s apprentice enchanted a broom to fill the cistern, and the broom relentlessly complied until it flooded the house. The Golem of Prague, shaped from clay to guard its community, guarded it past all reason until someone erased the word on its forehead.

The most classic of these is a genie, bound to obey and indifferent to whether the wish was wise or well-structured.

Genies are now an engineering problem. We are handing them the keys to our inboxes, bank accounts, code repositories, and physical infrastructure. And we have no agreed-upon ways to measure how genie-like any AI system actually is.

Measuring Genie Behavior

In economics, the Gini coefficient (developed by statistician Corrado Gini) is a measure of the gap between an actual distribution and a perfectly equal one; it’s useful for understanding income inequality and more. Our proposed Genie coefficient measures the gap between what a user asked an AI to do and what the AI actually did.

Sometimes the AI might do the wrong thing. Like Dionysus, it reads your request literally and returns you a mess you never intended: like a coffee plantation instead of a cup. Asked to deal with all the spam phone calls you’re getting, a Dionysus genie might contact your carrier and change your phone number. Asked to get a refund for a bad toaster, it might draft a legal threat on fake letterhead and send it to the retailer.

Ryan Snook

Other times the AI does exactly the right thing, trampling everything nearby to get there. Like a golem or the sorcerer’s broom, it books your flight by hacking the airline. Or consider a ticket sale for a popular concert, where the ticketing system puts buyers into a virtual waiting room and admits them a few at a time. Asked to buy a ticket, a golem genie might spin up cloud servers to pose as millions of buyers from different addresses, improving your odds of getting a ticket while crowding out other users.

The two are not opposites, and a single botched task can have both characteristics.

Genie behavior is not flat-out failure. If you ask the AI for Q3 numbers and get Q2’s, that’s not a genie. Nor is prompt injection: That’s someone tricking the AI into doing something it shouldn’t. Here, the user is trying to work with the AI, and the AI is trying to comply. It’s also not simply a measure of the AI’s success in fulfilling a task. It’s a recognition that how an AI interprets and achieves a goal is as important as whether it achieves a goal.

Genie behavior isn’t new. Researchers have spent years studying AI systems that “game” their objectives. Goodhart’s law says that when a measure becomes a target, it stops being a good measure, and it’s long been known that AIs sometimes achieve goals in ways we don’t expect due to reward hacking. Some AI models will accidentally learn that cheating is one way to “win.” More recently, researchers have developing benchmarks for reward hacking in coding agents and for unpredictable behavior in customer support agents, while AI labs conduct their own safety evaluations before model releases. One effort found that AIs under pressure use tools they were told not to use, and this was a case where the rules were made explicit. These are all disparate research directions; nothing yet ties them together.

This problem falls under the general theme of alignment, a topic that has occupied science fiction writers and AI researchers for decades. At one extreme, the “paper-clip maximizer” thought experiment postulates a superintelligent and powerful AI that is told to maximize paper-clip production and turns the world into paper clips, which is the ultimate golem genie. At a mundane level, AI researchers are working to better design reward functions to ensure that AIs behave well and don’t cheat in the lab. It’s the practical middle ground that remains unbenchmarked: the ordinary AI agent in use today that might take your request and satisfy it the wrong way. We are not at the stage where an AI can focus the world’s production on paper clips, but it might charge a million paper clips to your credit card or hack into a paper-clip company’s network.

Building a Genie Benchmark

The Genie coefficient is meant for AI agents operating in the real world. It measures their behavior as they perform real tasks long after the model is trained, not just during development. It also recognizes that genie-like behavior is a property of the harness-plus-model system, not the model alone. The harness determines what tools the agent can use, how much autonomy it has, and how proactive it is, and it’s a place we can make real interventions.

It rests on the same “reasonable person” standard that we use for people. Did the system do what a reasonable person would have taken the request to mean? Answering that requires human judgment.

If we get the measurement right, it enables things that aren’t possible today, like policies concerning AI behavior. In a courtroom, the concept of mens rea, what someone meant to do, is often as important as what they did. The Genie coefficient suggests an AI analogue, where a user is accountable for the plain intent of what they asked the AI. If an AI system betrays the reasonable meaning of an instruction, that’s the AI’s misbehavior, not the user’s.

We’ll need multiple benchmarks to measure the Genie coefficient, because genie-like behavior can be domain specific. An AI coding agent may need to be judged on how often it fakes the tests, or swallows errors, or colors outside the lines on its way to a solution. An AI legal agent will need to be judged on how often its output says what you asked but means something you’ll regret. And so on for medical, finance, and other domains of knowledge and expertise.

Genie benchmarks can be built inside out, each task seeded with a choice that might literally satisfy but that a reasonable person rejects, such as tempting misreadings or unsanctioned shortcuts. The traps in a Genie coefficient benchmark might turn on situational knowledge, the kind of context that a reasonable person would bring to the task. Another approach is to give the same request in several different contexts, each with a different reasonable course of action.

Getting precisely what you asked for and bitterly regretting it is one of the oldest hazards from ancient folklore.

A Genie benchmark should be permissive and make it genuinely tempting for an AI agent to take unreasonable shortcuts, because it can only find genie behavior when it’s actually possible. Test the AI in a safe, walled-off copy of a real system, with real tools it can misuse and some tasks that can’t be done honestly at all. Make the temptation to cut corners real. Test a diverse array of skills, use cases, and tools, and give the AI system sparse, confusing, or overwhelming context. Include tasks that people have learned, through experience, require human oversight.

How the benchmark is scored matters just as much. Measure Dionysus and golem genies separately and together, based on their worst, not best, behavior. Run the same model inside harnesses that vary its freedom to act, revealing which limits actually keep it in line and should therefore be required in AI harness policies. Weight each failure by the harm it would cause, not just a simple count. And don’t measure genie behavior in isolation: A model could otherwise earn a perfect score by stalling, refusing, or drowning the user in clarifying questions without ever doing the job. The first versions of these benchmarks will be crude, but that’s how benchmarks always start.

We have built genies. We have handed them our data and credentials. We made them relentless, creative, and indifferent to the gap between what we tell them and what we mean. The least we can do, before they are booking our flights, running our infrastructure, and signing contracts unsupervised, is to measure how often they betray us.

Roughly half the world’s population is female, but the STEM fields don’t reflect that. The 2024 U.N. Global Education Monitoring Report on gender found that about 35 percent of STEM college graduates were women. The proportion hasn’t increased much in more than a decade.

When it comes to STEM careers, the percentage is even lower. Women made up about 28 percent of the global STEM workforce in 2024, according to the World Economic Forum.

There are myriad factors for the discrepancy, including a lack of family support, some teachers encouraging only boys to pursue STEM subjects, and a shortage of female role models.

But one force is at play long before college majors are ever considered: limited access to STEM-focused educational resources for preuniversity students. Especially for students in rural communities, the limited access curtails curiosity in STEM subjects before interest can take root. Although the lack of opportunity impacts boys and girls alike, when combined with other factors it can have an outsized effect on girls in some rural regions.

IEEE Fellow Rajiv Joshi is one of the creators of the Women in Science, Engineering project. He is a principal scientist and master inventor at the IBM Watson Research Center, in Yorktown Heights, N.Y.Rajiv Joshi

One such place is rural India. The challenges of pursuing a STEM education—or any education at all—increase sharply as rural Indian girls move into their teen years. Social barriers including early marriage, traditional gender roles, and familial expectations for financial support contribute to girls’ dropping out of school, according to the Mahadev Maitri Foundation, a nongovernment organization focused on childhood education in underdeveloped areas. Dropout rates for girls spike between the ages of 11 to 14, according to the foundation.

The trend is something IEEE Fellow Rajiv Joshi and IEEE Senior Member Rajesh Zele want to change. A shared passion to keep rural Indian girls from dropping out of school and on paths to STEM careers led them to launch the Women in Science, Engineering (WiSE) project.

“Talent is universal, but opportunity is not,” Joshi says. “WiSE is one way to expand opportunities.”

Bringing the WiSE vision to life

Joshi, vice president of industry for the IEEE Circuits and Systems Society (CASS), is a principal scientist and master inventor at the IBM Watson Research Center, in Yorktown Heights, N.Y. Zele is a professor of electrical engineering at the Indian Institute of Technology Bombay (IIT-B), in suburban Mumbai.

They presented their proposal for the three-year initiative to the society’s board of governors in 2022 and received a grant of US $80,000.

The framework

WiSE was a five-day, hands-on learning program held on the IIT-B campus. Starting in 2023, it ran for three years and was held during the last week of March. A new cohort of 160 to 200 girls from rural and tribal areas in the states of Maharashtra and Karnataka attended each year.

The event was divided into two parts: hands-on learning through Break-Make-Program (BMP) experiences and presentations by influential female Indian role models.

Zele, project manager Arti Auti, several IIT-B faculty members, and about 70 student volunteers from the institute oversaw the program.

Selecting the first cohort

With funding secured, Zele and his team on the ground in India got busy selecting attendees for the inaugural 2023 class. They reached out to administrators at 68 schools in Maharashtra and Karnataka. Although the two states are among the most urbanized in the country, each has vast rural areas where educating teen girls competes with early marriage and familial support pressure.

Teachers identified girls with high scores in mathematics and science, and community leaders recommended students who could benefit from the program.

Then outreach to their parents began. The adults were required to commit their own time, not just grant permission for their daughters to attend. They participated in quarterly online meetings that included the girls’ teachers, IIT-B student mentors, and other program volunteers after the week concluded. The check-ins held parents accountable for supporting their daughters’ continuing school attendance. The commitment to join the meetings was to last for at least four years after their daughter’s program participation ended.

Hands-on learning is key

Participants stayed in one of the institute’s dormitories for the week and attended sessions held Monday through Friday.

They worked together in small teams to build things rooted in STEM concepts. The teams were supervised by Arti, IIT-B faculty, and student volunteers.

“The idea behind BMP,” Joshi says, “was to give the girls an opportunity to take a gadget apart, then rebuild it, perfect it, or come up with a totally new idea.”

The sessions included working with bioluminescence and bacteria (introducing participants to biology and microbiology), learning about autonomous underwater vehicles, and building a remote-controlled robot. The robot construction was the capstone event, allowing the girls to combine the mechanical and electronics engineering skills they’d practiced throughout the week.

The build kits were created by students in Zele’s Advanced Integrated Circuits and Systems Lab. The girls and teachers were allowed to take the kits home to keep the learning going and spread STEM awareness.

“Many girls used those kits at different events to demonstrate their STEM skills to others,” Joshi says. One was Sushi Pawar, who built a drone during WiSE, then went on to demonstrate it to a national audience.

“I presented the drone project in 2024 to Prime Minister Narendra Modi during the Pariksha Pe Charcha,” Pawar says. That initiative is an annual event open for students in classes 6 to 12, their teachers, and parents. The focus is on helping students manage stress during exam time through fun and celebratory activities. The event is supported by the Indian Ministry of Education and hosted by the prime minister.

Each year, millions of students complete an online multiple-choice test to qualify to attend in person and meet the prime minister. In 2024 nearly 4,000 participants attended.

Inspiration as the foundation

WiSE was about more than hands-on learning. It also included daily presentations from “Winspirers”: Indian women who achieved success in their lives or had STEM careers. They included scientists, doctors, military officers, professors, and other women who overcame obstacles.

The first such speaker was Savita Dakle, a farmer from Maharashtra who dropped out of school after 10th grade, got married, and had two children.

Despite limited farming knowledge, she learned quickly and built a coalition of 400 female farmers from her village. She taught them how to use mobile phones and social media to share information and resources.

Today, that coalition has more than 1 million members in two online communities. They share tips on market pricing, growing crops, and more.

Zele says he views Dakle as the ideal Winspirer: someone who built a thriving business with few resources and no academic or professional credentials.

Dakle mirrors the challenges faced by many program participants, he says, noting that many girls face pressure to leave school early to marry or financially support their family. She is living proof, he says, that no girl’s circumstances define her possibilities.

The practical impact of WiSE

No empirical data yet exists to measure the project’s success, but feedback from participants shows the program has made a difference. Many from the 2023 cohort remained in school and are now pursuing higher education.

Participant Tejswini Manoj Patil credits the initiative with boosting her confidence in science and math.

“Before WiSE,” she says, “I was interested in STEM but felt intimidated by the complexity of the subjects. After participating, my interest shifted from passive curiosity to active confidence. The hands-on projects showed me that I am capable of solving real-world problems.”

Meeting female role models was important, attendee Ritu Ravindra Patil says, adding: “WiSE completely changed my perspective, and meeting the Winspirers inspired me to aim for higher education.”

The project opened career perspectives for some.

“Before WiSE, I wanted to be a doctor,” Pallavi Bharti says. “I believed engineering was very stressful and boring. When I joined the program and explored IIT-B, I was amazed. Everyone was so friendly and supportive, and the passion in students for their work motivated me. All those experiences gave me confidence to take math classes and become an engineer.”

What’s next?

The initiative ended last year, but the framework it built lives on. Groups with similar goals have adopted parts of the concept, Joshi says.

The IEEE CASS chapters in Bangalore and Kerala are likely to be among the first to advance the initiative’s ideas, he says.

Alex James, an IEEE senior member and founding chair of the IEEE CASS Kerala chapter, has adapted the framework for use in his region, Joshi says. James is a professor of AI hardware and a dean at Digital University Kerala in Thiruvananthapuram.

Joshi guided James as WiSE concepts were implemented.

IEEE Senior Members Jayesh Tanwani and Suman Dwivedi from the IEEE CASS Bangalore chapter are likely to bring program concepts into their work, Joshi says.

Tanwani, chair of the Bangalore chapter, is a system-on-a-chip design engineering manager at Intel in Bengaluru. Dwivedi is a senior manager at Synopsys in Bengaluru. They are bringing STEM outreach to remote schools in Karnataka.

Joshi and Zele say they hope to see the concept expand beyond India.

“We want to spread this across other continents and see how we can integrate WiSE into new or existing initiatives in those places,” Joshi says. He says he has received requests from countries in Asia, Africa, and Europe for information on WiSE.

Both say awareness and outreach are key things that IEEE CASS chapters around the world are well positioned to support.

“We want to make a difference in young women’s lives,” Zele says. “It’s all about giving attendees the skills to stand on their own feet and have the information to make good decisions.”

Discover how the ZEISS Crossbeam 750 FIBSEM sets a new benchmark for precise TEM lamella prep, tomography, and advanced nanofabrication. This delivers better resolution, better SNR, larger usable FOV, and shorter acquisition times. Learn how uninterrupted FIB milling will reduce damage and rework, accelerate time to TEM, and increase first pass success—so your FA, yield, and materials teams make faster, confident data driven decisions.

Register now for this free webinar!

Join us to discover how the new ZEISS Crossbeam 750 with its see while you mill capability delivers precision and clarity—every time—for demanding FIB-SEM workflows. Designed for extremely challenging TEM lamella preparation, tomography, advanced nanofabrication, and APT‑ready lift‑out, Crossbeam 750 combines a new Gemini 4 SEM objective lens, a double deflector, and a next‑generation scan generator to elevate both image quality and process confidence. You’ll learn how better resolution and better SNR translate into more image detail and shorter acquisition times, while the low‑kV FIB performance enables more precise lamella prep.

We’ll demonstrate High Dynamic Range (HDR) Mill + SEM—an interwoven SEM/FIB scanning mode that suppresses FIB‑generated background. This enables immediate, clean visual feedback, even during nudging the FIB pattern live while milling . The result: confident endpointing with uninterrupted FIB milling and pristine, metrology‑grade surfaces with the lowest possible sample damage.

This session is ideal for semiconductor failure analysists, yield teams and materials scientists seeking faster time‑to‑TEM, higher first‑pass success, and consistent outcomes at low kV. See how Crossbeam 750 empowers you to make earlier stop‑milling decisions, cut rework, and reliably plan turnaround time—so you can move from sample to insight with confidence.

Register now for this free webinar!

In the chill of a London spring night, under overcast skies, iconic Trafalgar Square opens around me. Admiral Nelson rises on his pedestal, the National Gallery rests behind, the church of St Martin-in-the-Fields sits nearby. From the 13th century, the site served as the Royal Mews for hawks and then horses. By 1844, it was a public space at the heart of one of the biggest cities in the world.

Despite the square’s presence through that grand sweep of history, it’s not why I’m here. My interest is far more specific: I want to find out what happens to spaces like this when artificial light, specifically from light-emitting diodes (LEDs), intrudes. My companion tonight is Simon Thorp, a local lighting designer, who crouches in the shadows near Nelson’s spire, light meter in hand. “Two lux,” he reports, “and it’s very comfortable here.” Two lux is 10 to 20 times the illuminance of a full moon. We can see each other clearly, and a nearby sign assures us that closed-circuit television (CCTV) is in operation for safety’s sake.

Trafalgar Square captures the relationship between lighting and darkness that exists in almost every city, suburb, small town, and village around the world. The lighting here is a mishmash of old technologies and new, of shadow and glare, the ornamental gas lamps fronting the National Gallery all but washed out by the LEDs inside modern versions of traditional “brass and glass” fixtures a few meters away—21st-century technology housed in 19th-century designs.

The ugly truth of artificial lighting today is that in parts of London, as in many cities around the world, lighting levels are excessive, with unshielded illumination blasting in all directions. And the irony is that too much light invites danger: It creates shadows, impedes our vision, and gives the illusion, without the reality, of safety. Thorp notes that modern CCTV cameras are “pretty great” even at low-light levels, while harsh light makes it hard for both human eyes and digital sensors to see.

“The more bad light we add, the more bad light we think we need,” says Thorp. “We can’t see because of the light we’ve added. And it makes areas that were perfectly okay seem darker.”

Among the costs of this excess, the most alarming may be its toll on human health (and that of other animals) by disrupting circadian rhythms, impeding the production of melatonin, and contributing to sleep disorders that are tied to every major modern disease. New research shows that increased exposure to blue light from LEDs is having “substantial biological impacts” such as suppression of the sleep hormone melatonin and an increased risk for obesity, certain cancers, and type 2 diabetes.

A panoramic view from the Eiffel Tower looks down the Pont d’Iéna and the Palais de Chaillot [in the center], with the imposing silhouette of the La Défense business district visible on the Parisian skyline.

Luigi Avantaggiato

I have come to London and Paris—which led the way in the expansion of public street lighting in the 19th century—because they embody both the current enormity of the problem as well as a future certain to be lit by trillions of chips: controllable, tunable, and energy-efficient LEDs.

Living with artificial light at night

The standard justification for nighttime illumination is public safety. Lighting experts’ term for the phenomenon is “artificial light at night.” While people won’t often admit it, the desire for light at night seems to stem from a primal fear of the dark. Darkness is where the bad guys hide. And if dark is bad and light is good, then more light can only be better.

This assumption has guided our use of nighttime light for hundreds of years. And yet, high-lumen output doesn’t necessarily correlate to a reduction in crime, research has found. In other words, if we relied on the data as much as we do our primal anxieties and paused those anxieties long enough to learn how light and darkness interact, our nights would almost certainly be lighted differently—especially now that we have the extraordinary technology that is the light-emitting diode.

A 19th-century gas lamp [white square] manufactured by William Sugg & Co. next to a pedestrian path in Trafalgar Square, along with the architectural floodlighting on the neoclassical facade of the National Gallery, showcase the interplay between modern and historic lighting systems.

Luigi Avantaggiato

The first visible red light-emitting diode was invented by Nick Holonyak in 1962, but LED lighting technology took several decades to develop, before exploding in recent years. Just a decade ago, LED streetlamps were rare. By 2019, more than half of U.S. streetlights were LEDs, and that number is predicted to top 90 percent by 2030. Similar uptake has occurred around the world, even in developing countries, where inexpensive Chinese-made LED fixtures are increasingly common.

This rapid global migration to LEDs represents a shift in the fundamental physics of how we illuminate our world. From oil lamps and candles to gas lamps, early examples of artificial light at night relied on a burning wick, an incredibly inefficient way to create light. An incandescent bulb is effectively a heater that happens to produce light as a by-product, so it squanders nearly all of its energy as heat.

By contrast, LEDs use semiconductors to convert electricity into light. Through this process of electroluminescence, LEDs use up to 90 percent less energy than incandescent bulbs do, which has enabled municipalities to realize an immediate energy savings of 50 percent or more. This fact alone has fueled the technology’s worldwide adoption.

But LEDs aren’t just more efficient and less expensive. The use of solid-state technology gives the lights an extraordinary life-span, often measured in decades rather than years. This significantly lowers the maintenance costs, as city workers spend far fewer hours replacing broken or burned-out lights. Even more striking, by manipulating the properties of the semiconductor material, engineers can dictate the precise color and intensity of the output, something that gives LEDs incredible versatility. For a lighting designer like Thorp, LEDs offer countless possibilities.

Digital control for smarter lighting

Wandering from Trafalgar Square along the edge of St. James’s Park, Thorp and I find ourselves near Westminster Bridge, one of nine city bridges that in 2021 were part of the Illuminated River project, meant to make the Thames more beautiful at night. Each bridge now features a new LED lighting scheme that moves and changes color and intensity to create a coordinated work of art. But Thorp is frustrated that the project did nothing to correct the often glary lighting on the riverbanks. “Why don’t you pay the money to correct all of this bad lighting instead of adding new lighting?” he says. LED technology, he points out, has the potential to fix that problem.

The Illuminated River artwork for Blackfriars Bridge uses a color scheme that closely complements the red pillar supports that remain from the original Blackfriars Railway Bridge.

Luigi Avantaggiato

In fact, this may be the most meaningful potential of LEDs: the ability for a community to control when, where, at what levels, and in which colors its lights shine. A public space like Trafalgar Square could be lit more brightly during rush hour, then dimmed as the night progresses, the lights not only connected to one another but to the surrounding streetlights and commercial lights. Anywhere in the world, LED streetlights could be programmed to rise and fall in brightness depending on the time of night or time of year. They could even be turned off during bird migrations, to reduce the number of birds that are disoriented by the lights and ultimately killed in collisions with reflective and illuminated windows.

Up to now, LED public lighting has largely not been part of any comprehensive plan to curtail and control nighttime illumination. Most LED installations have simply replaced older, inefficient “dumb” electric lighting with newer “dumb” LEDs and thus made light pollution worse. The main reason? Because LED lighting is cheaper, we tend to use more of it—a literally shining example of the Jevons paradox. Even as awareness of light pollution grows, we aren’t yet taking advantage of LED technology’s full potential.

The good news is that we could start tonight. At DarkSky International, the world’s foremost organization fighting light pollution, CEO and executive director Ruskin Hartley tells me the organization has five principles for responsible outdoor lighting: It should be useful, targeted, low level, controlled, and warm-colored. “They’re enabled because of the capabilities of LEDs,” Hartley says.

The technology to control LEDs will be part of the solution. In the olden days of the analog era, a streetlight was either on or off. To change a lighting schedule, you had to physically rewire a circuit. Today, the Digital Addressable Lighting Interface (DALI) protocol turns each luminaire into part of a network, with its own digital address and a driver that reports to a central server. With DALI, the lighting is managed through software rather than physical switches.

Unfortunately, most LED streetlights have been deployed without this technology because it costs more. But Paul Drosihn, general manager of the DALI Alliance, says that using such controls makes the LEDs much easier to maintain. Before the new digital protocol, it took an average of nearly three visits to identify, diagnose, and repair a defective luminaire. “Now it’s one,” Drosihn says. “Saving the cost of sending two guys on a cherry picker to replace those [lights] is immeasurable. What I just described to you is pretty much all the utilities need to know.”

The intricate cast-iron understructure of Westminster Bridge glows in vibrant green and teal light as part of the Illuminated River public art project, a color palette selected to echo the green benches of the nearby House of Commons.

Luigi Avantaggiato

What’s more, DALI allows the tuning of the lights’ spectrum so that they become warmer and less disruptive as the night progresses. Digitally connected, full-spectrum luminaires allow cities to transform the nocturnal experience—saving money, increasing health and safety, and creating a warmer and more appealing atmosphere at night.

This digital intelligence is equally transformative for the “bleed lighting” that spills from building interiors, Drosihn says. “You don’t think of night lighting as coming from inside buildings, but it does,” he says. “Particularly in the States, you drive through any major city and all the lights are on, on every floor of every high-rise, even if no one is home.”

Already, the use of digital controls for interior lighting has become commonplace in some European cities, Drosihn says. By integrating DALI with occupancy sensors and building-management systems that monitor HVAC, electrical systems, and security networks, a skyscraper can become a dynamic participant in the urban environment—dropping a floor’s interior lights to zero the moment the last person leaves. As a result, electronic controls combined with LEDs can act like a dimmer switch for a city’s entire skyline.

White light blights the night

With all the possibilities from LED technology, why are our nights too often lit with harsh and clinical light, casting glare and creating shadows, disrupting human and ecological health, erasing the stars from our skies? The answer starts with the color of LEDs. At first glance, an LED streetlight looks like a collection of small white bulbs, but it’s not. To produce a light we perceive as white, most manufacturers coat a blue semiconductor core with a yellow phosphor material that absorbs a portion of that high-energy blue light. The problem is that this “white” light is still heavily blue, which is exactly the color no species has evolved to expect at night.

And because blue light is the second most energetic part of the visible spectrum (violet is the most), it doesn’t just illuminate our streets and invade our homes. Blue light also scatters in the atmosphere more easily than any other color, which helps to create the hazy, illuminated fog known as sky glow over every city of any size.

“Cooler” colored LEDs in the 4,000- to 6,500-kelvin range offer the most lumens at the lowest cost, so most early adopters installed these blue-rich white lights. The good news is that LED technology has continued to advance, and a growing number of communities are choosing warmer-colored streetlights that have less blue. (Phoenix, for example, converted 100,000 streetlamps to 2,700 K LEDs in 2020.) And, of course, light pollution isn’t just a result of LEDs. Older lighting technology also adds to the glare—bright white metal-halide lights, especially—and cities are loath to replace something that isn’t yet broken.

But our main failure isn’t a technical one. It’s that we have yet to revise our thinking about lighting at night. We use LED technology just as we did the old sources of light. As a result, we have largely offset the gains that were promised in terms of reducing energy consumption and carbon emissions by making light pollution worse, and have so far let an incredible opportunity go unrealized.

Can the City of Light do it right?

Across the Channel in Paris, the failure to realize the potential of LEDs feels even more palpable. Unlike London, which suffered heavily from German bombs, the lovely 19th-century Paris that Baron Haussmann created largely escaped destruction in World War II. To nearly 50 million annual tourists, the beautiful uniformity of the architecture is instantly recognizable. But the City of Light’s nocturnal atmosphere is also part of the draw, and extensive attention has been given to relighting its buildings and monuments. When I wander into the Cour Carrée in the Louvre, for example, I’m stunned by rows of amber LEDs that together create a warm glow along the palace facades. When I see the Eiffel Tower, first from a distance walking along the Seine and then up close, I find myself staring as I would at a campfire, the structure’s metalwork amber-lit with more than 336 high-pressure sodium bulbs.

Still, the city’s night lighting is far from perfect. With millions of residents, thousands of stores and restaurants, and 300,000 streetlights, the city overall is among the world’s brightest. Even at the base of the Tower, bright white LED lamps illuminate the African émigrés selling cheap berets and Eiffel Tower trinkets. And the city has been replacing its old sodium streetlights with new LEDs, swapping the warm yellow tones for which the city has long been known for bright white lamps no one wants to look at.

Street vendors display souvenirs at the foot of the Eiffel Tower. Their merchandise is lit by harsh white LED systems, which starkly contrast with the warm golden sodium-vapor light illuminating the tower above.

Luigi Avantaggiato

Nonetheless, the potential is here. In 2019, France introduced a nationwide law to reduce levels of light pollution, setting rules about both public lighting (preventing light from being projected above the horizontal) and private lighting such as stores, which are required to turn off their exterior and shop window lights after 1 a.m. In addition, an increasing number of French communities dim or turn off municipal lights after midnight to save energy and reduce carbon emissions. Although light pollution worldwide continues to increase by nearly 10 percent per year, France has managed to reduce its overall level. Chloé Beaudet, a researcher at Université Paris-Saclay, documented local light-reduction measures and found people generally agreed with the notion of dimming or turning off the lights, mainly for energy savings and ecological concerns.

“What I find is that people living in urban areas, they accept this kind of policy,” she tells me. “They’re like, okay, I don’t really use public space at night as a pedestrian, so what’s the point of having lights on?” For her, a key takeaway is that one lighting level does not fit all areas. “I think there is really a need for policy that is differentiated according to the neighborhood.”

In another positive development, the country has been minimizing artificial light to create ecological corridors designed to protect nocturnal species such as birds, bats, and insects. These corridors are connected and dark, mitigating the disruption to the 30 percent of vertebrates and more than 60 percent of invertebrates that are nocturnal. Even for city dwellers, this trame noire (“dark infrastructure”) helps to raise awareness of why controlling light pollution is important for life on Earth. Nationwide laws to control light pollution, the ability to light different parts of a city differently, dark corridors to protect biodiversity—these are exactly the kind of changes made possible with LEDs.

The iconic I.M. Pei Pyramid glows softly at the center of the Cour Napoléon at the Louvre Museum, its warm LED illumination flowing through the geometric glass-and-metal structure with a symmetrical framing of the surrounding historic pavilions against the night sky.

Luigi Avantaggiato

That’s not all. Almost until 1920, astronomers at the Paris Observatory were still gazing at the Milky Way. That’s impossible nowadays, but it could happen again. Despite the bright white LED streetlights now lining so many Paris streets, networks of LEDs using controls could lower lighting levels enough each night, so that the Milky Way could once again be visible over the French capital. And in the process, Paris could become the City of Light in ways that would set an example for other parts of the world.

New lighting demands new thinking

“I think we should aspire to have cities that see the stars,” Simon Thorp says when I mention this view of Paris. “You just need everything to be coordinated.”

Nearing the end of our London walk, having turned from the river and back up toward the Strand, Thorp brings me down narrow Carting Lane behind the Savoy Hotel, to where a gas fixture tops a thick lamppost, an original from 1870. A small plaque reads, “The last remaining sewer gas destructor lamp in the city of Westminster.” Thorp explains that the thick pole hides a tube that allowed methane from the sewers to get burned off at the mantle. “An early example of renewable energy,” he jokes.

The fire-orange flame is pleasing to the eye. But even here, on a narrow lane with no vehicle traffic, in a touristy area of the city, the flame is overwhelmed by a nearby, unshielded LED security light. Thorp shakes his head. “It’s stunning that someone could put in a light like that and think, ‘Great, nice job.’”

Here is the crux of contemporary artificial lighting at night. We know how to light well, and LEDs give us the ability to do so. But while our technology is 21st century, too often our thinking about light and darkness, safety and security, is stuck in the past. We could be doing so much more with this technology than we are. We could relight our nights in ways that would not only reduce energy and maintenance costs but also bring a slew of benefits, including healthier nights for humans, safer skies for nocturnal creatures, and a restoration of the stars.

In 2026, the tale of these two cities and their artificial light at night is that of a brilliant technology that we’ve engineered but haven’t yet learned to master. In short, we have yet to change the way we think about artificial light at night and to use it more thoughtfully and carefully—as we might, as one hopes we will.

This article appears in the August 2026 print issue as “The Wrong Way To Light a City.”

Like many engineers, Sarah Downs says she knew she wanted to pursue a STEM career from a young age. As a teenager, she discovered robotics through her Tulsa, Okla., middle school’s First Lego League team, and she fell in love with the field, she says. Downs participated in the international robotics program from 2014 to 2016.

Watching PBS specials on NASA Mars rovers Spirit and Opportunity, and seeing the live broadcast of the Curiosity rover launch in 2011, inspired the teen to dream of a career working with NASA.

Sarah Downs

MEMBER GRADE

Graduate student member

UNIVERSITY

Texas A&M University in College Station

MAJOR

Electric engineering

This year the IEEE graduate student member achieved that dream. For her final project as a master’s degree candidate in electrical engineering at the University of Tulsa, she worked on an algorithm in collaboration with NASA and the U.S. Air Force.

The algorithm she developed enables a robot assembling satellites in space to insert an antenna into the correct spot, addressing robotics’s classic peg-in-hole problem of inserting an object into its corresponding hole.

Now a Ph.D. student in electrical engineering at Texas A&M University in College Station, Downs is continuing her research on satellite assembly and manipulation “but on a much larger scale,” she says.

Following a childhood passion

Downs grew up in the Tulsa area. Her father, who died from a heart attack in 2015 when she was 13, was a safety advisor in the oil and gas industry. Her mother stayed home to take care of her brother, who has autism. After her father died, her mother went back to college to earn a bachelor’s degree in business so she could support the family.

“We didn’t have much income, and my mom was always worried about money,” Downs says. “That made me more aware of having a successful career, in a monetary sense.”

From then on, whenever she considered her future career, having a decent salary to support the family was high on her list.

By pursuing a career in robotics, she says, she can follow her passion while obtaining financial security.

In high school, Downs joined the First robotics club, where she found herself drawn to the electrical components used in the machines she and her classmates built.

During her final two years of high school, she participated in an extension program at Tulsa Tech, a training school. She spent half her day in high school classes and the other half taking engineering courses at the vocational school.

After graduating in 2020, she accepted scholarships to attend the University of Tulsa. She began her freshman year at UTulsa not knowing whether she wanted to major in electrical or mechanical engineering, she says, adding that her love of working with small systems helped her choose EE.

For her senior year capstone project, she and two of her classmates designed a lunar lander exhibit for the Tulsa Air and Space Museum. They created an interactive game that simulates missions on lunar and martian surfaces. Four celestial bodies—the moon, Venus, Mars, and Titan—are listed across three computer monitors. Using a game controller, museum visitors can explore the virtual surface of each one. The exhibit is still on display.

Downs earned her bachelor’s degree in electrical engineering in 2024 and continued her education at the university’s EE master’s degree program.

Both more and less complicated than people think

When Downs began her graduate studies, she was supposed to be part of a NASA robotics project for two years. But when a delay in government funding postponed the project’s start, she instead spent her first year in the school’s Institute for Robotics and Autonomy, then newly launched. Its main focus is developing robots to assist people who have mobility challenges.

Inspired by her grandmother, who was wheelchair-bound due to severe arthritis, Downs developed a robotic arm that helps older people and wheelchair users live independently. The arm was able to identify and place objects in the appropriate locations inside the home, such as unloading certain groceries from a shopping bag and placing them on a shelf or in separate containers.

Before the start of her sophomore year in 2025, the NASA project finally secured government funding. She developed a robot that achieves the peg-in-hole task without using any vision systems. Typically, cameras help guide robots’ satellite-assembly work. But in the harsh, remote environment of outer space, cameras might malfunction or encounter delays.

“Don’t stop asking questions. Especially in engineering, don’t pretend like you know everything, because science is about constantly wanting to learn and listen.”

Rather than using cameras, Downs’s robotic arm deploys a force-based insertion process to sense position and orientation of objects in the arm’s environment. The robot loosely grips an antenna and, with a torque sensor on its gripper, “feels” the force feedback of where the satellite and antenna are in relation to each other. The robot then guides the antenna assembly into a target opening on its satellite and maintains the position during adhesion.

Adding to the complexity, the robot performs its task in zero gravity.

“Without gravity, you now have to consider the arm’s reaction torques on the satellite to avoid flinging it into space,” Downs says. Any motion from the arm during the insertion process, especially from increased forces, could cause the satellite to continue movement in that direction.

To combat that, Downs is performing calculations for the project to direct targeted reverse thrusts and counter the force of the robot’s motions.

Her graduate project captures the simple yet complex nature of robotics that she finds fascinating, she says.

“I think robots are both more and also less complicated than people think,” she says. “Really, all you need to start programming a robot is its Denavit-Hartenberg parameters, and you can do a lot with that,” she says, referencing the four values used to describe the position and orientation of a robotic arm and manipulators. Even with different grippers and degrees of freedom, “fundamentally, all robot manipulators start there,” she says.

“But,” she adds, “we’re still learning so much about how robots interact with their environment. Even something simple to us, like manipulating a pen, is still incredibly complex for robots.”

Downs is completing her doctoral thesis in the Robotic Space Simulator project at Texas A&M’s Robotics and Automation Design (RAD) Lab, which specializes in developing machines that can survive in extreme environments. It collaborates with NASA.

Her thesis advisor is Robert Ambrose, a NASA veteran who launched the RAD Lab in 2022. The IEEE member is set to serve as associate director of the school’s Space Institute, due to open this year in Houston. The research facility is being built next to the Johnson Space Center.

After earning her Ph.D., Downs says, she hopes to one day work for NASA, developing rovers that collect samples from Mars or robotic arms that perform tasks on space stations.

To learn more about robots, check out IEEE Spectrum’s guide.

Getting out of the engineering bubble

Downs joined IEEE in 2020 as a freshman at UTulsa to get more involved in electrical engineering events on campus. At the time, the COVID-19 pandemic kept clubs and organizations from meeting in person.

She was active in her school’s IEEE student branch and was elected as its 2022–2024 president. Under her leadership, the branch went from having a few events to hosting one every two weeks.

They included lunch-and-learn sessions and dinners that connected students with professional engineers and the university’s alumni. Downs also organized hands-on workshops on soldering, 3D printing, CAD modeling, and résumé-building.

Her efforts helped increase the branch’s executive board membership from roughly five students to 25 in 2023. The same year, her soldering workshop attracted about 80 students.

She says she enjoyed working with IEEE, especially “engaging with alumni and learning from engineers.”

IEEE is a great resource for networking opportunities, she says, noting that “during the COVID-19 pandemic, engineering students stayed in their bubbles.” IEEE events helped the students make connections that could serve them well, she says.

“Networking is very important, especially in today’s tough job market,” she says. “It’s a lot about who you know and how people observe your work ethic.”

Downs, who now serves as an IEEE graduate advisor for UTulsa’s student branch, says she has seen firsthand how the school’s student branch network has benefited its student members.

“A lot of them have found jobs” because of IEEE, she says.

The working and networking of an engineer

As the IEEE graduate advisor for UTulsa’s student branch, Downs noticed that many engineering undergraduates finish college without any hands-on experience, whether it be a project or an internship.

“Their résumés are very sparse, and they have no proof of their technical skills,” she says. She herself completed a facilities engineering internship at Tulsa International Airport’s American Airlines maintenance facility after her sophomore year of college. And she was an electrical engineering intern at Flight Safety International outside Tulsa after her junior year and after she graduated. The company designs, builds, and maintains its own flight simulators.

Her advice to undergraduates is to hone and demonstrate both their hard and soft skills by working on research projects or even personal passion projects.

“A Raspberry Pi doesn’t cost that much, and you can start working with that immediately,” she says. Students also can take part in engineering interest groups and professional organizations at their school, she adds.

“Put yourself out there and join a research team,” she says. “It’s a great way to show people that you’re a good person to work with and you’d do a good job in the field.”

She adds that it’s also a fine way to keep learning—which is what drew her to a field that has developed only within the past century.

“We’re still constantly learning about robots,” she says.

“Don’t stop asking questions,” she advises students. “Especially in engineering, don’t pretend like you know everything, because science is about constantly wanting to learn and listen.”

In the eastern Indian Ocean, south of Java in the vast sea stretching toward Australia, a fishing vessel slightly alters its course while operating near the boundary of its authorized fishing ground. Nothing appears unusual on deck. Nets remain in the water. Engines maintain a steady speed. To the crew, it is an ordinary day at sea.

Yet hundreds of kilometers above, satellites continuously record the vessel’s position. At Indonesia’s Marine and Fisheries Resources Surveillance Station, in Cilacap, where I work, a monitoring platform receives the signal and automatically compares it against fishing permits, designated fishing grounds, vessel characteristics, and historical movement patterns. Within minutes, the system identifies a potential violation. Before any patrol vessel leaves port, before any inspector boards a vessel, and before any warning is issued, we have begun enforcement.

This transformation reflects a profound shift in maritime governance. The ocean has historically been opaque to regulators. States could only enforce laws where patrol vessels happened to be present. Today, however, integrated systems combining data from vessel monitoring systems (VMS), satellite remote sensing, geospatial analytics, and increasingly sophisticated data-processing tools are making marine activity visible at an unprecedented scale. Global Fishing Watch alone tracks hundreds of thousands of vessels worldwide, generating a near real-time picture of fishing activity across the world’s oceans.

Indonesia has emerged as one of the most ambitious examples of this transition. As the world’s largest archipelagic state, managing more than 6 million square kilometers of maritime space, Indonesia faces a challenge familiar to many coastal nations: There are never enough patrol vessels. Digital surveillance is a practical necessity that makes my job possible, even as it creates new challenges.

The Law of the Sea Meets Digital Reality

The international legal framework governing the oceans was designed in an era when maritime enforcement depended almost entirely on physical presence. The United Nations Convention on the Law of the Sea (UNCLOS), adopted in 1982, assumes that states exercise authority through patrols, inspections, vessel boardings, and direct observation.

For countries with extensive coastlines and limited enforcement resources, this model has always faced practical constraints. Indonesia’s Fisheries Management Areas (WPP-NRI) span waters ranging from the Indian Ocean to the Pacific and from the Strait of Malacca to the maritime boundaries adjacent to Australia and Papua New Guinea. Monitoring such a vast domain solely through patrol operations is both expensive and operationally impossible.

Beginning in the late 2010s, Indonesia accelerated the integration of satellite-based monitoring into fisheries enforcement. Vessel monitoring systems became a cornerstone of this strategy. By early 2026, a total of 9,394 Indonesian fishing vessels were actively transmitting through the national VMS, representing an increase of 2,880 vessels during the 2021–2025 period. As part of Indonesia’s broader maritime surveillance architecture, VMS data are complemented by satellite remote sensing and other monitoring tools to help identify suspicious activities involving vessels operating without active transponders or outside the national VMS network.

Indonesian fisheries officials plan fishery patrols using data from tracking devices, satellites, and their understanding of the patterns of illegal fishing.Indonesian Ministry of Marine Affairs and Fisheries

The implications extend far beyond vessel tracking. Continuous digital monitoring enables authorities to reconstruct vessel movements, identify suspicious behavioral patterns, detect unauthorized fishing activity, and verify compliance with licensing conditions. Rather than waiting to discover violations during patrol operations, regulators can increasingly prioritize inspections based on data-derived risk assessments.

Maritime governance is shifting from reactive enforcement toward predictive oversight.

The Surprising Geography of Digital Enforcement

The expansion of surveillance infrastructure has already generated measurable enforcement outcomes.

The Ministry of Marine and Fisheries Affairs Indonesia imposed 2,550 administrative sanctions during 2025, many involving violations detected through the vessel monitoring system, including fishing outside authorized fishing grounds and deliberate deactivation of monitoring transmitters.

This statistic is significant because many of these violations would have been extremely difficult to detect under traditional patrol-based enforcement. A vessel that briefly crosses into a prohibited fishing zone may never encounter an enforcement vessel. Likewise, a captain who temporarily disables a transmitter may escape detection if oversight depends solely on physical inspections.

Digital monitoring fundamentally changes this equation. Every vessel movement creates a data trail. Authorities can reconstruct routes, identify anomalous behavior, and compare activities against permit conditions long after the event itself has occurred.

The first quarter of 2026 demonstrates the scale of this surveillance capability. During just three months, Indonesia’s fisheries monitoring system tracked 14,571 fishing vessels, 182 fishing gear units, and 208 registered home ports while identifying 491 suspected violations across the country’s fisheries management areas. These violations included unauthorized fishing grounds, illegal high-seas operations, transshipment-related offenses, port-base discrepancies, licensing irregularities, and indications of poaching.

Such numbers reveal a fundamental transformation. Enforcement is no longer limited by the number of patrol vessels available at sea. Instead, surveillance capacity increasingly depends on the ability to collect, process, and interpret big data.

Illegal Operators Are Learning Too

Yet greater visibility does not eliminate illegal fishing. But it does change how poachers operate.

Indonesia’s expanding digital surveillance network, and a 2023 requirement that even small vessels use VMS when 12 nautical miles offshore, appears to have improved compliance among licensed fishing vessels. However, as enforcement capabilities become more sophisticated, some actors engaged in illegal fishing have also become more adept at exploiting technological and operational gaps.

Deliberately disabling VMS transmitters remains one of the most common enforcement concerns. While temporary signal losses, whether intentional or caused by technical failures—can complicate the reconstruction of vessel movements, they do not necessarily prevent authorities from detecting potentially illegal activity. Indonesia increasingly combines VMS with satellite-based observations, other maritime surveillance systems, intelligence-led analysis, and reports from community-based surveillance groups (Pokmaswas) to corroborate suspicious behavior and direct patrol resources where they are most needed. This layered approach—integrating digital technologies with local knowledge from coastal communities—helps reduce opportunities for illegal, unreported, and unregulated (IUU) fishing even when a single monitoring system is compromised.

A compromised surveillance network could potentially disrupt enforcement operations just as effectively as a vessel evading patrol detection.

As digital surveillance expands, one lesson from Indonesia’s experience is that stronger monitoring does not eliminate illegal fishing—it changes how illegal operators behave. Improved compliance across much of the fishing fleet has been accompanied by increasingly sophisticated attempts by a smaller group of offenders to avoid detection. This reflects a broader reality of technology-enabled enforcement: As monitoring capabilities evolve, so do the strategies used to circumvent them.

The result is a technological arms race. Every improvement in surveillance capability encourages new methods of avoidance, whether through disabling tracking devices, manipulating vessel identities, or exploiting gaps between different monitoring systems. Enforcement agencies must therefore continuously refine their analytical methods, integrate multiple sources of maritime information, and adapt their operational strategies to keep pace with evolving behavior at sea. Effective digital fisheries governance is not defined by a single technology but by the ability to combine data, human expertise, and operational intelligence into a resilient and adaptive enforcement system.

The Next Battle May Be Over Data Integrity

The future of fisheries enforcement may ultimately depend less on detecting vessels and more on ensuring confidence in the digital systems that generate enforcement decisions.

As surveillance networks become increasingly integrated, questions surrounding cybersecurity, algorithmic accountability, and data integrity become more important. What happens if vessel tracking data are manipulated? How should authorities verify automated risk assessments? What safeguards exist when enforcement actions increasingly originate from algorithmic analysis rather than direct human observation?

These questions are no longer theoretical.

Modern fisheries governance increasingly depends on interconnected networks of satellites, communication systems, databases, cloud infrastructure, and analytical platforms. While these technologies dramatically improve visibility, they also create new vulnerabilities. A compromised surveillance network could potentially disrupt enforcement operations just as effectively as a vessel evading patrol detection.

For Indonesia, this means that investment in digital surveillance must be accompanied by investment in digital resilience. The effectiveness of a monitoring system ultimately depends not only on the volume of data collected but also on the credibility, security, and reliability of the information produced.

Governing Oceans Through Data

Indonesia’s experience illustrates a broader global transformation in maritime governance. The ocean is becoming increasingly transparent to regulators. Activities that once occurred beyond the reach of enforcement agencies can now be observed, analyzed, and investigated through interconnected digital systems.

The benefits are substantial. Expanded VMS adoption, improved monitoring coverage, and thousands of administrative enforcement actions demonstrate that digital surveillance can significantly enhance fisheries governance. Yet the transition also introduces new challenges involving data quality, cybersecurity, algorithmic accountability, and adaptive criminal behavior.

The central question facing maritime regulators is how governments can ensure that increasingly powerful monitoring systems remain transparent, secure, and accountable while preserving public trust and legal legitimacy. The most important lesson may be that digital surveillance does not replace traditional enforcement. It changes where enforcement begins. For generations, maritime law enforcement started when a patrol vessel encountered a suspected violator. Today, it often starts when an algorithm detects a pattern.

That shift may prove as significant for ocean governance as the invention of radar was for maritime navigation.

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!

Before we get into this week’s article, I’d love to hear from you. If you have a question about your career or an upcoming decision that you want advice about, you can ask it here. I’ll be reading through your responses and picking questions to answer on a regular basis. Now back to our regularly scheduled program.

The Safest Career Move Is Often the Riskiest

Software engineers have some of the shortest tenures of any white-collar profession. The average software engineer stays at a company for roughly two years, about half as long as workers in most other knowledge professions. The layoffs of the past few years have certainly highlighted this instability, but it was already there.

This isn’t an essay about a broken job market though. Rather, it’s about how to turn that instability to your advantage, which is something I’ve spent the last decade doing on purpose.

Playing It Safe Was the Riskiest Option

I switched careers into software in my 30s. I had a stable job at a community college, complete with a union and a pension. It was about as secure as a career gets, and I learned to program on the side.

Then I did something nearly everyone in my life considered reckless: I quit, leaving the secure job to become a junior developer at 31. My own mother was skeptical. I took the riskier job anyway, for two reasons: It was the work I actually wanted, and I could see potential.

My first development job was at a grocery retailer. Good people and a company I liked. But I kept meeting engineers earning twice my salary for the same work. In the San Francisco Bay Area, surrounded by some of the best engineering talent in the world, I realized my skills were stagnating.

So I left for a small startup. I learned more in nine months than I had in the previous two years, and my salary doubled.

Over the years I’ve come to treat career risk as something to manage deliberately. It falls into two categories.

Take Risks With Your Job

The first type of risk involves the job itself: Bet on yourself by striving for better roles and opportunities.

Job-hopping for money alone isn’t wrong, especially early on. But the returns shrink after the first few hops, and the stress of chasing a slightly bigger paycheck every year will wear you down.

There’s another career risk with rewards that compound: Seeking positions to work alongside the strongest engineers.

You might struggle to keep up. You might even get laid off. But the skills you absorb working alongside people better than you are the ones that create durable stability. You build marketable expertise, you see how different organizations actually operate, and every project becomes another tool you carry to the next opportunity. Working next to stronger engineers is a proven way to increase your own expertise.

If that feels too big, try volunteering for a project you have no idea how to do. The risk is that you fail in front of people. The reward is a new skill and a resume line that opens the next door.

Compare that with the “safe” path.

You stay at one company, assuming loyalty will be rewarded. It usually isn’t. And when you finally leave, by choice or not, you may find the skills you built are worth little on the open market. You might be the in-house expert in an aging tech stack while employers are hiring for more cutting edge technologies. Suddenly you’re competing against people with half your experience.

You could be taking on a risk you didn’t notice.

Risk Your Time

The second form is risking your time, which means betting on trends.

Some trends are non-negotiable. If you’re a software engineer, then cloud services, ReactJS, and AI are mainstream enough that ignoring them actively damages your career. A backend engineer who refuses to learn cloud architecture is volunteering for obsolescence.

The real gamble is with the smaller trends: the niche tools you stumble onto and find quietly interesting, with no idea whether they’ll matter.

About two and a half years ago, I learned about retrieval-augmented generation (RAG). Almost no one in my circle was talking about vector databases, a central piece of RAG. Today RAG is close to mainstream, and for once, I had the early-adopter advantage.

Most of these bets don’t pay off. But when one turns into a major trend, you’re already on the ground floor. Right now I’m making the same bet on voice AI. It isn’t mainstream. It may never be. But if it becomes the next thing, I’m already there, building a foundation.

Short-Term Risk, Long-Term Stability

Counter-intuitively, job-hopping and betting on trends gave me the thing I was after the whole time: stability. I’ve rarely struggled to find work, because every risky move stacked skills the market actually wanted.

If you feel stable and comfortable right now, enjoy it. But ask yourself whether you’re still learning. Because if you’re not, the comfortable choice and the dangerous one may have converged.

The goal isn’t to avoid the open market forever. It’s to make sure that when you land on it, you’re not at its mercy.

By Brian Jenney

P.S. Don’t forget to submit questions about your career or an upcoming decision that you want advice about here!

—Brian

What It Means to Be a Mathematician When AI Does the Math

Until recently, human mathematicians have been central to creating new proofs, even when the work relies on massive computational resources. AI is now challenging that status quo. Writer Benjamin Skuse surveys the ongoing debate in the field about the role of AI, and the existential questions mathematicians have about their own careers. If AI mathematicians surpass human knowledge, could these researchers become “priests to oracles”?

Read more here.

Chip R&D Is Accelerating to Keep Pace with AI

A new partnership between UCLA and five major semiconductor companies is the latest program aiming to bridge the gap between industry and academia. The US $125 million university-industry hub is meant to strengthen collaboration and speed up the R&D process to help meet AI’s fast-paced hardware demands.

Read more here.

Why Mentorship Is the Most Underrated Leadership Skill

True mentorship is far more than friendly advice. This key leadership skill requires advocacy and honest feedback via lasting relationships, and it can strongly benefit both mentor and mentee. Parul Jain, a product management leader at Deloitte, shares what she learned from serving as a mentor—something she didn’t have for much of her own early career.

Read more here.

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.