This course is the one that I really need to be blogging most regularly given that it can go in unexpected directions. And now we have the snow day, so no class tonight. Maybe that will give me time to regroup and think about how I can best keep this up after spring break. Until then, here's a summary of what's been up.

A couple weeks ago, we focused on subsumption architecture with an emphasis on low-level behaviors. One idea I wanted to stress was both the power and limits of the level 0 behaviors. To that end, I had students play with BugWorks, a nice system from Chris Thompson (perhaps no longer in development) that, at its core, allows one to build Braitenberg vehicles. (As an aside, check out the picture of Braitenberg here. It's exactly what I expect him to look like.) While systems of interacting bugs display fascinating behavior (no surprise in a Braitenberg world), we would have trouble thinking of them as possessing anything more than rudimentary intelligence. Instead we find interesting system-environment interactions. The bugs respond in reasonable fashion to the sensory input, reasonable meaning that the reaction is readily understood in terms of the sensors and how they're connected to the bugs' motor skills. But any intelligence we attributed to the system is simply that: our attaching meaning to the bugs' sometimes complex actions.

That led us into the long Saturday class, half spent with Avida and the other with Lego Mindstorms. With Avida (we actually used the limited Avida-ED edition), the goal was to make plausible the idea that working (not necessarily efficient) algorithms could be evolved through random mutation combined with selection. The selection was based on rewarding creatures that could perform the indicated logical operation while still reproducing before being killed by another creature. The result was that creatures found efficient ways to perform the logical operations and more efficient ways to copy themselves. To my mind, the Avida-ED simulation doesn't have the richness of behaviors that Tierra has, perhaps because the creatures must have the same size. But that's made up for by the visualization tools, although my students weren't always happy with those (particularly trying to make out what was happening in the tiny pictures of the registers). A variety of interesting behaviors was seen, and their next task will be to analyze how that works for a complex creature. Much of our next Monday class was reviewing the Saturday work with Avida-ED in terms of the general concepts of artificial evolution.

And then on to the Lego robots. The main task was to build a robot that could simultaneously execute multiple behaviors, a task that the software makes pretty easy to do. It's only slightly harder to build subsumption techniques into the system by having communication between the behaviors--having a level 1 behavior suppress or inhibit a level 0 behavior. And that's what they'll do next.

With the Legos, I've observed that students almost always start building before they think about the software or the overall design. That's why we were going to spend tonight (and now will spend our class two weeks from tonight) on Pfeifer's and Scheier's design principles. I hope that for the final project, I can get everyone to think before they jump. Although anyone who has watched me program knows I don't follow that advice. 

So now there's a paper due, I meet with all the students individually over the next two weeks, and then back to design.

Well, not quite everything since I have some catching up to do. When I last blogged this class, we had just been working on Coulomb's law as an entry to the shell model for the atom. Since then, we've gotten all the way up to Ar!

The Moog-Farrell approach is as much one about developing skills in scientific explanation as it is building a reasonable model of atomic structure. By examining the trends in first ionization energy, a single piece of data for each atom, we can hypothesize the existence of the first three electron shells. The first shell holding a maximum of two and the second shell holding a maximum of eight falls out of this data cleanly. For me, the hard part is getting students who have seen this structure before to realize that they don't truly know it's reasonable until they have experimental supporting it. Too often, they confuse being told something with knowing that thing (and our students aren't the only ones who fall into that trap).

But we can quickly build a model that is consistent with the ionization energy data and just happens to be the standard model for filling electron shells. The model gives concentric shells that, through the introduction of core charge, provide good fidelity to the ionization energy trends of the first three periods. And fidelity to those energies for the main group elements if one is willing to accept that the core charge stays the same while the valence shell increases by one as we move down a main group. (It would probably be good to introduce a little bit of trends in reactivity to help that part of the model development slide by more easily.)

What's clever is that we can now grab an independent data set, the covalent and ionic radii, and show that the trends in those values are also consistent with our developing model. The model therefore has predictive as well as explanatory power.

I further like that the model is clearly not perfect. It doesn't explain well the anomalous behavior of IE in going from N to O, or from P to S. When we talk about spatial distributions of electrons, we can do a little better at explaining what's going on in the chalcogens, but that won't be a completely satisfactory explanation either. And that's what gives us the motivation to build better models.

One thing I'm learning in working with this model is how much I dislike the term "shielding" in the atomic structure context. A common student misconception is to think that the nucleus only has so much pulling power that must be divided among all the electrons in the atom. While that's consistent with some experimental data (larger size and smaller IE for anions compared to the neutral atom), it's of course inconsistent with Coulomb's law. But using a word like shielding reinforces the misconception. It implies that the valence electrons don't feel as much of the nuclear pull, and that's not true. In the model, the valence electrons feel the full pull of the nucleus, but they're also repelled by the core electrons, thereby offsetting the nuclear pull. Core charge then gives a rough guide to the balance for the valence electrons of nuclear attraction and core electron repulsion.

Another issue at this point is that students want to know how many electrons each shell has. So far, the data is consistent with a model that has two electrons in the first shell and eight electrons in the second shell. I've seen students then try to build electrons past the third shell by just going 2, 8, 8, 8, ... without recognizing that we have no data (yet!) that tells us how many electrons can go in the third shell.

And that brings us to transition metals. I'll put off discussing electromagnetic radiation; we only need that to have an understanding of photoelectron spectroscopy. That will get us to K, Ca and Sc, and then we have the rest of the periodic table.

But most of that will happen after break; test this week...

Because the test was upon us, we just started conditionals with simple if statements. Nothing spectacular, but it allowed Chris to play with modifying pixels in images based on their color, following some of the ideas in Chapter 6 of Guzdial and Ericson (GE).

What's of greater interest is what comes next. We've got some time to spend with if, if-else, if-else-if, but once we've got that down, then we've established almost all the foundations we need: variables, types, arrays, objects, classes, loops, conditionals. So what comes up next?

• GUIs: We believe that students needs to understand how they work behind the scenes. What are events and how are they handled? How can they build a GUI that uses those events effectively?
• Extensions to other classes: These concepts are not specific to images, so how can they be utilized with other objects. The GE work with sounds may help with that idea.
• Building one's own classes: We've currently been working with adding methods to a pre-existing class. What if we want to build a class from scratch?
• Building one's own application: See the GUI above.

Coming out of the first half of class, I now see how important it is for students to type in as many methods as possible to their MyPicture class, as GE suggest. But I also believe that it's important for them to comment those methods as well in order to express their understanding of the code. For the many students who can't yet read code and immediately understand it, the commenting exercises are tremendously valuable. But like so many seeming valuable exercises, you've got to put in the time.

A snow day and the conclusion of the second annual Guilford Undergraduate Symposium (GUS) makes this a good time to catch up on blogging. The symposium went extremely well. I don't think we had any last minute cancellations; from what I've heard, all the sessions were well-attended; and feedback was uniformly positive from faculty, students, trustees, and even a few parents who were on campus.

A trustee used the best verb: The students were empowered. Because they felt ownership of their work, they were confident and enthusiastic in presenting it. And I spoke with several students who made clear that this kind of independent, original work is a life-changing experience. The quality of the work was great across the board.

By February 2010, can we get more of this to happen? There's certainly work that students could have presented last Friday, but chose not to (perhaps because it was work-in-progress, as if there's any other kind). Can we build experiential learning even more into our curriculum? These are the experiences that students will build from and remember, particularly when it includes the opportunity to present to their peers.

Last year, I would have been happy with 25 presentations; the first symposium had twice that many. This year, I wanted even more than the 67 presentations we had last Friday. The potential is there for much more. Even without a collegewide experiential learning requirement, so much is being done in so many departments as part of the standard curricular requirements. I'm looking forward to GUS 2010.

Yes, I'm behind in blogging. That's a shame because we're in an interesting place in this class. We finished up nested loops this week with several image manipulation exercises, following Guzdial and Ericson: copying, copying with cropping, rotating and scaling (up and down). With the first test coming up next week, Chris and I had a conversation about transferability. Having worked with the image examples, how much are students able to transfer their understanding to objects that aren't images? My take right now is that students who follow our advice (comment on their own every example we give them, start the assignments early, come talk with us) will be able to do that, and other students struggle. That's not significantly different from what I see in other classes; students say "I understood when I (or the professor) did it in class," but if they don't do the work outside the class to solidify that understanding, things won't go well. That's not a particularly novel insight, but I do believe that for those students who do the work outside of class and take advantage of the help we want to give them, the multimedia focus will work much better.

Last semester while teaching Chem 111 at just about this point in the semester, I sat in on class session for both introductory biology (the molecular and cellular version) just as Tom Tucker was discussing the molecular structure of proteins and lipids and introductory physics just as Don Smith was discussing the discovery of the electron and the Rutherford experiments establishing the basics of atomic structure. In that week, I could see how the atomic models we develop avoid discussing key features that the physicists learn, but also that the biologists use only the chemical knowledge they need for making their points about living systems.

When Don Smith and I talked, he explained this well (and I'm about to paraphrase his words badly). It's the distinction between the what and the why. In chemistry, we need to know what the nucleus is and how it's structured; we don't need to know why the protons and neutrons stick together. But we also have to develop an explanation for why, for example, atoms come together to form single, double, and triple bonds with different bond strengths. The biologist then need to know about the different bond strengths, but not necessarily the reasons for why they're different. All of this can bleed together at the fuzzy boundaries between the disciplines, but I find Don's categorization a useful one.

All of that is a lead-in to what we did in Chem 111 yesterday and today: Begin developing a model of the atom following the very nice Moog-Farrell treatment. The idea will be very simple: What's a straightforward model of the atom that provides an understanding of ionization energy data (and, later, photoelectron spectra)? The answer is wonderfully simple, but the process of building the model is as important as the model itself.

And really all we've done so far is apply Coulomb's law to electron-electron and proton-electron interactions. The conclusion we want to draw is simple: We need to put energy in to pull electrons and protons away from each other; we get energy out (sometimes as motion of the electrons) when two electrons repel each other). That we designate the first as being a result of the proton-electron system having negative potential energy and the second as the electron-electron system having positive potential energy is simply a convention. (That convention being that the potential energy is zero when the particles are really far apart.)

The early stumbling point is that many students have only seen energy as a positive number. The idea that there can be negative energy is really confusing. We did an aside into kinetic and potential energy today to try to work through that, and I'm sure I'm going to have to repeat that. I also try to make the point that when we discuss small numbers, we mean numbers that are close to zero on the number line--small in magnitude. I want students to get beyond the conception that if numbers like 10⁹ are really big, a number like -10⁹ is really small. Both are really big numbers; they just have different signs.

If all that sticks, we'll be ready to define ionization energy on Tuesday and then start looking at the experimental data.

No new material today, just review. Watching students work through the concepts is a reminder that programming is another discipline where building conceptual models is important. Norman's Design of Everyday Things was important in my realization both of the need for these models and the difficulty of constructing them. Such models are necessary in both the disciplines I teach, and I haven't found a reliable way to help students construct them. I need to reread Norman.

We tried a new kind of exercise today, and I'll be interested in seeing how much the students gain from it and how much we learn about what the students know. Continuing with the theme of nested loops from last time, I had written a couple simple methods to draw a horizontal grayscale gradient and a horizontal gradient between any two colors. But the students didn't know in advance what the methods did. Their task, working in teams of two, was to run the methods and then comment the code.

As with everything we do in this course, the exercise took longer than expected. But I already learned a lot watching the students work on the code. I can see the uncertainties they have about what individual lines of code do. The reason for declaring object variables with null initialization for reuse later is eluding many of them (and may not really be necessary for them to understand at this point). I had to do a little math in the code to limit the color components to [0,255], and I'm pretty sure our students don't fully understand how that works. But I think they do see how the loops work, and that's really the main point. We'll know more once we read their comments.

Chris ran review session 1 this morning, and I'm doing the second one on Thursday morning. Chris said that he worked through the concepts outside of the image classes and methods that our book uses, and he thought that was helpful. I'm interested in seeing what happens in my review session; perhaps the concrete work with the images can get in the way of developing understanding of the abstract concepts. But maybe we just need to figure out how to apply the method better.

One thing that is true: Students are much more excited about the output of these programs than any of the text-based examples I used in the past. I think we can work from that.

Even with assignments such as working in NetLogo and analyzing the Game of Life, I'm sensing some restlessness during class because we're talking rather than doing. And while the six hour class on Feb 21 will be more doing, we still have some ways to go before we get there. Showing such strong examples as Karl Sims's creature evolution (embedded at Evolving Virtual Creatures: The Definitive Guide) or older (Cog) and newer (Domo) examples from Brooks's lab only goes so far. But we need to talk about goals before we start designing.

Pfeifer's and Scheier's overview of agents in Chapter 4 of their text stimulated more discussion than I expected, in large part because, as a student noted very early on, we need a working definition of autonomy. Pfeifer and Scheier focus their discussion on complete autonomous agents, and several in the class questioned the necessity of having such things and the ability to create them.

In particular, we discussed how much we (i.e. humans) lose their autonomy because of the necessity for productive interactions with the environment. One student noted early on that a fully isolated human was hardly autonomous because it would quickly break down. But too much reliance on the environment, including other agents, certainly puts a drag on one's independence. This discussion helped establish the point that autonomy is a continuum with the implication, not explicitly stated in class, that the designer has to decide what level of autonomy is appropriate for the desired goals. (We did an aside into Kelly at that point; while his book is titled Out of Control, he more often discusses what is the reasonable balance between control and a hands-off approach.)

The discussion of the particulars Pfeifer and Scheier say should be taken into account when designing (embodiment, self-sufficiency, adaptability, situatedness) took an interesting turn. I sensed a consensus that adaptability was likely the most important of these. Embodiment is necessary, but almost trivially so if you're working in a real world. Self-sufficiency and situatedness are also important, but become so as a means to being able to adapt. What appears to be most compelling is the ability of an agent to make do with its environment, no matter how that environment changes. If it can roll over everything in its way, then it must be doing pretty good. Is it intelligent? Maybe, but at least it's successful.

That led to the last bit about universality. Agents can and should be adaptive, but we can never design them to be adaptive to every possible circumstance they may encounter. We just want to make sure that we can take account of things that might reasonably be expected to happen. Cut down on the design and programming; increase the likelihood of success in the agent's niche.

Next week, we delve more into Brooks, start designing something, and look more into artificial evolution in the Kelly, Pfeifer and Scheier, and Avida perspectives.

Demo rule #1: Make sure your graduated cylinders do not have holes in the bottom before you pour liquid into them. Unfortunately, I'm pretty sure that's the main message everyone took away as green liquid started pouring out of the bottom of the cylinder. Since there was no glass anywhere, the cylinder must have been broken previously and put back into its plastic base. Still, it's pretty impressive that someone was able to break the bottom of a 100 mL graduated cylinder that cleanly.

The real subject of the day was molarity. We started with a qualitative discussion after wiping up the green liquid. Two different volumes of solution take from the same container have the same concentration. As one student said, the ratio of green dye molecules to water molecules is the same in each case--they came out of the same bottle!

But when I added 25 mL of water to 25 mL of the original solution, that ratio clearly changed. The solution becomes less concentrated in the green dye molecules. The qualitative understanding of concentration isn't that bad.

Where we get hung up is making it quantitative. We want to use concentration as another means of counting molecules. This activity does a nice job introducing the concept of molarity and helping to develop an understanding of having concentrations of both ionic molecules and component ions in a solution. Just about everyone seems to be getting that 0.50 M NaCl(aq) is less concentrated in Na⁺ ions than 0.30 M Na₂SO₄(aq).

Where we'll have problems is that two key concepts, dilution and limiting reactant problems involving molarity, are only in the exercises. Without discussing those in detail, we lose track of the importance of molarity in counting molecules. Molarity:volume::molecular weight:mass. If I can get that point across, we have all the stoichiometry we need for the rest of the semester.

Thursday: I try to get that point across. And we start building atoms.