Dr. Peter Andrew Jamieson — Academic in an information evolution

← Back to Meanderings, Opinions, and Articles

Get Off My Lawn: Is That Our Real Problem with AI?

September 2026 · written with the help of Claude (Anthropic)

There are two things that I think AI presents us. (By AI I mean what we call Artificial Intelligence, though my preference is Algorithmic Intelligence, of the GenAI sort.)

First, my framing. I teach undergraduate engineers in higher ed, and I wonder what changes for them (let alone for us). In that exploration, one of my major goals, and the way I think I can help understand this, is to use the technology and see how it changes how I work. So I’ve done a deep dive. Much of what I’ve built I call dark factory building: the basic idea that I’m no longer writing the code. Instead, I’m creating software, web pages, hardware (I’m mAIking), and so on, more from the imagination than from the low-level implementation.

It seems we’re all upset with AI, old and young alike, except for a small portion of us. Is that because we really like our lawns, and we’re worried that AI is stepping on them? I think the fear looks different depending on where you stand. Get off my lawn, or my future lawn!

The old, those of us who have been doing things for a while, are upset because the automation is now on our lawn. We spent years cultivating our craft, and now someone can produce something that looks almost as good, cheaper and quicker.

The young are upset too, but for a different reason. They worry they won’t get a lawn at all. The path we took, the years of practice, the entry-level work, the slow building of relationships, seems to be getting automated, and the pipeline might be drying up. How do I get my lawn? And even when the tools (really not tools but agents, as Yuval Noah Harari reminds us) let them make something cheaply, they’re left wondering whether it counts, whether anyone will value it, and whether they actually know how to do the thing they just made.

So everyone is upset, just about different sides of the same fence.

And the small portion who aren’t upset? They seem to be the superusers, the people who have embraced the fact that they can now do more, and who have the semantic knowledge to do more. I think the answer to both fears is the same, and it comes down to that semantic knowledge and bridging it to the up-skill, or the superpower.

The automation is on our old lawn

I don’t think that we, as white-collar workers, really do that much, even though we say we are busy. Most of us claim we are doing a lot, and we might be doing a lot, but much of that might just be busy work. That observation comes from my own experiments, and from trying to understand ideas like those Professor Chad Jones lays out in “AI and Our Economic Future.” What is work? If capitalism takes labor beyond each individual’s own needs and puts that extra labor into the capital machine to create services and products, it’s amazing how much of that machine is bureaucratic moving and reskinning of information: the busy work.

Previously, when a disruptive technology started eliminating other types of labor, we white-collar workers (let’s call us the academics, or the perceived thinkers) responded simply: this is a moment where you need to skill up. Now automation has entered our lawn, the space of “thinking.” And we, the white collar, have a difficult time with it, because now the automation is on our lawn too. Where’s the next up-skill? It’s a fascinating moment. Period.

The gates are opening, and that’s the problem, for the young

For a young person who wants to enter the world and start creating in our spaces, this should be very exciting. This technology allows a product to be created that might not have been accessible before without a significant amount of time and money spent to build the artifact: music, a movie, a book. The book is probably the cheapest of these, but creating any one of them still costs a great deal in developing your skills, getting better, investing the time, and learning the craft. In many of those spaces you previously couldn’t do it by yourself. You had to spend significant time learning the craft, developing relationships, finding funding, and so on. As I argued in my last post, the culmination of that process is developing judgement. (Oddly enough, in that post the spelling “judgment” was the tell that I didn’t write the document alone. I’m more Canadian: judgement.)

Now many of these ideas, these artifacts, can be created with the help of AI. That is a disruption for the people who have gone through the process and are now part of an industry, and the pipeline into those industries is getting just as blurry for the people coming up and trying to get in.

Again, it’s “stay off my lawn,” right? I’ve spent many hours on my lawn. My lawn looks nice. But now you can go make your lawn cheaply, it looks almost as good as mine, and you can release it relatively quickly. For those of us who tended our lawns the long way, that is very, very concerning. For those trying to get lawns, the old, established paths to a lawn don’t seem to be simple, transactional paths anymore.

But the young aren’t simply celebrating. If the tool can produce the artifact, what is the young person’s work? The old at least have their years of practice to fall back on, and they have the semantic knowledge and pretty good judgement (the path to the superpower). The young are being handed the tool (again, not really a tool; it’s an agent) before they’ve built the knowledge that tells them whether its output is any good, and they can sense that there’s no guidance on how to build it.

From automation to intelligence

I first came across this idea listening to Andrej Karpathy’s conversation with Dwarkesh Patel, “We’re summoning ghosts, not building animals.” If we look at the progression of automation from a GDP standpoint, it seems inevitable that intelligence would eventually be next in line. Since the Industrial Revolution, we’ve constantly been automating. And now part of that automation is “intelligence.” It’s a Pandora’s box that I don’t think we can close. So assuming it’s here, how does it change things? How can we, either the old or the young, accept it and build a new lawn?

In my case, I come from a world of using language to design. When something can be designed with language, for example software programming or designing hardware chips, we usually use language because it lets us accelerate the design, capture hierarchy, repeat component processes, and build complexity from simplicity. Well, GenAI, trained on our plethora of documented designs, is excellent in this language space, to the point where, in my exploration, the dark factory, as described by Dan Shapiro in “The Five Levels: from Spicy Autocomplete to the Dark Factory,” really does change the game. (Nate B Jones gives a good walkthrough of Shapiro’s levels in “The 5 Levels of AI Coding (Why Most of You Won’t Make It Past Level 2).”) I’ve been exploring what alumni are doing at places like Microsoft, NVIDIA, Apple, and Google, and listening to The Pragmatic Engineer, and I keep hearing professionals in the software space describe a moment, around December 2025, when everyone who worked in this space realized the game had changed, specifically for software.

I’ve had a number of alumni visit recently. When I asked them how it’s changing what they do at work, one of them told me he used to be a product manager, and now he’s a developer again, a builder again, and he’s loving it. So the question is, how do you get someone to develop that excitement about using the automated tool to help them build things? I really do think it all comes down to this: these tools allow us to build things faster. And how does that happen? I can see that my former students have built up, in their workplaces, a set of semantic knowledge that gives them the superpower.

Semantic knowledge as the path to the superpower

The best answer I’ve come up with is semantic knowledge. I’m borrowing the term from cognitive psychology (or whatever name that field wants to go by) and from Yaman, Tian, and Lindström’s paper “Semantic knowledge guides innovation and drives cultural evolution” (PNAS, 2026). They describe semantic knowledge as the associations that link concepts to their properties and functions, and they find it guides innovation and drives how our culture develops. In other words, our brains build semantic knowledge of how things work and how they’re put together. You have to develop it. You can’t just say, I would like to build an app, or I would like to build a microchip. Something has to happen where you see that process, and you can’t just see it and say, I understand how it’s done. You actually have to engage and do it. I’m going to call that doing the first-principles exercises of education.

The second-principles exercises, then, use our semantic knowledge, judgement, and system design skills to imagine a good thing. With that imagined good thing in hand, we ask: How does it work? What is its structure? How will it be tested? In what order should it be built? Then, through the process of making it with the automated agent, we make the thing. Those of us who have deep semantic knowledge, the kind that comes from making things from first principles and practicing, can then use agents as a superpower to create.

When I first started building this way, I felt powerful and told others it was “like being a god.” But the powers were neither omniscience nor omnipotence. So I realized it’s more like having a superpower. Our ECE and CSE faculty at Miami quickly engaged in benchmarking our own courses, asking how GenAI would do in them (ECE results, CSE results, and our prompting methodology and taxonomy). No surprise: it did well.

The question that remains, and I will argue it’s the crux of all this: how do I give this superpower to my learners?

The first-principles exercise is how we’ve always done it in education. At the undergraduate level (K–12 too), we aren’t at the edge of human knowledge in a given space, so we usually know the answer. My colleague in math, Louis DeBiasio, captures this premise by asking which century a student gets to in undergraduate mathematics, which is a wonderful framing of the exercises. What we all tend to provide is exercises for the learner to take the concepts, apply them, and get to the answer. The answer was always in the back of the book, or the professor knew how to do it, and what textbooks gave us was a structure for making that happen for an individual. That’s the first-principles exercise. And the notion that a teacher causes that transformation (the BEST teachers can do this!) is a myth. The reality is that the learner is given a scaffolded space and time, so that their semantic knowledge can make the next addition to their brain.

The weird thing now is that GenAI can pull that curtain back and show us the entire experience, summarized, in a form that can be read and psychologically felt as “I learned.” If I don’t know how to do something, let’s say a physics question, I can pose it to an LLM, and with its framework and tools it can generate an answer. I call this pulling back the curtain, because now, as learners, we can see all the steps of the solution to a problem, whether that be mathematical, scientific, or essayed (we’ll call that the logical written treatise). And the weird thing is, as a human, when I read the steps, I think, oh, I understand that concept. I can see all the steps. They’re logical. Maybe there’s an error somewhere, but there are always errors, whether a human or GenAI does the work. And in that reading, I get this feeling that I’ve learned it.

It’s the same with bad study skills. I highlight the text, and it feels like I read and understood the ideas. Look, it’s highlighted! Brain releases feel-good chemicals. Human is happy they learned. Versus: try to solve the problem. I can’t solve it. Try it again later. I still can’t solve it. Brain feels bad. No reward. Painful. The learning doesn’t feel like it’s working.

So there’s a real temptation as a learner, because learning is hard (and painful), not to do the first-principles exercise. Instead, we can take the second-principles approach: we read the solution, and then we think we’ve learned. And because of that cognitive offloading, a term that came into my consciousness through Shaw and Nave’s paper “Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender” (2026), our semantic knowledge doesn’t develop.

The key questions that emerge on our road to superpowers

Three key questions follow from this:

The crux of that last question is knowing when to work from first principles and when to work from second principles, with the AI’s help. And when is the first-principles approach good enough?

I’ve faced these same questions in math. I remember listening to a TED talk by Conrad Wolfram, “Teaching kids real math with computers,” in which he describes mathematics as four steps. Step one: pose the right question. You look at the constraints and understand what you’re trying to solve. Step two: turn that real-world problem into a mathematical formulation. Step three: compute. How do we solve the equations, whether analytically or numerically? And step four: take the solution back into the real world and ask, does it make sense?

I know that isn’t mathematics. Mathematics at the high level is about taking a set of rules, and the logic on top of them, and proving or disproving certain things. But in the space of mathematical application, which I’ll call engineering, we use math and its techniques to model the real universe. So the question becomes: how do I learn which mathematical techniques can model the real world?

And that’s relatively easy to see. That’s not to say these math courses aren’t hard for an 18- or 19-year-old brain (let alone younger or older brains), but if you can sit in the forest of learning, you start to see Newton’s physical mechanics and its relation to calculus, and then you understand how one can model motion in our world. The relationship is nice. Then, usually in Calc 2, we start seeing other ideas, and in Calc 3 additional variables, so that we can model change in a broader sense.

But as a student learning these ideas, you sit so close to the trees, as I say. Our educational game finds that fuzzy nuts are hard to count, so we ask for simple nuts to count, and those are really Wolfram’s third step. You’re just doing the computation.

I might even phrase it this way. When I was doing Calc 1, Calc 2, Calc 3, differential equations, and algebra (plus what I called the Magic Show of complex numbers), I didn’t really understand what I was doing. I was following along as best I could, knowing there was a test coming that would check whether I understood what I’d been told, written down, and read. I would look at the problems and try to do them, and in most cases I couldn’t. So instead I would look for completely solved problems and reverse engineer how the computation was done to get the correct answer as shown. There were excellent resources out there that helped me, like the Schaum’s Outline series, or I would go to the library and find other textbooks with more examples. But I didn’t really understand the what or the why of what I was doing. For steps one and two, I was pattern matching the problem to the approach. For step three, I was algorithmically executing the calculation based on that pattern match. And for step four, I didn’t really care whether the answer was correct or made sense in the real world.

I did this throughout my engineering education, until my intellectual maturity grew (thank you to my mentors for surviving an idiot) and I started asking questions. Okay, what is this derivative? What is this integral? This Fourier guy had some pretty cool tricks. This transistor is really small. What, you can put them together and build a physical algorithm thing?

I feel like the general answer in the learning of mathematics is that you just need to be in the space, playing. The more computation you do, which is the part the actual computer can do, the more likely that at some point you get an emergence of number sense. That might be true, and it might not be, because I’m a single narrative sample point of that learning. So we might call computation, step three of mathematical learning, fundamental, in that you have to do it to develop number sense. That seems strange. And now we’re probably asking whether that is the first-principles exercise, when we should have spent more time on steps one, two, and four. How do you create the problem? What is the problem for? How do we translate it into math? What are we trying to understand here? And similarly on step four: Does it make sense, especially when we get back into this real universe? Are the answers grounded in truth? Can I estimate? Can I look at something and say, yes, that probably makes sense?

With practice of first-principles exercises, there is a moment (I don’t know when) that the superpower emerges: the transformation. Unfortunately, there’s no radioactive spider that bites you, and the transformation isn’t a discrete point in time. Which leads me to the call to question…

Come on, lawn owners (higher ed): invite me to your cool new park, or build one for us to visit

So it really comes down to this question: how do we do the modern first-principles exercises, as we always have, and how many times should we do them so that we develop the semantic knowledge to understand how to build the new artifacts (the systems), and to teach and learn the superpower of second-principles creation? I don’t have an answer. I actually think it’s a really good science experiment (Scrooge dollar signs should now appear in administrators’ eyeballs) that someone should design and try to solve. (Why not me? I’m happy to help, but I’m driving enough buses right now and lack the dressing up or clout to achieve such noble, serious work. And maybe the nuts are too fuzzy to count.) It’s one of those social science experiments that someone like me has the capability to design, but I have neither the trust of funders nor the capacity to run it alone. But I will argue it is higher ed’s most important question, for our aspiring lawn owners, the undergraduates, and our current lawn owners, the elite, to answer.

References

This post was written by me, with the help of Claude (Anthropic) for editing and for tracking down the citations. I’m not sure I’m writing since I’m dictating to it?

Peter Jamieson is an Associate Professor of Electrical and Computer Engineering at Miami University. More of his work, publications, and projects are at drpeterjamieson.com.