🎧🍌 The Future of AI, Science, and Learning | Karthik Duraisamy, University of Michigan
Inside Michigan's new partnership with the OpenClaw Foundation, how AI has transformed scientific research, why some academics think AI is a fad, and Karthik's code red for students
Karthik Duraisamy is one of the most highly regarded professors at the University of Michigan. He is a prolific scientific researcher and his classes are legendary among Michigan’s engineering students.
Karthik co-leads the University of Michigan’s newly created Institute of Agentic Computing. This is Karthik’s first public conversation on the new institute, which will serve researchers and developers building and applying agentic AI to advance scientific discovery, engineering, and beyond. It will also address broader issues of responsible development and governance, and serve as a central node for managing developers and maintainers of the OpenClaw platform.
We talk about how AI is transforming scientific research, two exciting scientific discoveries made with AI that were demoed at ClawCon in Ann Arbor on April 16th, how universities actually work, why some academics think AI is just a fad, how AI has changed education, and the code red advice Karthik gave his students a few weeks ago.
This is the first time I’ve had an academic on The Peel. Let me know what you think and if I should have more on the show!
Support this Episode’s Sponsors
Numeral: The end-to-end platform for sales tax and compliance.
Flex: Sign-up for Flex Elite with code TURNER, get $1,000 here.
Amplitude: AI analytics. All you have to do is ask.
Inquire about sponsoring future episodes here.
Timestamps to jump in:
0:25 The Institute for Agentic Computing
4:27 OpenClaw Foundation and Lobster Compute Company
8:19 How Universities actually work
12:33 ClawCon in Ann Arbor
15:24 Two scientific discoveries made with ScienceClaw
20:06 How AI is speeding up scientific discovery
25:42 Supporting AI and OpenClaw development
29:55 Why universities function like VC funds
34:29 How universities get money from the government
40:55 Why some academics believe AI is a fad
46:17 Biggest bottlenecks in AI today
49:26 How AI will change the world
53:10 Karthik's Code Red advice for students
59:19 Separating learning and doing
1:03:10 Ways COVID and AI impacted college students
1:14:53 How the role of universities is changing
1:23:21 Why college classes suffered from grade inflation
1:26:05 How AI is actually impacting the job market
1:32:49 Karthik’s advice for students
1:39:16 Winning two NCAA basketball national championships
1:43:04 Almost dying in the Grand Teton National Park
Referenced:
Karthik’s bio at UMich
ASU iPhone video
Find Karthik and LinkedIn
👉 Stream on YouTube, Spotify, and Apple
Transcript
Find transcripts of all prior episodes here.
Turner Novak:
Karthik, welcome to the show.
Karthik Duraisamy:
Happy to be here.
Turner Novak:
Yeah, thanks for coming on. I think it’s going to be fun. You guys just announced a bunch of things at the University of Michigan. We’re going to talk about that and then also talk a little bit about how AI is changing research, education, finding a job, the economy. But to kick things off, what did you guys just announce at the university?
Karthik Duraisamy:
So first of all, I think it’s an incredibly exciting time to be alive given everything that’s happening, not just around AI, but in science and research more broadly. At ClawCon, we announced the Institute for Agentic Computing, which is a partnership between the OpenClaw Foundation and the University of Michigan. The goal is to develop responsibly powerful agentic frameworks that people can use for a wide range of things. There’ll be a core team developing agentic infrastructure, and then a large number of people working with those developers to apply those frameworks across many different fields. Any field humans have ever touched, I think, will be agentified.
Turner Novak:
And how did this come about? The creation and the thinking around starting this?
Karthik Duraisamy:
Yeah. So first of all, OpenClaw is one of the most popular software frameworks out there right now.
Turner Novak:
And what is OpenClaw for someone who doesn’t know?
Karthik Duraisamy:
So Peter Steinberger and his team introduced OpenClaw in November. At its core it reduces friction, it basically gives you a very strong personal AI assistant that you can use to automate a wide range of tasks. A lot of the initial use was people running it on their laptops to automate manual things: emails, communications, files, that kind of stuff. But that’s scratching the surface.
Then people started using it as a social agent. I use OpenClaw to train my own agent, and now my agent can talk to your agent. They start doing interesting things, some of which we control and some of which the agents do on their own. The way I think about it: you had chatbots, which came on the scene about three years ago. Over the last year they’ve become extremely powerful. But those are just giving you ideas. You type something, it gives you an idea, you use it, you go do something.
Turner Novak:
It’s like better Google, sort of.
Karthik Duraisamy:
Yeah. Much better Google that can also give you cognitively powerful things, put ideas together. Those are chatbots. And then you have agents that act. It’s not passive, not just giving you information. Agents act upon information.
Turner Novak:
It’s like saying, “Hey, go order me a pizza.” And it goes and calls Domino’s on the website, places your order, and you get a pizza.
Karthik Duraisamy:
Exactly. Maybe it’ll deliver the pizza too. Soon.
Turner Novak:
The robots. Yeah, the autonomous robots.
Karthik Duraisamy:
Yeah. And that’s another thing, agents are not restricted to software. They act in the physical world too. People are already controlling their own personal robots with OpenClaw.
Turner Novak:
Oh, really? I didn’t know that was happening already.
Karthik Duraisamy:
Yeah. So anyway, if you think about the progression: personal chatbots, then personal agents, and I think the next evolution is social. The social and economic infrastructure around which our society is organized, all of that may now be agentified. I could have a few agents of my own that encode some of my skills and expertise. You could have a few of your own. They can talk to each other, collaborate, and it can be very decentralized. If you want one sentence: OpenClaw is like an operating system for the agentic world.
Turner Novak:
And OpenClaw is open source, correct? And there’s a foundation attached to it.
Karthik Duraisamy:
Correct.
Turner Novak:
How does this all relate to the institute and the university?
Karthik Duraisamy:
OpenClaw is an open source project and that’s really why it caught fire. There are probably more than three million users right now, and it happened in almost no time. The OpenClaw Foundation was formed to make sure OpenClaw stays open source. And many of the core developers, like Peter Steinberger’s team, will be in the institute as well. Think of that as the core layer, and then there’s a layer around them: people at the University of Michigan, collaborators, pretty much anyone around the world who’s interested in taking these frameworks and adapting them to their specific domains.
Turner Novak:
And I know we got connected through Dan in the investment office at the University of Michigan. They’re involved in this a little bit, they funded something, I think they funded a company that’s sort of a third party also helping. Can you explain what’s going on there?
Karthik Duraisamy:
Right. So the University of Michigan’s fund is heavily involved in the OpenClaw Foundation. And they also formed a new company called the Lobster Compute Company.
Turner Novak:
Lobster Compute. That’s awesome.
Karthik Duraisamy:
The OpenClaw Foundation is purely doing open source, basically supporting the developers. And Lobster Compute is the investment wing. That’s where money goes in for startups and other things.
Turner Novak:
It might be interesting then for people to understand how universities work. I had no idea this is how it works until we talked the other day. Can you explain the whole setup, university, institutes, departments, all of this?
Karthik Duraisamy:
Yeah. Even without the Institute for Agentic Computing, universities are fascinatingly organized places because they do many different things.
Turner Novak:
Yeah. It’s not just teaching classes.
Karthik Duraisamy:
It’s not just teaching classes.
Turner Novak:
It’s like a small percentage.
Karthik Duraisamy:
Yeah. Advancing research, innovation, startups.
Turner Novak:
You’ve got sports teams.
Karthik Duraisamy:
Sports teams. Hospitals, especially at Michigan. These are very complex organizations. But if you want to break it down, the core of the university lives in departments. The department of physics, aerospace engineering, sociology. That’s where students get admitted, take courses, get their degrees, mainly from a teaching perspective but also research. And research is not siloed. If you’re in aerospace engineering, you don’t just work on aerospace engineering. A lot of the interesting problems come when areas intersect.
Turner Novak:
So it could be the intersection of ethics in aerospace or something like that.
Karthik Duraisamy:
In fact, we have a research area in space ethics.
Turner Novak:
Really?
Karthik Duraisamy:
So you’re not far off. The engineering school and the business school have a joint program, for instance. Think of departments as the low-level units where faculty are hired, tenure is given, students are educated. On top of departments we have institutes. Institutes bring together different departments, different researchers, different students. The institute I direct is called MICDE, Michigan Institute for Computational Discovery and Engineering. We have faculty and students from around 40 different departments, all exploring different aspects of computing for science. Institutes sit on top of departments and bring people together to do interdisciplinary research.
Turner Novak:
So how many departments and institutes are there at the University of Michigan?
Karthik Duraisamy:
That’s a hard question. I don’t think anyone knows.
Turner Novak:
Okay.
Karthik Duraisamy:
I’m just kidding. The rough number is about 200 departments. And one of the amazing things about Michigan is pretty much all 200 would be in the top 10 of any ranking you can imagine. It covers all areas of human activity. Institutes would number in the dozens. There’s the institute I direct, the Institute for Agentic Computing that we just announced, the Institute for Firearm Safety, the Institute for Social Research, which is the largest social science organization in the entire world.
Turner Novak:
Yeah. My mother-in-law actually worked there for probably about a decade.
Karthik Duraisamy:
Wonderful. So institutes of various sizes and scopes bring together faculty from many disciplines to go after some grand challenge problems.
Turner Novak:
So you just announced some research at this thing called ClawCon, by the time someone’s listening to this it’s already happened. What is ClawCon for someone who’s never heard of it?
Karthik Duraisamy:
Think of it as a gathering place with a spectrum of people, core developers and heavy users of OpenClaw, but also people who are just genuinely curious about what’s happening. Typical tech meetups cater to a very specific audience. ClawCon is more democratic. It doesn’t distinguish between an ordinary person interested in AI and somebody who’s a big developer. You have powerful keynote demos that show how these things can change science. You also have very basic things like, what is OpenClaw? How do I install it? It runs the whole gamut.
Turner Novak:
So it’s a meetup conference kind of thing, a bunch of enthusiasts and experts and heavy users coming together to spend time, do some demos, do some presentations.
Karthik Duraisamy:
Correct. Mainly focused on getting people to meet, getting people to talk, showing some demos, maybe sparking some ideas.
Turner Novak:
So you just announced some new research. What did you announce?
Karthik Duraisamy:
Yeah, this is pretty amazing. Some of our colleagues at MIT and our group have been working together on scientific discovery using new AI agents. The MIT collaborators developed something called ScienceClaw, the big question it’s trying to answer is: how can science change when you’re combining humans, agents, and powerful tools? And all of it runs completely decentralized. It’s the next generation of collaborative science.
We showed a couple of examples at ClawCon. The first one: what do cricket wings, baroque choral music, and composite materials have in common? We had agents in biology, agents that understand material properties, and agents that understand music, all collaborating in a decentralized way, finding a part of design space that was unexplored and ending up discovering some incredibly useful material resonators.
Think of those agents as experts who know a lot about their own topic but not about the synergies between topics. When you bring them together, interesting things happen.
Turner Novak:
What could you use it for? What’s something someone might actually make with this?
Karthik Duraisamy:
Think about resonance. If you have a material excited by a disturbance, you want it oscillating at a certain frequency. The operational use is very broad. It’s not like this was discovered today and I’m going to use it tomorrow. But if you can now design material resonators for any property you want, and agents are finding the process to achieve that design, that’s a real step forward.
The second application we showed is superconductivity, which is more straightforward to understand. If you can pass energy across a medium with zero loss, no dissipation, that’s a superconductor. The problem is pretty much all known superconductors only work at extremely low temperatures, close to absolute zero. Minus 100, minus 200 degrees. Just not practical for most applications. We’re using agents to search for superconductors that work at much higher temperatures, room temperature, ideally. I don’t want to give the impression that we ran the agent and won a Nobel Prize. This is the first step in a long chain. Next, you have to build the material, test it, get information, come back. But this first step is actually the hard one, because the possible space of superconducting materials and configurations is enormous. Without these newer techniques, searching that space would take a very, very long time.
Turner Novak:
So it might be interesting to talk about the technique, but even before that, the old technique. How would you discover a new hypothesis in science? And how is that changing with AI?
Karthik Duraisamy:
Let’s break down the scientific process. You make observations in nature. You write down a theoretical model that explains the observation. Then you manipulate that model to get the property you want. These are fairly simple models at first, still the mental map.
Turner Novak:
This is the scientific method we probably learned about in high school.
Karthik Duraisamy:
Exactly. Theory, observations, experiments. Then you go to more detailed models, the kind that a lot of people in the institute I direct actually run. Very detailed models of a particular physical process. You gain more insight, you run an optimization. But computation is different from reality, so then you go build whatever you’re designing, do the experiments, take measurements, and iterate. All of these steps still have to be followed in the age of AI.
Until recently, all of these steps were done in sequence by different people. A theoretician might take years on the first part. Then a specialist in computation runs the models and optimizations. Then a specialist in measurement handles the experiments. Then you put it together and maybe you have an outcome.
Turner Novak:
Are you waiting for other people to be done with their phase while you’re working on multiple things at a time?
Karthik Duraisamy:
Generally, yes, it was sequential. Even before AI, some of that had been made more simultaneous. But what AI has done, and promises to do in more areas, is make all of it run at the same time. Especially with specialized agents, you can do decentralized science. You don’t need to know everything about every domain. I still think expertise matters, but it accelerates the whole process.
Turner Novak:
So how does that actually work? Did you need AI to do this? What does it unlock?
Karthik Duraisamy:
That’s the thing. AI is an enabler. Think about AI having access to skills and tools. Skills is my scientific expertise, encoded as a set of rules, literally in a text file.
Turner Novak:
So you’re an expert on aerospace physics. You know everything about how aerospace intersects with the world. And that’s all you know, and you don’t know anything else.
Karthik Duraisamy:
Yeah. Plus I have a certain kind of insight, a certain flavor of the scientific method, that I can also build into that. My approach might differ from yours because we have different tastes, different perceptions, different fields. All of that can be encoded as skills. Agents have access to skills.
And then there are tools. A simulation software that takes a real-world problem, turns it into a computation, helps you understand and manipulate reality virtually. Tools could also be experimental facilities, a lab setting, robots doing repeated experiments, all connected and coordinated by AI. Human expertise plus AI plus tools is what makes this happen. Sometimes the reasoning capabilities of these models are so powerful, especially the last six months or so, that AI seems to be identifying some of the human skills on its own. But I still think human expertise matters. Short answer: AI with access to tools and skills is what makes all of this possible.
Turner Novak:
And this intersects with the institute. What is the institute going to be doing practically on a daily basis?
Karthik Duraisamy:
At least at the beginning, the core job is making sure the open source development of OpenClaw is done well and maintained properly. There are so many users, and they need confidence that this is going to stay open source and keep being developed.
Turner Novak:
And there are two other people running the institute with you.
Karthik Duraisamy:
Yeah. For now the leadership team is myself, Professor Brad Orr from the University of Michigan, and Kurt Skipstad from engineering. That’s the initial team. If you think about the core piece, there are the core developers and maintainers. Then there’s the next layer, like my students or colleagues’ students who want to apply this to specific problems, not just using the software but adapting it for specific use cases. And the reason it makes sense to center this in a university is we have people in every possible discipline, economics, sociology, medicine, engineering.
Turner Novak:
So you can just say, “Hey, we need legal experts,” and there’s just a person who works for the school who’s the world expert on that thing.
Karthik Duraisamy:
In many fields, the world expert is sitting right here. But the other important point is it’s not just about developing and applying software. There’s also the physical piece, robotics is important and we have an amazing robotics department. But also developing this responsibly, and in a sense preparing society for what’s coming. Because a lot of the apprehensions people have about AI aren’t even about the technology, it’s about how fast it’s moving and how it disrupts existing structures.
Can we design institutions that are ready for this? Can we do this responsibly? And I want to make it clear: this isn’t just the OpenClaw Foundation and the University of Michigan. That’s the core, the central node. For this to succeed, we need representation from everywhere in the world. Think of it as a meeting place for that.
Turner Novak:
So could I get involved, as a person who doesn’t work at the university or OpenClaw in any way?
Karthik Duraisamy:
Absolutely. There are three million users of OpenClaw right now, probably around 500,000 GitHub repositories. In a sense they’re already part of the ecosystem. This centralizes some of the most important effort, but it’s a window to developers and contributors around the world. And the other thing I want to emphasize: nobody can predict with any confidence how the technology is going to evolve in the next year or two. So we’ll be very adaptive to the changing scenario, but the goals are clear.
Turner Novak:
Yeah, this might be completely outdated in six months.
Karthik Duraisamy:
I don’t think the core methods will be outdated in six months. A particular piece of software might change, but things do last longer than that. Many other things will change in unpredictable ways, though.
Turner Novak:
You mentioned that these institutes are almost like a VC fund in a way. Can you explain that?
Karthik Duraisamy:
When people ask what a professor does, they think our entire job is teaching. It’s certainly not. There’s teaching in the classroom, mentoring research students, mentoring PhD students to do original research, leading a research group, running a lab doing world-leading work in your particular domain.
One way to think about almost every professor at a top research university is like a startup founder. You’re recruiting some of the best students in the world. I just completed my PhD admissions, I got 200 applications and accepted maybe two. And you’re competing with MIT and Stanford. Just like a startup, you’re fighting other places to recruit top talent. And all of that costs money, so you raise it. Just like every startup founder does.
Turner Novak:
It doesn’t just show up?
Karthik Duraisamy:
It does not show up, unfortunately.
Turner Novak:
So how does the funding work at the University of Michigan?
Karthik Duraisamy:
Normally the largest portion comes from the federal government, NASA, National Science Foundation, Department of Energy, Department of Defense, NIH, etc. We write proposals on research ideas. It’s very competitive, maybe one in five proposals gets accepted on average. At Michigan it’s higher, but one in five is the norm.
Turner Novak:
And this is basically saying, “Hey, government, I have this idea. Here’s the impact it could have on the world. Give me $10 million to work on figuring this thing out.”
Karthik Duraisamy:
Pretty much. You present evidence: here’s my past work, here’s some preliminary results that show a new direction is promising, here’s my five-year plan. Then you ask for funding. Very competitive.
Turner Novak:
What makes it so competitive?
Karthik Duraisamy:
There are so many excellent universities in the world. It’s not just Michigan. There are so many other research groups pushing the envelope. Anyway, coming back to the original question, being a professor is like being a startup founder. And institutes that bring together different faculty can be thought of as incubators or VCs. In MICDE, the institute I direct, we often identify an interesting direction of research that isn’t mainstream yet. We bring together faculty, students and postdocs, build critical mass around that area, and give some seed funding. We have something called the Catalyst Grants Program.
Turner Novak:
So this is without going to the government.
Karthik Duraisamy:
Exactly. Smaller amounts, not $10 million, but say $100,000. People can use that to explore an idea, and then the institute helps those professors put together a bigger proposal. We recently won a $20 million center from the Department of Energy, which we’re very proud of.
Turner Novak:
What was that?
Karthik Duraisamy:
It’s called the Predictive Science Program. A new center called CPRIME, Center for Prediction, Reasoning and Intelligence for Multiphysics Explorations. Expertise, computation, and AI coming together to address a real problem. So yeah, institutes as incubators, that’s a pretty good analogy.
Turner Novak:
What does a good research proposal look like? How do you know if something is worth spending time on?
Karthik Duraisamy:
We don’t send proposals in a completely blue-sky sense. Some foundations say “give me your best idea,” but that’s rare. Normally the federal government has specific requirements around a topic.
Turner Novak:
So they have like a request for research they put out.
Karthik Duraisamy:
Yeah, it’s called an RFP, request for proposals. Someone may be interested in nuclear fusion, propose ideas that can improve the efficiency of fusion.
Turner Novak:
So there’s someone at the federal government level who’s in charge of dishing out these grants. They might say, “We want work done in nuclear fusion,” and “We have $2 billion earmarked and we might fund 20 projects.”
Karthik Duraisamy:
Generally it’s not $2 billion, but something like that. The right order of magnitude: funding a PhD student for a year costs about $100,000. A full PhD is around $500,000 over five years. Faculty members typically have five to ten students. So median research budget per faculty member is probably around $500,000 a year.
To give you a sense: Michigan has the third-largest research program in the US by dollars. Total research activity is about $2.2 billion per year. Not all of it is federal, I’d say about 60% is. The university itself puts in around $700 million a year, which I believe is more than any other university spends on its own research. Then state funding, industry funding. A whole range.
Turner Novak:
This is basically like a corporation outsourcing its R&D to a university, saying, “Do this for us, we want to make products from it”?
Karthik Duraisamy:
Sometimes they’re looking for a specific solution, like outsourced R&D. Sometimes they’re looking for good ideas. Sometimes they’re looking for due diligence because we’re experts and we know how to judge things. It’s a combination.
Turner Novak:
And what’s the benefit for the university? What do you get out of doing research that some other company benefits from?
Karthik Duraisamy:
Well, first of all, they pay us to do it.
Turner Novak:
Okay, fair.
Karthik Duraisamy:
That’s a start. But beyond that, after Stanford and MIT, Michigan is the third-largest producer of spinout startups from university research. About 32 or 33 startups per year. So it’s not like money comes in and we just produce papers and graduate students. Innovation is happening, and some of our colleagues have done really well in that area. But we don’t take money just because it is money. Every faculty member is interested in furthering the boundaries of knowledge in their area. If the funding is aligned with that, that’s the way of having impact. Training students is impact. Solving problems that somebody cares about is impact. Pushing the boundaries of research for the sake of it is also impact.
Turner Novak:
So with a startup, the university owns some of the equity, and when there’s an exit, it goes back to the university? How does that work?
Karthik Duraisamy:
If it’s a startup spinning out of research that happened at Michigan, funded by the government or otherwise, yes, the university takes some equity and some part of the IP, royalties, etc. But I have a startup myself, and the university is actually very fair. They’re not here to make money off it. They genuinely want to help innovation grow, want the impact of faculty and students to be higher. There’s certainly some economic benefit too, but that’s not the primary thing.
Turner Novak:
So I think it might be interesting to talk about how the world is changing because of AI. You mentioned there are a lot of academics who almost don’t believe in AI. It seems like you’re all in on it. You’ve basically created initiatives to lean into it and you’re using it to do research. But some people think it’s a fad. What’s going on there?
Karthik Duraisamy:
I think with something like AI, multiple things that seem contradictory can actually both be true at the same time.
Turner Novak:
Okay. How so?
Karthik Duraisamy:
Some very learned professors say, “I asked AI this question and it hallucinated a nonsensical answer, therefore it doesn’t work.” But they did that two years ago. A lot has happened since then. The core thing is you have to separate the marketing and hype from the actual model capabilities and the scientific value. Sometimes people mix those issues. In some domains, AI is already incredibly useful, coding, mathematics, theoretical physics. In others, it’s still catching up. Both things are true.
Turner Novak:
So how is it good in those things? In your domain, where are you seeing AI being extremely useful?
Karthik Duraisamy:
Why are coding and math more favorable for AI? Because if the AI does something wrong in those domains, you know immediately. If it produces wrong code, it won’t compile. Or you run it and get the wrong answer. The feedback from the output of the model back to its thought process is very direct. These are called objective metrics. That’s where AI is already very powerful.
And the other thing that’s been remarkable about the latest frontier models is their ability to compose ideas from different fields and bring them together seamlessly. Half-baked idea here, half-baked idea there, they see patterns and merge them. That’s genuinely powerful. And many people use AI only in chatbot mode. They don’t take advantage of the tools that can be built around these models. In mathematics, for instance, there are things called interactive theorem provers. AI can suggest something, it goes into the theorem prover, and the theorem prover gives important feedback back to the AI. If you’re just using it as a chatbot, you’re missing a lot. We come back to the same formula: human expertise plus AI plus tools. That’s when you see the real benefit.
Turner Novak:
So it’s not just, “AI, cure cancer,” and it goes and does it?
Karthik Duraisamy:
Not possible. In some domains, AI is proving mathematical theorems that people hadn’t touched before, either for lack of attention or lack of patience. But cancer, fusion energy, most hard practical problems, those have to go through the full scientific process. If AI identifies a drug candidate, that has to be verified computationally, then verified through trials.
Think about bottlenecks. When the bottleneck is purely cognitive, like math, which is all cognitive, AI can probably go all the way soon. But in many problems the bottleneck isn’t just cognitive. Physics has to agree. You have to build an expensive experiment to test your idea. In those cases, you still need human plus AI plus tools. There’s something called Amdahl’s Law: if there are 10 units of work and AI completes 8 instantly, the remaining 2 are still going to slow you down. Just because you’ve handled 90% of the work doesn’t mean the rest happens on its own.
Turner Novak:
So what are some of the bigger bottlenecks we’re running into?
Karthik Duraisamy:
I don’t want to give the impression that everything cognitive has been solved. Current AI models still have many limitations. They’re good at language and reasoning, pretty good at math and computation. They’re not as good as humans at spatial reasoning or physical reasoning, that’s where physical AI and robotics lags behind language and math. The bigger bottleneck for truly hard problems, like actually discovering a superconductor I can use tomorrow, is when you interact with the real world. You have to build something and test it. AI doesn’t do that on its own. Not yet, anyway.
Turner Novak:
So what are some of the big ways you think the world is going to change as AI gets better?
Karthik Duraisamy:
It’s a very broad question, and if anyone answers with absolute certainty, they’re probably not being honest. You can only talk about likelihoods. Here’s how I think about it: we’ve built the entire economy around scarcity and friction.
Before the internet, information was scarce. It was a commodity. Think about the major inflection points in the history of intelligence, language, writing, printing, the internet. Each one broke down information barriers and made certain things less scarce. But until large-scale AI, knowledge and cognition were still scarce. You had to basically learn from age five to age 22 to become competent in a field. And now something you can get for the cost of a Netflix subscription gives you not just information, but knowledge and intelligence. The scarcity of certain things is getting wiped out.
The other piece is friction. A lot of the economy was built around the fact that moving something from here to there required someone to pick it up and move it. If you wanted to buy a house, you went through a realtor. That’s friction. Sometimes friction is good, actually. But AI is removing a lot of it. The value propositions we used to place on different things is changing before our eyes. People who were extremely good at remembering things, at doing math, at a particular type of skill, those were very valued. I still think there’s value to many of those skills, but the proposition will shift.
Turner Novak:
And you gave a “code red” to your students. What was that?
Karthik Duraisamy:
About a month ago I brought all of my PhD students into a room, a dozen of them, and said: these reasoning models, even before we’re talking about agents, are able to reason through things at the level of my expertise in areas where I’m one of the world’s experts. I’ve had many weekend projects where I have a research idea, describe it loosely to an agent, and it does the research, explores configurations, writes code, tests ideas, comes back with results, writes reports. Things that would have taken me four or five months, getting done in a weekend.
I don’t want to say all of my research can now be compressed into a weekend. But many of these tasks are things I couldn’t give to a second-year PhD student at Michigan, and we accept around 5% of applicants. These are among the best in the world. If AI does some of those tasks as well as a second-year PhD student, it raises a question.
Turner Novak:
So what’s one of these tasks that AI is now doing at the same level or better? Just give me an example.
Karthik Duraisamy:
Before I do, I want to say a PhD student is not just about completing tasks. A PhD is not just task completion. PhDs will still exist. Original ideas exist. There is value in training people, and people come up with ideas in ways that are very different from what AI does. So I don’t want to equate a PhD student to just executing a set of tasks.
But as a professor who wears many hats, researcher, teacher, startup founder, institute director, my time is very splintered. Generally, if I have a research idea, I’d work on it over a weekend or a few weekends. If it seemed reasonable, I’d say to a student, “Hey, why don’t you look at this?” That’s how we did research until two or three years ago.
Turner Novak:
So you’d come across something worth spending a ton of time on and suggest the idea to someone.
Karthik Duraisamy:
Yeah. And sometimes the students themselves go through that process, they work on something for a month, come to me, and their idea is better than mine. Regardless, going from ideation to actually having a research idea that makes sense takes time. Especially in fields like mine, I’m a computational scientist, so most of what I do involves wrangling equations, computing them, looking at physical phenomena, modeling them. Many research ideas in my field can now be tested out very quickly. And as the agent works through ideas, even if you’re guiding it, it can give you connections that steer your own thinking in new directions.
Long story short, I called my students in to say: until around December 2025, I always talked about these tools in terms of future potential. A year ago, absolutely horrible. Six months ago, less horrible. Now, decent. That was my framing. But they’ve reached a threshold where the ideas coming out of these models and the way they’re reasoning is about as good as leading-edge research. If we’re not using these tools right now, we’re missing out.
But there was another part of the code red: you develop a lot of intuition by doing things the slow, rigorous way. There is value in students not using AI and developing their own thinking. I still want students to have original ideas, do things the hard way, make mistakes, learn from them. If you completely remove that friction, you lose intuition for how things work. But if you completely ignore these tools, somebody gets there faster than you. It’s a very hard balance and it’s happening in every field. Code red was about showing them how powerful these tools are, but also not forgetting to do things thoroughly.
Turner Novak:
Yeah. So it’s basically telling them they need to completely master AI and also master not using AI at the same time?
Karthik Duraisamy:
Don’t substitute AI. Maybe we have to separate the learning phase from the creating phase, even though those overlap. The only way you learn physics is by doing problems, by working through them. AI knows the answer, but if you skip that friction, you lose the intuition it builds. At the same time, don’t be oblivious to the tools. They can be used in the right way.
Turner Novak:
It reminds me of the CFA charter. It’s kind of like a CPA but for investments. There are three exams. Level one is basically everything you’d learn in a finance undergrad, one test. Level two is like a master’s in finance. Level three is more theoretical, closer to a PhD in finance.
At the end of it, you need to be able to answer something like: Paul and Linda are in Canada, they have a portfolio in the US with some investments in Bangladesh and France in local currency, they want to hedge in Mexican pesos, they have a kid going to college in 18 years, here’s the portfolio size, here’s the growth target. Now give them a recommendation. And you have to do all these hand calculations, what hedges do they need, inflation, expected returns across different asset classes.
At the end of the day, you’re never actually going to do all of that by hand in real life. You’d Google it or hire an expert. But I had to go through and learn all this stuff and it sucked. And I thought, I’m never going to actually use any of this. But it does give you some intuition for how to think about these things. In investing or finance or business, there’s usually a spreadsheet that someone’s looking at to make a decision, but you need to know what goes into that spreadsheet. Same thing in physics: you need to know what’s influencing the outcome. You don’t have to solve it by hand. But you do need to know how it works.
Karthik Duraisamy:
Yeah. By going through that process, doing it by hand, getting it wrong, having your professor tell you what you missed, that’s how you actually build the intuition to see a problem. Intuition is always useful, but the judgment you need to give can now be informed differently.
Turner Novak:
How have you seen the way students learn change over time? And how do you think it’s going to change going forward?
Karthik Duraisamy:
We’ve had three shocks in the past few years. One is COVID.
Turner Novak:
Oh yeah. I almost forgot about COVID.
Karthik Duraisamy:
I know. People went online, high schools were less rigorous in certain ways, and you saw it show up at the university. Even in 2022, the incoming undergrad class, I could see some mathematical deficiencies. I’d even call it persistence: the willingness to keep going at a problem. I could see a bit of that dropping. And you could say it had been happening over a longer period anyway, probably Google and smartphones started it. I don’t want to say students aren’t great, because they are. But certain levels of mathematical rigor and persistence, especially at the undergrad level, we could see it declining a bit. And then COVID was a real shock on top of that.
Just as we were recovering, we got ChatGPT in November 2022. At that time the models were pretty bad, hallucinating constantly, but they could still do a bunch of things. The bigger moment to me is the last six months, where at the undergraduate level, however hard you make a question, the best AI models can do it in pretty much any field.
Turner Novak:
So any kind of take-home test situation, kids are just acing everything?
Karthik Duraisamy:
I’m actually impressed by the honesty and integrity. We have a tradition in the School of Engineering: for the past 150 years, no exam has ever been proctored. The professor hands out the exam in class and steps outside the classroom. We’ve done it for 150 years and I still see much of that persisting. But without a doubt, students are using AI tools to study and to replace some parts of their thinking. I’m not naive, some students are probably fully relying on those tools. But what we end up missing is that persistence, that struggle.
I teach a class that students find to be one of the hardest in my department, but also one of the most enjoyable, because the subject itself is beautiful.
Turner Novak:
What’s the subject?
Karthik Duraisamy:
Aerodynamics. You learn how wings generate lift, how to design wings for certain properties, how much power you need to move an aircraft. You learn math, physics, and engineering all in one particular way. It’s quite abstract, some students struggle, but everyone enjoys it. It’s a hard class, but students learn a lot.
Six years ago, I’d give a homework problem and half the class wouldn’t even know how to start.
Turner Novak:
That would be me.
Karthik Duraisamy:
I don’t take pleasure in torturing students, but I knew it was excellent for learning.
Turner Novak:
Can you give me an example of something I’d get from you where I just wouldn’t know where to start?
Karthik Duraisamy:
I teach how aircraft fly and how those trailing vortices work, when a plane flies through clouds, you see them roll up behind the wing. And in one homework problem I’d say: “Here’s a flock of geese flying in a V formation. Tell me how efficient flying in this V formation would be”, without having mentioned many of the relevant details ahead of time. Many students wouldn’t know where to start. Even with Googling you could get some help on certain things, but some problems I’ve given, Google has nothing. Students spend a few hours just figuring out how to begin. But you learn an enormous amount in that struggle. The hour and a half where you go nowhere is actually where you learn. And now that’s gone.
Turner Novak:
Because you can just type it into ChatGPT.
Karthik Duraisamy:
Any question, any course, any level, any university, including problems from the most famous mathematician alive, AI will either just solve it or recommend six or seven directions to start. So yeah, something is being lost.
Turner Novak:
The geese question, I’m curious now. What would I even be trying to figure out?
Karthik Duraisamy:
I told you how a single aircraft flies and how the trailing vortices look. Then you treat each goose as a little aircraft with a certain mass it needs to support to fly. You create a bunch of these little airplanes with the corresponding vortices and optimize over the whole configuration. I’d give you the size of the geese, the spacing. Maybe that’s not the hardest example, but for some students even that connection is hard to make.
In general, it is in that friction and struggle that you actually learn. If that’s being replaced, you don’t build as much intuition or judgment. Like that CFA scenario: doing it persistently, rigorously, by hand has a lot of value. And that’s being replaced.
Turner Novak:
And the interesting thing is where software comes into play, all these little mini calculations you have to make. If you make one mistake anywhere in the chain, it breaks the whole thing. A calculator chains those steps instantly. AI just extends that further, it creates all those calculations automatically. So the skill becomes knowing how to use AI and an agent to solve a problem, not how to use a calculator.
Karthik Duraisamy:
The meta, right? But what if one of those agents is doing the wrong thing? You have a sequence of them and one is wrong.
Turner Novak:
So you still need to understand what the agents are doing.
Karthik Duraisamy:
Yeah. And the other way to think about it: if everything can be automated, what value do you bring to the problem? Anyone can do it.
Turner Novak:
Do you think we’re going to have to learn how to create and manage and run agents? Is that what education is turning into?
Karthik Duraisamy:
I actually don’t think so. Did we need a lot of education to use ChatGPT 3.5 when it came out? No. But you needed a lot of common sense to use it effectively.
Two weekends ago, a student came to me with a brilliant idea. He understood it better than I did. Over the weekend I was wrangling with it, so I asked an AI agent, described the problem, gave it all the context, it wrote code, and I wanted it to create a visualization to help me understand what was happening. It did everything well, but the visualization was off. The angle it was showing me was wrong because it had no spatial reasoning. So I said, “Help me help you, describe the view you’re showing me with some numbers, and make the figure interactive so I can rotate it. When I rotate it, those numbers update. I’ll tell you the best angle.” Human and AI working together.
People say, “The great education is how to use AI.” No. In a couple of months, using an agent is going to be clicking a button. The skill is how you synthesize what you know and use your common sense to wrangle with AI. Learning how to use AI technically, that’s never going to be the hard part.
Turner Novak:
So how does the role of the university change, with education becoming either more or less important?
Karthik Duraisamy:
It’s been changing for decades already. We used to be gatekeepers of information that nobody else had. Then slowly the internet and online courses started eroding a bit of that.
Turner Novak:
Like when MIT put their classes online.
Karthik Duraisamy:
That’s one of the most seminal moments. I think people should talk about it more.
Turner Novak:
When was that? About 25 years ago?
Karthik Duraisamy:
Around the early 2000s. MIT took all their class notes, videos, lectures, homeworks and put them on the internet, free, for anyone in any part of the world. Did it change the world enormously? For certain people, yes. But it was passive. It was revealing information, not teaching. Now AI is giving you cognition and knowledge. A lot of what universities used to do will be less valuable than before. As keepers of knowledge, you had to go talk to the expert. That knowledge is now pretty much encoded, in a skills file or agentified in some way. We’ve been keepers of knowledge for thousands of years, and now suddenly this thing you pay $20 a month for knows most of it.
But universities still have a big role. Just having access to content doesn’t mean people learn. Just because knowledge is accessible for $20 doesn’t mean people extract the most from it. Universities still add value by bringing young people of a certain age together in a certain environment. Putting 25 people in a class and introducing friction in a certain way, deadlines, exams, making you think. People learn from people. There’s personal mentoring, especially at the graduate level. And you have access to specific facilities. Michigan has the nation’s most powerful laser, called ZEUS, right here. We use it to study theoretical physics, plasmas, material properties. That doesn’t exist inside ChatGPT.
My work is computational, but many colleagues have physical labs, one of the best battery design facilities in any US university, for instance. These are specialized resources. Many discoveries and innovations come directly from university labs. The president of Arizona State has this great one-minute video where he picks up an iPhone and says there are 800 technologies in it developed at university labs. Apple adds value by putting things together, but the underlying work came from somewhere.
And then maybe now more than ever: credentialing. Universities, for better or worse, filter. Even to get into Michigan, the acceptance rate is around 10%. After you come in, you go through a program where you’re credentialed and graded. I think those credentials become more important now, because there are easy ways to complete work without truly doing it.
Turner Novak:
Because they prove that you know the topic, or at least that you’ve gone through the motions of learning it.
Karthik Duraisamy:
Partly, yes, even just getting in is a signal. And I’m not saying that’s a good thing, but it is what it is. You go through the program at various levels of rigor, the university credentials you. But I would say grade inflation is probably going to slow down now.
Turner Novak:
I’ve seen those charts where the average GPA used to be like a 2.3 and now it’s like 3.8 or something.
Karthik Duraisamy:
Like 3.6. Harvard had big rules about too many people getting As. I think these distinctions will grow and evaluation will be taken more seriously, partly because now there are easy ways to complete work without doing it yourself.
Turner Novak:
Do you know what was driving grade inflation? From your perspective as someone actually giving out grades.
Karthik Duraisamy:
Many things. Students are genuinely better prepared coming into college now than 20 years ago, that’s one factor. And in most universities the expectation has shifted to “if I work hard, I get an A,” which many times is actually correlated with outcomes. But I’m not naive, if you’re paying $60,000 a year, I’m not saying that’s why grades are inflated, but if you dismiss it completely, you’re being naive. All of these things are probably true simultaneously.
Turner Novak:
And you think the inflation is going to stop? Level out or come back down?
Karthik Duraisamy:
I’m seeing a movement where things are either leveling or starting to come back down a bit. Harvard actually pulled it off, showed that 70% of students were getting As, professors pushed back, brought it down. Students complained about mental stress. It’d be stupid to dismiss that claim. It’d also be stupid to completely ignore what the professors are doing. Both are probably true.
Turner Novak:
For a lot of people, the reason you go to university is to get a job, increase your chances of success, get the credential, get the stamp of approval. What do you think is going to happen to the job market over the next decade?
Karthik Duraisamy:
Nobody can predict, but some things seem clear. You’re seeing about 100,000 tech layoffs per year over the last couple of years, and in 2026 we’re already at around 100,000. That’s a large number, but we have about seven or eight million tech workers, it’s roughly 1%. I don’t want to minimize the pain it’s creating, but I don’t think it’s going to be as dramatic as some people are saying.
Turner Novak:
Some people would say AI replaces everything and everyone’s unemployed.
Karthik Duraisamy:
That’s probably going too far, at least in the short term. The economy is built in a way where these impacts take longer to penetrate. I don’t think 20% of the population will be unemployed in the next two or three years. Unemployment is around 4.5% right now, maybe it goes up by one point, which is actually pretty bad. But I find it more concerning for fresh graduates getting into jobs. Many entry-level skills in certain domains can be automated. But if you skip developing those entry-level skills, you never become an expert. It’s a chicken-and-egg situation. And people talk about things like everyone having a billion-dollar startup, but where’s the market? Who’s buying?
Turner Novak:
You’re talking about the one-person billion-dollar startup.
Karthik Duraisamy:
I’m sure there’ll be a few. I think there might already be one. One of my colleagues, Jerry Davis in the business school, has colorful thoughts on what he calls “zero-employee unicorns”, a completely agent-run company. I don’t think it’s out of the question, but those will be exceptions. In summary: people already in good jobs will probably find AI helps them be more productive and do more. But I worry more about entry-level jobs. That’s the bigger concern.
Turner Novak:
It’s just harder to get that first job, because instead of a manager hiring someone new, you just use AI and software.
Karthik Duraisamy:
Not in every profession, and not completely. But you’d have to be naive to think it won’t have an impact.
Turner Novak:
One of the recent guests on the show talked a lot about the US healthcare system. He made the point that throughout American history, we’ve had these job-program-like industries, manufacturing was a built-in employment mechanism. Healthcare is kind of that way right now. A lot of cities and regions have some healthcare system as the largest employer, nurses, admin, people who help move patients around hospitals. A lot of those roles exist partly to employ people. Even if AI makes things more efficient, the government is largely funding it and isn’t going to say, “Let’s eliminate these jobs.” So there are multiple layers of friction protecting those roles.
Karthik Duraisamy:
In the short term, I agree. But I can’t see clearly five years out, and things can get non-linear in ways that are hard to predict. You’re right that many jobs are structured in ways that will protect people for a while. There will still be layoffs, but I don’t think it’ll be as catastrophic as some people suggest, at least in the near term. My bigger concern remains entry-level jobs. That’s the one I keep coming back to.
Turner Novak:
What advice are you giving your students? “Hey, if you want a job, here’s what you have to do.”
Karthik Duraisamy:
It’s hard advice, but it comes down to: whatever you study, get more rigorous about it and add real value. Get really good at what you do. I don’t tell them to use an AI tool. That’s not the priority. The priority is going through the process, building intuition, and then using the tools out of necessity and common sense. I’m at an age where many of my friends have kids in 10th grade thinking about college. People used to think computer science was a sure shot to an amazing career.
Turner Novak:
Yes, becoming a millionaire, coasting forever.
Karthik Duraisamy:
So I tell them: it doesn’t matter what field. Even if you study computer science, there’s theoretical computer science, there are so many things beyond just coding. But if people ask me openly, “What should we study?”, maybe I’m biased, I’d say: physics, chemistry, biology, mathematics.
Turner Novak:
Those are all things AI can do really well right now, right?
Karthik Duraisamy:
AI can do well up to a certain point. But most of the unsolved grand challenge problems in humanity involve those fundamental sciences. And there are deeper questions, what is the nature of life, where did we come from, what is the nature of reality, those aren’t going away. There’s no chance someone produces a unified theory of physics tomorrow and that’s it. So: study things for their intrinsic merit, do it really well, build up your basics. The problems in those fields are never going away.
But also: get really deep, and know how to synthesize things across disciplines. Don’t skip the fundamentals. You need to know something really well to have genuine expertise. Not everything is automatable. AI can do your math, physics, chemistry, and biology homeworks. But the hard skills will matter, as long as you can synthesize information across different fields.
Turner Novak:
Yeah. One piece of advice I always give people: niche down a lot more than you think. If somebody just says, “I want to be a scientist,” that could be anything. But if you say, “I want to be a scientist who works on making paper cups more durable, the strongest paper cups ever”, I know exactly who you are. If I ever come across a problem requiring that, you’re the first person I think of.
It’s like being a soccer coach trying to improve your team. You’re at tryouts and there’s this person who is just insanely good at throw-ins. Maybe they’re not the best at everything else. But you might make the team because of that one thing. Or corner kicks, someone so good at that one piece that they make the squad because of it.
Karthik Duraisamy:
Right. The ideal profile is something like a T. Someone who knows a little about everything is like a flat rectangle. Someone who only knows one thing is like a vertical I. You need enough breadth to see how things connect, but you need to be really deep at some things. That value isn’t going away.
That said, it also won’t be simple. Being smart, being good at something, that alone used to be enough to make it in the world. Those skills are still useful, but not sufficient on their own anymore. The value proposition is shifting toward people who solve real problems. The cost of generating an idea is approaching zero now, because AI can generate ideas. Not all good, but some. What matters is taking that idea and going all the way through, actually solving something people care about. Earlier it was, “He’s smart, he knows math, he can do calculations in his head.” Now it’s, “Okay, so what? What are they actually doing with that?”
Turner Novak:
And you’ve actually seen two different college basketball national championships firsthand as a member of the school. What is that like?
Karthik Duraisamy:
Amazing. I’m a sports junkie. I enjoy college basketball more than most sports, though it’s not my top one.
Turner Novak:
What’s your top sport?
Karthik Duraisamy:
Soccer. I’m incredibly passionate about it.
Turner Novak:
That was actually a lucky guess when I gave the soccer example.
Karthik Duraisamy:
Anyway. When I was a graduate student at the University of Maryland, we won the national championship in men’s and women’s basketball, not the same year, a couple years apart. And I got to experience it again at Michigan a few days ago. A couple of my professor colleagues and I were out on South University with all the celebrating students. I’m sure they had no idea we were faculty. But it was amazing. 64 teams, straight knockout, it’s a real accomplishment. And sports brings out certain emotions that most things don’t. Where else do you see 10,000 people completely happy, forgetting everything else, all celebrating together? To be part of that is something else.
Turner Novak:
Yeah. I’ve seen videos of robots playing sports and people say “soon robots will be better than us at sports.” But I don’t actually want to watch robots play sports. Do you think human sports will ever get fully replaced?
Karthik Duraisamy:
I don’t think so. Computers have been better than humans at chess for 20 or 30 years. Chess is more popular now than it was when computers first beat humans. There’s something in human pleasure that comes from watching people compete. That doesn’t go away.
Turner Novak:
Yeah. There’s this concept called the bionic games. Like the Olympics but you’re allowed to enhance, drugs, robotic limbs, whatever. If you lost your arm and got a replacement that’s stronger than a natural one, you’d be allowed to compete with it. That kind of thing.
Karthik Duraisamy:
There’s a market for everything. One of the most popular sports in the world is motor racing. That’s already human and machine.
Turner Novak:
So we’ll probably just get new sports. That’s what actually happens.
Karthik Duraisamy:
Yeah, that’s very possible.
Turner Novak:
And I think you told me once that you almost died hiking in a national park. What happened?
Karthik Duraisamy:
My wife and I travel a lot, and there was a time we did some serious trekking in different parts of the world, Argentina, here and there. One time in Grand Teton National Park, it was a snowy day and somehow they gave us a permit to climb and pitch camp. An overnight trip. A ranger was coming down as we were heading up and said, “I’m quite shocked they gave you a permit, weather conditions aren’t great.” We said we’d be fine, we’d done plenty of hikes. He said, “Don’t make me come rescue you up top,” and gave us some vague instructions about keeping to the right of the trail if the path disappeared.
We kept going. Light snow at first, we enjoyed it. Then it got heavier, the trail disappeared, and we couldn’t follow whatever instructions he’d given. I fell into a hole, not too deep, and then my wife fell into a separate hole about 20 minutes later. Some of our backup socks got wet. Then it snowed heavily. We found something that looked like a trail, pitched camp pretty high up, and it snowed more. Because of the falls, socks were soaked through, even the backups. My wife’s hands were going numb. A few more hours and she might have lost some fingers. And then coyotes started howling, and she said, “I don’t want to die being eaten by coyotes.”
This was probably 2005 or 2006. Cell phones weren’t reliable out there, but I kept trying and somehow reached a friend in College Park, Maryland. He was able to call the rangers. They came up in the middle of the night with hot drinks and helped us break down camp. And it was the same ranger who had said, “Don’t make me come rescue you.”
Turner Novak:
Oh geez. Was this midnight or the next day?
Karthik Duraisamy:
Around midnight or 2 AM. He was having dinner with his wife when he got the call. So yeah, we’ve had some adventures.
Turner Novak:
Any upcoming trips planned?
Karthik Duraisamy:
Some trips to Japan and Europe. Nothing big on the hiking front.
Turner Novak:
More laid back?
Karthik Duraisamy:
More laid back, yes.
Turner Novak:
Well, this has been awesome. Thanks for taking the time to chat. A lot of fun.
Karthik Duraisamy:
It’s a pleasure. We touched on so many different topics. You were certainly a good host.
Turner Novak:
There’ll be a lot of data out there for the LLMs to train on. Maybe we can teach the language model something.
Karthik Duraisamy:
I don’t know how much real insight I had, but I’m sure I compressed a few things.
Turner Novak:
Yeah. It’s a lot of fun. Thanks for being here.
Karthik Duraisamy:
Thank you for having me.
Stream the full episode on YouTube, Spotify, or Apple.
Find transcripts of all other episodes here.