1. Podcasts
  2. Daniel and Kelly’s Extraordinary Universe
  3. Big Math with Steven Strogatz
Daniel and Kelly’s Extraordinary Universe

10 Nov 2026 · 1h 12m · 23 chapters

Episode artwork

Big Math with Steven Strogatz

In short

The episode explores “big math” (math that exceeds human cognitive limits) and how it’s powered by computers—especially vectors, matrices, and higher-dimensional thinking—plus what this means for AI and the future of math education.

Guests/backgrounds

Steven Strogatz, Cornell University professor of mathematics; National Academies of Science member; known for accessible math writing (The Joy of X, Infinite Powers) and co-hosting Quantum Magazine’s The Joy of Y. Hosts Kelly Wienersmith (parasites and space; studies math in grad school) and Daniel (particle physicist).

Key claims

Big math is what happens when math meets computers and involves millions to trillions of parameters. Vectors/matrices “package complexity” so humans can reason about large systems. AI doesn’t replace mathematicians’ roles; it changes what proofs/decisions are needed, and math professors will still exist. Matrices are central to GPUs and machine learning because neural networks learn by updating matrices via backpropagation.

Notable examples

Untying knots by “lifting” into a higher dimension; topoisomerases using knot theory to infer DNA enzyme actions; moon landing guidance using state vectors and matrices; Burj Khalifa structural modeling with massive force/stress calculations; GPUs originally built for graphics now training large language models; Google’s PageRank using a matrix representation of the web.

Written by AI. May contain mistakes.

Chapters

2:42

The Role of Math in Society

Exploration of how math underlies our world and its future with AI.

“Whether or not you enjoy math and all of its puzzles, it's undeniably everywhere.”
4:40

Introduction of Guest: Steven Strogatz

Introduction of mathematician Steven Strogatz and his background.

“And we have the perfect guest on the show today to talk about that.”
8:12

Discussing Big Math and Its Implications

Deep dive into the concepts of Big Math and its relevance to modern technology and cognition.

“Let's jump right into the episode and see what Stephen has to say about this question.”
14:00

Understanding Higher Dimensions in Math

Explore how higher dimensions can help solve problems that are otherwise impossible.

“They're different locations in the fourth dimension.”
15:29

Describing People and Features

Learn about using feature vectors to uniquely specify and describe individuals.

“Like we can use other coordinates, color, time.”
17:21

Topology and Its Practical Applications

Discuss the role of topology and knot theory in biological systems, specifically DNA.

“like a person's demographic information, and thinking of it as a big bag of numbers.”
19:17

Mathematics Behind Engineering and Construction

Discover how mathematical concepts like vectors and matrices apply to engineering projects.

“index finger, and now link those two circles like links of a chain.”
21:37

The Moon Landing: Math and Human Ingenuity

Examine the critical role of mathematics in the success of the moon landing and the balance of human control.

“Can you tell us about why we need things like vectors and matrices and why that's important for engineering projects like building skyscrapers?”
28:00

The Role of Big Math in Decision Making

Discover how big math influences critical decisions in military and AI contexts.

“So I guess your point is that we couldn't have gone to the moon without big math that was bigger than what we could do ourselves.”
29:17

Introduction to the Burj Khalifa Discussion

Explore how matrices are utilized in engineering skyscrapers like the Burj Khalifa.

“And when we come back, we'll talk more about using matrices to build skyscrapers, talk about network theory, optimization, and all sorts of other fun stories from Stephen's great book, Big Math.”
Show all 23 chapters
32:20

Mathematics Behind Engineering and AI

Learn about the significance of matrices in engineering and their connection to AI.

“author of Big Math and professor of mathematics, dirty mathematics, apparently, at Cornell.”
42:04

The Power of Matrices in Technology

Explore how matrices underpin advancements in technology, especially machine learning.

“A matrix can tell you who's connected to who.”
46:06

Historical Insights on Matrices and Their Origins

Learn about the historical roots of matrices and their ancient applications.

“That's amusing thing number one about matrix.”
54:14

Mathematics and AI: A Changing Landscape

Discuss how AI is reshaping the field of mathematics and its perception.

“And Kelly has projected her negative feelings about algebra onto people she's never met from thousands of years ago.”
56:00

The Artificial Split in Mathematics

Explore the historical distinction between pure and applied mathematics and its implications.

“Did Isaac Newton 2 ,000 years later than Archimedes, did he have a list?”
58:25

The Role of AI in Problem Solving

Discuss the potential for AI to revolutionize how mathematicians approach problems.

“My 24 year old and 26 year old daughter taught me about Riz.”
1:00:21

Human Versus AI: Understanding and Curiosity

Analyze the differences between human curiosity and AI's potential to ask questions.

“Because a moment ago we were saying AI is a tool.”
1:02:55

Assessing Understanding in Education

Evaluate the challenges teachers face in ensuring true understanding among students.

“I think they will soon be better than us at exposition, at teaching, at asking questions, all of it.”
1:07:35

Identity and Community in Mathematics

Examine how identity among mathematicians is shaped by community and competition.

“I personally think this is the biggest source of strife and is similar to what's happened with other outmoded communities in the past.”
1:10:05

A Math Challenge from Childhood

Stephen shares a childhood anecdote about a math mistake that motivated him.

“Yeah, Bill and George in seventh and eighth grade.”
1:12:06

The Nature of Mathematical Identity

Discussion on how technology impacts professions and the identity tied to them.

“For a little quick context, so my husband and I, in the mornings, we talk about how different technologies have impacted the cultures of different communities of professionals.”
1:14:13

The Future of Math Education

Exploration of the future of math teaching and the enduring human connection to math.

“It probably won't be a career like I've had, but I do believe we will still have teachers and students of math who are human beings.”
1:18:31

The Sublime Nature of Math

Stephen discusses how math provides certainty and comfort in an uncertain world.

“So I think, you know, if you learn about math, even if you're not the best on the planet, this puts you in touch with something pretty close to the sublime.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On Daniel and Kelly’s Extraordinary Universe, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00This is an iHeart Podcast. Guaranteed human.

0:30for people age 12 and up who weigh at least 88 pounds with moderate to severe eczema, atopic dermatitis, that is not well controlled with Rx topical therapies or who cannot use those therapies. Don't use if you are allergic to EBCLIS. Allergic reactions can occur that can be severe. Eye problems can occur. Tell your doctor if you have new or worsening eye problems. You should not receive a live vaccine when treated with EBCLIS. Before starting EBCLIS, tell your doctor if you have a parasitic infection. Paid partnership with Lilly. Respect your time. Ask your doctor about Epclus and visit epclus.com or call 1-800-LILLY-RX or 1-800-545-5979.

1:04This is Jacob Goldstein from What's Your Problem? Running a business is hard enough. Don't make it harder with a dozen apps that don't talk to each other. One for sales, another for inventory, a separate one for accounting. That's software overload. Odoo is the all-in-one platform that replaces them all. CRM, accounting, inventory, e-commerce, HR, fully integrated, easy to use, and built to grow with your business. Thousands have already made the switch. Why not you? Try Odoo for free at odoo.com. That's odoo.com. So, everyone on board? Yeah, I think so. You can't see it, but everyone in this room is nodding.

1:45They all agree the launch plan's perfect, but Julie has an idea how it can be even better. So she asks Copilot Cowork to help her overnight. Hey, Copilot, gather the past three months of retail sales reports, surface key insights, and build a deck outlining a partner strategy by tomorrow morning. Morning, team. I found a way to improve the plan. Check slide four. Get help with complex tasks with Copilot Cowork. Microsoft Copilot. Work lives on. This is Oz Velocian from Tech Stuff. For many kids and teens, math class is a source of stress. And a lot of that stress comes from learning at the wrong pace.

2:20That's why I want to tell you about Brilliant. Brilliant is the online math tutor that's visual and interactive. It helps students actually understand math step by step and gives them guidance when they need it. Brilliant was built by experts from MIT and Harvard and is trusted by millions of teachers and parents. Sign up at brilliant.org slash tutor and get 20 % off for back to school. That's brilliant.org slash tutor.

2:53Whether or not you enjoy math and all of its puzzles, it's undeniably everywhere. Not only does it underlie the laws of physics, perhaps even revealing the source code of the universe itself, but we rely on it to build our world. Everything around you needs math to be designed, to be built, and to persist. Today on the podcast, we're exploring the role of math in our lives, how it has shaped society and your daily experience, even and especially when it goes beyond what can fit into the mind of a single human being. And of course, we'll dig into the question of the future of math in the face of AI's rapid improvements in pure math skills.

3:35Has math let us build tools that put mathematicians out of work? Welcome to Daniel and Kelly's Extraordinarily Mathematical Universe.

3:59Hello, I'm Kelly Wienersmith. I study parasites and space. And when I was in grad school, math was like my favorite thing to dabble in on the side. I would actually go over to the math department and sit in on the Ph.D. math classes. And they would be very, very kind to me when I asked really dumb questions. Hi, I'm Daniel. I'm a particle physicist and I love math. And I'm also glad that AI has not yet taken my job. Oh, that's a sad start for the episode. You know, I don't think AI has taken mathematicians' jobs yet, but we're going to be talking about the future of AI and math today. We certainly know that 2026 will always be remembered in math as a year that math has changed a lot.

4:42Yeah, yeah. And we have the perfect guest on the show today to talk about that. We do. We're talking to Steven Strogatz. He's a famous mathematician and podcaster and author, and he's here to talk about his new book, Big Math, which I read and really enjoyed and is out now. Y 'all should check it out. But we're also going to ask him questions about what it's like to be a mathematician in the age of AI, how that's changing, and what he foresees for the future of AI and math. Yeah, it's kind of interesting that you and I are taking such different paths from the beginning. I was trying to go the beauty of math and the amazing way it feels to solve a math problem you've been banging your head on.

5:20And usually you're the one with the beautiful flowery language. But today, you're leaning into the catastrophizing. Well, it's fascinating because there's so much of human science, especially in mathematics and theoretical physics, that leans on the flowery language, the beauty, the search for elegance. And AI is taking a very different approach. It's much more brute force, much more like, just get to the answer. We don't care if the proof is 600 pages of impenetrable gobbledygook. We just want to know yes, no. And I think that's a lot of the tension in the mathematical community right now. It's like, what is math about?

5:58Is it about learning truth about mathematical questions or is it about the understanding that comes along the way? Is it really count if you've learned a fact but don't digest it in a simple and elegant way? And so I think Stephen's book is really well-timed because even though it doesn't talk a lot about AI, it helps us understand the role of math in the world and the role of math in science and especially the power of computers to use big math to change how we live our lives. Yep. And like usual, we checked in with our extraordinaries to see what they thought about one of the main topics for the episode.

6:34So I asked people if we will still have math professors in 10 years or in 50. Here's what folks had to say. If there are no math professors, then no one will be able to understand what it is AI has proven or disproven or how we could use it. So yes, there will still need to be math professors in 10 years and in 50 years. Will there still be math for humans in 50 years? I hope so for all the people who want to be math professors in 10 or 50 years If there is not some sort of cosmic event that wipes us all out Yeah, I guess so Of course there will be I mean, for a number of reasons concerning about the efficacy of teaching And computerized learning But mostly because you can't prompt a human to give you the answers I think jobs like math professors and teachers are something that will always be needed.

7:28You still need a human creative mind because the AI tools aren't inherently creative and won't make anything new. Elegance of creative thought has yet to be replicated by AI. I have a feeling that it never will be. In 50 years, who knows what technology is going to be like. Yes, I think there will be math professors in many years to come, maybe hundreds. I think that there still will be math professors in 10 years and in 50 years. But they won't be finding proofs. They'll be deciding what needs to be proved. I think there will always be people who are fascinated with mathematics and other people who are willing to listen to them.

8:09Look at all that optimism. Not a single person thought there will be no math professors in 50 years. I live with someone who disagrees. I love that optimism. Let's jump right into the episode and see what Stephen has to say about this question. Because we talked to him about his book, and then at the very end, we asked him the same question about the future of math professors. It's my pleasure to welcome to the podcast Stephen Strogatz. He's a mathematician with a remarkable gift for making deep ideas accessible, vivid, and exciting. He's a professor at Cornell University and a member of the National Academies of Science, and he's made influential contributions to our understanding of complex networks while sharing the beauty of mathematics through his books like The Joy of X and the New York Times bestseller Infinite Powers, which is a personal favorite of mine.

8:56He also co-hosts Quantum Magazine's podcast The Joy of Y, exploring some of the biggest questions in math and science. His new book that we're going to talk about today is called Big Math, co-authored with Alex Townsend, exploring the mathematical ideas powering modern technology and a profound question. What happens when the tools we have created begin to outgrow our understanding? Stephen, welcome to the podcast. Thanks very much, Daniel. And hi, Kelly. Hey, nice to have you on the show. I think, Daniel, is this our first official mathematician guest on the show? I think it is. Wow. Yes. I think Stephen is definitely qualified as an official mathematician.

9:32What do you think? Maybe not in my circles. I'm on the very applied end of math, I might be verging into physicist or even engineering territory. Oh, wow. Uh-oh. So you're on the applied side of the pure versus applied math debate? I am as dirty as an applied mathematician gets. Oh, oh. Daniel, can we stop this interview? I just, this doesn't sound like a good idea anymore. Oh, you need a real mathematician? No, we're excited to have you on the show. Daniel has been telling me about your amazing books. Before you came on the show, he was saying he understands calculus, more deeply now after reading one of your books.

10:11That's true. So we're excited to have you on the show. And so tell me, what is the big idea behind big math? I would say it's that sometimes math gets too big to fit comfortably in our heads. You know, that as big as relative to human cognitive capacity. Think about, well, maybe most famously, we have trouble visualizing the fourth dimension, let alone the billionth dimension, you know? So when we talk about, say, in modern neural networks, it's not uncommon to hear of models with a trillion parameters. You know, the large language models that OpenAI or Anthropic have now, certainly hundreds of billions and even sometimes trillions of parameters, these different weights of the connections between the neurons or the different biases in the neurons.

11:00So if you have a trillion unknown numbers in math, we sometimes like to think of them as a single point. Just like in high school when you had the XY plane and you have an input X and an output Y and you graph a single dot with an X coordinate and a Y coordinate, that's two numbers on a two-dimensional space as a single point. And so similarly, when we deal with a trillion numbers, we sometimes think of it as a single point, but now there's a trillion different axes, which you can't possibly picture. So big math is – another way to put it is it's the math that happens when math meets computers.

11:42And so it's pretty modern. A lot of the ideas of it go back thousands of years. But once you scale them up to millions, billions, or trillions of parameters or unknowns, now you're firmly in the world of big math. But you must be comfortable thinking about higher dimensional problems and spaces. How many dimensions can you think about in your head? And how do you do that? Because I'm a physicist. I'm very physical. I have trouble thinking about higher dimensional spaces without trying to imagine a geometric setup. And of course, I can't go beyond three dimensions geometrically. Mm-hmm. Well, we're the same.

12:19I mean, I'm certainly the same. As far as really literally picturing it, I would get stuck at three dimensions myself. But, you know, we have all, in your training and my training, we do learn some tricks that let us go a little bit farther, not much. But sometimes, for instance, in the branch of topology that studies knots, how curves can have knots tied in them, and then you close the ends of the string. If you picture like shoelaces, you tie a knot, but then you attach the tips of the ends of the shoelace together. So the whole thing is a loop, but with knots stuck in it. And, you know, when you have a knot in a loop, you can't get rid of the knot by just wiggling and pulling on the string.

13:01The knot is firmly in there. You have to somehow break the string. But that's because knotted curves live in three-dimensional space. If you consider a knotted curve in a four-dimensional space, it turns out every knot can be untied. without doing any breaking of the loop. And here's how we can picture it. If you think of the fourth dimension as color. Okay. So like the knot could live, right now the knot lives in a space that's all colored blue. Okay. So the blue coordinate is not doing anything. Everything is blue. They're all at the same blue location in 3D spatial space. Right. Now, if when I push one strand of the knot next to another strand, I can let it pass through itself by getting a little bit pink at the moment where it's crossing the blue strand.

13:45So now, even though it looks like the strands are on top of each other, they're not because one of the strands is blue at that point and the other is pink. So they differ in the fourth dimension. So this fourth dimension is color and blue and pink are different locations in that dimension. They're different locations in the fourth dimension. And so the way like two dimensional planes can pass by each other in 3D space, they're in different locations. Now you're saying two three-dimensional strings can pass by each other if they have different colors. Right. Another way of saying it, this idea of like lifting off into another dimension to do something that would otherwise be impossible.

14:21Maybe a simpler example would be, imagine two rubber bands on your kitchen table. So just put a rubber band, a little one, and now take a big one, like the kind they give you at the supermarket for your blueberries so that the pack doesn't open up when you're taking it home. But it still does anyway? It still does anyway. Okay, but this is the ideal world of math. Wait, but no, you're a dirty applied mathematician. You should have a solution to that problem, Stephen. But the thing I wanted you to picture was a little rubber band on the table, completely engulfed by a big loop. Yeah. And now if you ask, can I get the little loop outside of the big loop while staying in two dimensions the whole time, staying on the tabletop?

15:06You can't do it, right? You could see the little loop has to bang into the big loop. But if you lift it up into the third dimension, it's lifted off the table. Of course, you can easily get it out. It's not trapped. So this is a common move in math that by going to an extra dimension, you could, you give things more room to play. And so anyway, but back to your question, Daniel, you said, how do I picture multiple dimensions? No, I think that's really fascinating. I'm giving you a flavor. Like we can use other coordinates, color, time. And you're doing this maneuver, which I think is really fascinating, which is you saying, look, I can't think in four dimensions.

15:39Let me add a new property to things in three dimensions, which represents sort of a fourth dimension. That's right. And it reminds me of what we do in general relativity, where we talk about spatial curvature, and we're talking about three-dimensional space or four-dimensional space-time. And we say it curves, and people are always asking, what does it curve into? And the answer is, well, it's an intrinsic curvature. We're adding a feature to every point in space, this metric, which tells us how it curves. and you're doing this maneuver, which I think is implicitly arguing that every intrinsic curvature is equivalent to like adding another dimension, right?

16:17Something like that. You can think of it as an extrinsic curvature instead. Yeah. It comes up a lot in modern applications of math where you may have heard people talk about features. Like if I wanted to describe you, I might talk about your properties as a person, you know, maybe where you were born, how tall you are, whether you have glasses or not, do you have a beard? What's your race? I mean, you could kind of specify a person. The fact that I live in California and not Virginia. All of that. That's unfortunate, yeah. But we can give a lot of different features that characterize a person.

16:55Now, how many would it take to uniquely specify you? You know, maybe I would need a hundred, a million, I don't know. But it's the point being that to describe something, you can give, in the jargon, we would call it a feature vector, right? A list of these numbers, a list of these properties, because a vector is a good way of bundling many properties or numbers together that all describe a single thing, in this case, you. So we're just used to this. And so it's a way of taking things in the real world, like a person's demographic information, and thinking of it as a big bag of numbers. Mm-hmm.

17:29Mm-hmm. So I'm excited that you specified that you are an applied mathematician because now I don't have to feel any shame when I ask this question. So things like topography and talking about, like, how do you untie a knot, is that just, like, a lot of fun to think about? Because, like, it sounds fun to think about. Or are there, like, applied implications of this ability to untie knots if you add dimensions? The untying of knots, if we add dimensions, I don't know that I've ever seen an application of that scientifically. But certainly topology itself and knot theory is really practical in mathematical biology.

18:08So here's an example. DNA, everyone famously, you know, it's certainly been described in all of our school textbooks as a double helix, right? Those are geometric properties. To say a helix, that's a curve that looks sort of like that spiral that goes around a barbershop pole. A helical shape. Each of the two strands of DNA is a helix. And they wind around each other many times in a real molecule. Like if it was a DNA from a virus, it could have 500 or 1 ,000 turns of the DNA in the double helix. And frequently, the double helix doesn't have loose ends. It's not like an open piece of string. It itself closes into a loop.

18:51So now you have these two curves that are doing helical turning, but they're circles, so to speak. I mean, if you see, I'm indicating here, listeners who are just doing this audio will have to picture me making my fingers in a swirly motion, drawing a helix. and sort of looks like if you took your two fingers and made a chain link, you know, like put your thumb to your index finger and make a circle, do that with your other thumb and index finger, and now link those two circles like links of a chain. Okay, that's in the jargon, we would say that's a pair of curves with a linking number of one.

19:28And you could instead go through the circle many times, you could go 500 or 1000 times, then you'd have a linking number of 1000. And why that's important is because that's real. That's what happens in viral DNA. It happens in our own DNA and our cells. We have these closed circular DNA molecules that are topologically very interesting. And the reason I'm going on this long rant is sometimes they have knots tied in them. So you're asking about knot theory and knotted loops. Nature already knows about this. We have knotted DNA. And we also have special enzymes, molecules called topoisomerases. is topo for topology, isomer for a different version of the same molecule.

20:11So these topoisomerases can actually do things like snip DNA like a scissors and then move a strand through like I did with blue and pink, except they do it very brutally by cutting the DNA, moving the strand through, and then resealing the DNA. And so where knot theory gets used in biology is sometimes an enzyme's action is unknown. The enzyme is a little machine. What is it doing to the DNA exactly? I gave you a picture with scissors and closing. But what's really happening? I mean, the enzyme is a protein. It's not literally a machine. I mean, it's a protein that acts like a machine. So people have used knot theory to infer enzyme action by looking at the series of topological products that are made.

20:56If you let the enzyme work on a DNA molecule longer and longer, it produces more and more elaborate knots. Yeah, okay, so you get the idea. By looking at the knots, you end up learning what the enzyme is doing. Very cool. Thanks. Very cool. Let's get back to your book, The Big Math. And, I mean, we all know that math underlies physics and everything else is just stamp collecting. But I think that in your book, you're making a different point, right? Not that math is the language of science, but the nature of these ideas, these mathematical concepts, is changing our world because of how it powers our science.

21:31And I was interested in what you said earlier about features and vectors and lists of numbers that we connect together. Can you tell us about why we need things like vectors and matrices and why that's important for engineering projects like building skyscrapers? Sure. It's a great question because vectors, if people have run into them, they may have learned about them in school as little arrows. We certainly use that picture in physics when we talk about vectors around, you know, in the magnetic field or the electric field. So they can be visualized as arrows. They could represent forces, velocities, that sort of thing.

22:06But in big math, the vectors, instead of having two or three components, might have, for reasons I said about features, They could have hundreds or trillions of components. So that's pretty common. And, well, why would we need that? Here's an example that's big but not so big. And it's also not so modern. Think about the moon landing. You know, here we are. We have Buzz Aldrin and Neil Armstrong in the lunar module. It's the year 1969, July, if I remember right. I was 10 years old. They're trying to land on the surface of the moon. And this is this incredible, daring feat of engineering, of aviation know-how, of courage.

22:49And inside the lunar module, which was that little spidery lander that's going to land on the surface after it detaches from the main spaceship, from the command module, the lunar module with these two astronauts in it has inside of it a tiny computer, the world's smallest computer up till that point. I mean, this is an era, remember, when computers used to occupy a whole room. Right. You know, to have a computer that could fit inside a suitcase, which is what the NASA instrumentation engineers developed. They invented personal computers before we had personal computers by like 10 years at least.

23:27And it was totally an incredible innovation and also very high stakes because this onboard guidance computer had to steer a spacecraft through the void with two lives on the line very far from home. I mean, it couldn't be controlled by, you know, mission control. It would take too long to get the signals up to that computer. Anyway, so you have this little onboard computer. And what's it doing? It's trying to keep track of the dynamical state of the lunar module. Now, what that means is this little spacecraft has three numbers that are telling you where it is in space. How far above the moon's surface is it?

24:09That's one number. How far east and west or north and south? That's two more numbers. So it has three numbers for position. It has three numbers for velocity. How fast is it moving in each of its three directions? It has three more numbers for how fast is it spinning and around what axis. I mean, you can see, just to describe the state of the lunar module, I've already listed nine numbers. There's also thrust. There's how much fuel is left. There's just a lot of things to keep track of. No human being could do this, right? I mean, that's already too many numbers to keep in our head. So this little, beautiful little computer is calculating every two seconds.

24:46It's not as fast as our modern computers. This thing has less computational power than your watch, your digital watch. Yeah. Anyway, the numbers I'm listing that describe the condition of the spacecraft, this is the state vector. To an engineer, an aeronautical engineer, to characterize the lunar module, you need to know its state vector and how its state vector is changing moment to moment. And this is what that little machine was doing quietly and indispensably. I mean, it's doing all the math to get these two astronauts on the moon safely. And I guess the problem is hard because to decide what to do next, you need to consider all nine numbers at the same time, right?

25:26And they all contribute. And so you can think of it as how you're transforming your vector of numbers into a new vector you're hoping to have in the future. And I guess you're relating those two vectors with a matrix or something. That's right. That's right. Thank you. Good. Exactly. It comes out to be something close to 20 numbers when you list them all. And those 20 numbers have to be updated moment to moment. And if they are off, which they can be because the spacecraft isn't always exactly where you meant it to be. Right. Then the sensors have to tell the craft where it is. Then the computer has to tell the thrusters whether to fire thrust or fire less thrust.

26:05Like if you do too much, you may overshoot the landing spot. Also, if you make too much of a correction, there's something called control theory involved here. This is like cruise control in your car. You know, like you're going up a hill, the cruise control will push on the gas to make you have more acceleration because now you're going uphill. It's harder. There are similar kinds of cruise control automatic steering mechanisms that this computer was in charge of. And if it updates too fast or if it overcorrects, the craft will start to wobble. If you don't correct enough, you could land in the wrong place.

26:39And this became very serious in the actual moon landing because Neil Armstrong could see when they were about 500 feet above the surface that they were going to hit a boulder patch. The moon is rocky as hell. And you can listen to this on YouTube. It'll stop your breath. I mean, and it literally got the mission control engineers after the famous, you know, the Eagle has landed. They said they radioed back to Neil Armstrong, you got a bunch of guys about to turn blue here. Because the radio went quiet. In fact, what happened was when Neil Armstrong saw that they were going to hit the rocks, he had to override the guidance computer because the computer didn't have any vision system.

27:23This is 1969. They didn't have good computer vision at that time. So Neil, being the fantastic pilot that he was, took control and steered the thing down. It's getting me emotional just thinking about it. It was such a fantastic combination of human daring and ability. The right stuff, as they say in the book and in the movie. And his heart was beating at about 150 beats a minute at that point. NASA was recording the whole thing. So there's a lot of math in the moon landing. But he kept his cool and he got it down. So that just, you know, right before the Eagle has landed, he says, it looks pretty rocky.

28:03Oh, my gosh. Yeah. So I guess your point is that we couldn't have gone to the moon without big math that was bigger than what we could do ourselves. Exactly. And also we still needed humans in the loop in order to land it, literally. We did. It's sort of a metaphor for our time, isn't it, that we're now seeing these giant autonomous systems which can make decisions about what mortgages we should or shouldn't be allowed to get, whether – maybe too grim to bring up, but let's be real. What targets should the U.S. military strike on the first day of the war in Iran? Unfamously or infamously, notoriously, they blew up a school with hundreds of little children.

28:46So some human who was in the loop got it wrong. Or maybe the AI got it wrong. We're still, you know, Congress is investigating what happened on that day. But it just shows that, you know, this is a very complex territory we're entering now where we're making a lot of serious decisions with the help of AI or with the help of our machines. And we sometimes get it wrong. This is – the real world is difficult. High stakes. High stakes, to say the least. Yep. Absolutely. All right. Well, let's take a break. And when we come back, we'll talk more about using matrices to build skyscrapers, talk about network theory, optimization, and all sorts of other fun stories from Stephen's great book, Big Math.

29:39Hey, it's Kelly Rowland. You may not know this, but I have eczema, so I get how it can steal your time. Why let eczema take over? Flare less with EbGliss. About 4 in 10 people taking EbGliss achieved clear or almost clear skin at 16 weeks, and most of these people maintained clearer skin at one year. EbGliss, Lubrikizumab LBKZ, a 250 milligram per two milliliter injection, is a prescription RX medicine for people age 12 and up who weigh at least 88 pounds with moderate to severe eczema, atopic dermatitis, that is not well controlled with RX topical therapies or who cannot use those therapies. Don't use if you are allergic to EpGliss.

30:15Allergic reactions can occur that can be severe. Eye problems can occur. Tell your doctor if you have new or worsening eye problems. You should not receive a live vaccine when treated with EpGliss. Before starting EpGliss, tell your doctor if you have a parasitic infection. Paid partnership with Lilly. Respect your time. Ask your doctor about EpGliss and visit epgliss.com or call 1-800-LILLY-RX or 1-800-545-5979. Morning, Carmen. Hey, Carmen. Morning. Quick question. Just one second. Sorry. Carmen's worked at this office for 20 years. She knows everything, which is why Carmen has to help with, well, everything.

30:48Hey, Carmen, what did Faye pitch in 2023? Who owns the Nordic account? Oh, uh, where's the printer? But Carmen's revenue report is due in five minutes, so she shows them WorkIQ. Hey, Copilot, who owns the Nordic account? And what did we pitch them in 2023? Get the context you need with Microsoft Copilot. Work lives on. Having peace of mind for your home and connected life is more important than ever. That's why Xfinity Shield has built-in Wi-Fi features like Xfinity Cybersecure, which provides protection against online threats and risky sites for every device. Plus, you can lock in one Wi-Fi price for five years.

31:24Guaranteed. It's added peace of mind so you can stream more of your favorite podcasts on iHeartRadio. Get it now at Xfinity.com slash Shield. Xfinity. Imagine that. Restrictions apply. Not available in all areas. Five-year guarantee with select plans only. Xfinity Gateway required. Buick knows confidence changes how you move through the world. That's why Buick designs SUVs with striking style and was named the most dependable mass market brand by J.D. Power. From the head-turning Buick and Vista to the seven-passenger enclave, experience the Buick SUV lineup at Buick.com. Buick, exceptional by design.

31:59Buick received the fewest reported problems in its segment in the J.D. Power 2025-2026 U.S. Vehicle Dependability Studies based on 2022-2023 models. For J.D. Power 2026 award information, visit jdpower.com slash awards.

32:19Okay, we're back, and we're talking to Stephen Strogatz, author of Big Math and professor of mathematics, dirty mathematics, apparently, at Cornell. Stephen, you were telling us why these problems are too big for us to solve in our heads, and we need computational support. I really enjoy the description of building the Burj Khalifa in your book. Can you tell us a little bit about why vectors and matrices are so important for solving such a big problem like the biggest building on Earth? Yeah, no, thanks a lot, Daniel, for the kind words about the book. The Burj Khalifa is the world's tallest building.

32:55It's a skyscraper that I think, if I'm remembering right, is more than twice as tall as the Empire State Building. It's astounding. Yeah. Yeah. How crazy is that, right? This is in Dubai. Yeah, it's in Dubai. It's, well, it's on ground that is not very favorable for building such a tall skyscraper. This is built in the desert of Dubai. And to get a building that tall on such a loose foundation required very heroic engineering innovations. And so the team that built it, rather than just build it and see if it's going to work. One approach. Yeah, okay. Which is one approach. You can see where computers would be helpful here.

33:41I mean, you could build a scale model. You could build a small one and see if everything looks good. And that's one option. Classical engineering does do things like that. But you can also build it in the computer. You can build a virtual Burj Khalifa following all the right geometry with all the girders in the right place. And so they did do that. They made a very, very intricate model of the structure inside a computer, thinking about all the forces on the girders, on the steel and concrete. So to describe all the forces, all the stresses, all the deformations, the strains, the twisting, to calculate all the properties of the building, you have to describe, like, how much are the different girders deforming?

34:26How much are they stretching or twisting? That's a number. And then how much force will result or how much torque in the girder as a result of all that? So there's a lot, a lot of numbers. It's into the billions of numbers that are being calculated to see, is this building going to be safe? Is it going to stand up? How much is it going to sway in the wind? You have to keep in mind that this in the desert, at this height, twice the height of the Empire State Building, the wind can get pretty serious. They have windstorms. They have dust storms. So all of that has to be taken into account. And it was in this very, very impressive bit of engineering that, you know, when you're in the building or looking at it, you don't think about how much math went into making it work.

35:11Do you know off the top of your head, this is maybe not a fair question, how much does it like wibble back and forth at the top? Because, you know, it needs to be able to move in response to the wind so it doesn't crack. It does move. But I feel like I would puke. So how much does it move up at the top? Do you know? I don't happen to know the number offhand. It's a good question because it is true that very, very tall buildings, these super skyscrapers, are flexible to some extent. And they do wiggle at the top a little bit. And in fact, sometimes we'll install so-called vibration absorbers at the top, which are giant masses that are sort of counter motions.

35:50So as the building moves, the vibration absorber might move the other direction to try to stabilize it. I don't know if the Burj Khalifa has a big vibration absorber feedback system in the top, but it does have a very special triple buttress design. If you look at it, it has sort of a triangular cross section. So it has these three long posts going way, way up. And it also has a funny cross section that changes as you move up the structure, which was designed to, as the engineers put it, confuse the wind. Instead of keeping a constant shape, the cross-section is changing in a subtle way as it moves so that the wind doesn't have a sort of constant target to start creating vortices and vibrations.

36:37We'll have to ask the wind on the podcast and see if it felt confused. Yeah, confuse the wind. What a great phrase. So we want to ask you later on about math's impact on AI. But first, I want to ask you more about the mathematical underpinnings of AI and machine learning, which I think are not widely enough appreciated. Can you walk us through why matrices are really important for computer graphics and how that ended up having GPUs power modern machine learning? Like, what is the connection between matrices, computer graphics, and machine learning? in the end. Okay. Yes, good. You have asked me a couple times about matrices, and I keep forgetting to say the word matrix.

37:21So, let's talk about matrices for a second, just quickly to go back to the Burj Khalifa, and then I'll try to get to your question about computer graphics and machine learning. So, a matrix, we can also think of it as a sort of machine. It's a mathematical function, really. That's the role it tends to play. It can play many roles, but one of its roles is as a machine that takes an input, which is a vector, a big list of numbers, which in the case of the Burj Khalifa might be the placements of all the girders or the displacements, how much they're stretching. And then it gives an output as a result of all that input.

37:57And the output would be what's the resultant stress in the girder. And often the relation between input and output follows a straight line. So there's a linear proportionality between, like people who took elementary physics will know this. We have the so-called Hooke's law that if you pull on a simple spring twice as far, it pulls back twice as hard. And that's a linear relationship if force is proportional to displacement. Anyone who's done archery knows this too. You know, you're pulling back on a bow string, it gets harder the farther you pull it back, harder to pull it, takes more force. So this linear relationship, when you have a linear relationship between input and output and you're dealing with many variables, that's what a matrix does for you.

38:42It's the machine that can handle these linear input-output relationships. Okay. So we see that in many places in science and technology, not just in properties of materials like steel, but in computer graphics. We have to think about – like there, the vectors might literally be a spin vector. Like picture Simone Biles, okay, the great athlete at the Olympics. She's doing her gymnastics routine. She's running. She jumps. She's tumbling in the air. To describe the state of a spinning acrobat, you need a vector that we call the angular velocity vector or the angular momentum vector. You need to know her position.

39:23All of that stuff is, again, vectors. And matrices govern rotational dynamics. That's something you probably teach, Daniel, in your junior-level mechanics course or something like that or sophomore level. So rotational dynamics is an old subject. It goes back to Euler. He wouldn't have been using matrices exactly. He would have written out equations one at a time. But now we do it with matrix notation. It's more compact. And actually, even before we used vectors, there was a fancier thing called quaternions. Quaternions were invented as a generalization of complex numbers to try to discuss instead of rotations in two dimensions, rotations in three.

40:05And quaternions are still being used today. They've been resuscitated in computer graphics. They're very fast to compute with quaternions when you want to show an image of, you know, maybe it's a video game and something's – maybe you've blown up a Jeep. It's a war game video and you blew something up and the Jeep is flying through the air and spinning. Some computer programmer somewhere is using quaternions or matrices to give a very fast visual update of this spinning, exploding Jeep. Are quaternions and matrices the same thing, or are they different ways of dealing with lots of information?

40:45I would say they're very closely related. Quaternions are historically older. They went out of fashion for a long time because they only work for describing three-dimensional motions. They don't generalize to higher dimensions. Whereas matrices can work in any number of dimensions with any number of variables. So matrices, which are also a product of the 1800s, came a little bit later than quaternions, and they're more powerful. We use them in quantum mechanics. We use them in, I mean, I guess I should say, I earlier said that a matrix you could think of as a big, it encodes an input-output relationship that's linear.

41:26But nowadays, like in large language models, which I think was maybe the second part of your question there, Daniel, you could use a matrix in various other ways. Like, suppose I want to say which artificial neurons are connected to which in the network. That's a very important thing, the wiring diagram of the neural network. whether it's a real neural network like in a brain of a person or the neural network of a tiny worm, C. elegans, you know, we have these networks of which neurons are connected to which. That description as a math issue, how do you describe a network with formulas? Well, matrices can do that too.

42:05A matrix can tell you who's connected to who. It's a big table of numbers in that case, and it can show who's connected and how strongly are they connected. And is it a positive connection? Like if it's neurons, does it excite the neuron downstream or does it inhibit it? You can put a plus sign or a minus sign. So matrices are tremendous, like maybe the most powerful idea in big math. In fact, we toyed, Alex Townsend and I toyed with writing a book that was just going to be called Matrix. Because the matrix, if there's a star of the show mathematically, it's probably the concept of the matrix.

42:39It can do so many things for us. It's literally a trillion-dollar idea. I mean, NVIDIA's entire company worth is on the power of specialized chips that manipulate matrices, which started out being really good for computer graphics until people realized, oh, the same chip is really good for machine learning because. I see. I see. That's what you were getting me to. I didn't take your hook, the lead debate. Yeah. No, absolutely. NVIDIA's chips were designed for doing very fast rendering in video games. Yeah. And they turned out to be exactly what was needed for the training of large language models and other neural network models.

43:18That's the thing. You know, when they learn from data, as they do when they're, say, learning to do Google Translate, to do very fast language translation of any language on Earth practically into any other language, how does the machine learn to do that? At first, it's not good. But as it makes mistakes, the errors that it makes get propagated back into its neural network through a process that we nowadays call backpropagation, which ends up updating a certain matrix or a set of matrices. And so learning to a machine is a matter of updating matrices. I thought that was a really fascinating development when they developed this technology for gaming, and then it turned out to be also really useful for machine learning and Bitcoin farming.

44:05And my son was a teenager at the time, and he was really frustrated by that because he's like, GPUs are really expensive now, and I just want it for my Fortnite rendering or whatever. But all these guys are buying them for machine learning and Bitcoin farming, and so now they're crazy expensive. I remember he spent like$1 ,000 on a GPU, and then two years later, he sold it for more money than he bought it for, which is unheard of. In computer world, usually things get cheaper exponentially, but GPUs, even outdated ones, were more expensive than they were originally. It's crazy, but it's the power of this matrix, right?

44:43They built a general matrix processing device, which turned out to be relevant for lots and lots of things. It's an amazing, unexpected windfall for them. Well, I feel like I want to mention a couple things, but I can also just let you ask a question. But since, I mean, the subject of matrices is so rich. Okay. Well, let's see if I can do it. First of all, the word. You know, the word sounds like such quintessential science fiction stuff, like the Matrix movies. Yeah. But the word is actually a beautiful old word that goes back to the word for mother. Really? Really. I mean, it's no accident that Matrix sounds like maternal or matriarchy.

45:17Oh, my gosh. What is a Matrix? We talk about a fiberglass matrix, you know, in composite materials. A matrix is a kind of environment that can give birth, literally birth, to new possibilities. Wow. A great mathematician named James Joseph Sylvester in the mid-1800s gave us the word matrix in its mathematical sense as something which in his case was giving birth to determinants, which are a certain mathematical object that we use for solving systems of linear equations. Back in those days, we don't use them for that anymore. But anyway, he was really interested in matrices as a kind of womb, like a womb that could give birth to something powerful.

46:02And so, yeah, it comes from the same word as mother or womb. That's amusing thing number one about matrix. Amusing thing number two, you mentioned money and how NVIDIA is such a profitable, you know, a trillion dollar company. So was Google in its day. And what made Google rich was that if you ask how our web pages connected, right, that was the old problem in 1995, let's say, when the World Wide Web was new. It was very hard for finding whatever you wanted on the internet. Searching was a real serious problem. And Google, Larry Page and Sergey Brin solved it By thinking about matrices, putting a number that tells you which web pages are connected to which and essentially rendering the graph of the searchable internet, the network, as a big matrix.

46:54And so certain properties of that matrix would then tell them when you do a search, which page should the machine show you? What's the right return for your query? Like my kids talk about searching it up. How does Google know what to show you? That's like a trillion-dollar calculation with matrices. That's what made Google rich to begin with. And is that still how they do it? They still use PageRank, it's called, named actually after Larry Page, not after WebPage. Oh, interesting. Yeah. So PageRank was the first great breakthrough that led to Google, but a lot has been done since then. So it's not just the matrix theory.

47:31There are many wrinkles now with neural networks and other things. So matrices gave birth to great fortunes, it sounds like. They did. Wow. They absolutely did. Let me follow up on that and zoom out a little bit. Why does developing a new mathematical structure like a network or a matrix let us do things we couldn't do before? Does it open new territory or is it just like a more efficient way to explore something we could have explored more clumsily? You know, like could we have done all this science without graph theory, without matrices, or would we have been like implicitly inventing it along the way?

48:07you know, like the way Newton invented calculus to get his physics to work. I think psychologically, these new concepts ease our cognitive burden. It makes it easier for human beings to think about when we have the right notation, the right concepts. It helps us, again, I come back to the inevitable human limits, you know, to err as human. We're not perfect computers. You know, you've probably had guests on talking about probability theory and how bad we are at assessing risk and how easily we're confused by conditional probability puzzles and that sort of thing. So we're quite imperfect. People who, for example, are happy to drive across country because they're scared of flying.

48:49Even though driving is much more dangerous, yeah. That kind of thing. We misestimate risk all the time. Yeah. So I think maybe because the concepts like matrices, like vectors, they give a way of packaging complexity into more manageable forms, easier for us to hold in our head. That's why these things developed historically. So vectors and matrices are from, in some version, from the 1800s. But actually, there are precursors of matrices as long as like 200 BC, 2 ,000 years ago in ancient China. The first matrix calculations were word problems in algebra books. They didn't call it algebra. That's an Arabic word from a millennium later.

49:29But the Chinese were doing problems that involved word problems about if I give you this kind of rice patty and then I give you a lower grade of rice patty and I'm trying to make edible rice and this number – it sounds like a high school algebra problem. It really does, except they were doing it in 200 BC. And the way they would solve these problems is they made a physical device, a board called a fang cheng, that is a wooden board with grid lines drawn on it. And they would put numbers in each cell of the grid. It looks like a spreadsheet, except it's a real piece of wood and it has numbers which are sticks.

50:10Instead of abstract one, two, three, they would literally put three sticks or five sticks or they had their own notation. And then a skilled calculator person would manipulate the sticks on the board and do what in modern language would be called Gaussian elimination or row reduction. For someone who's taken linear algebra, you'll know that this is how we solve linear equations today. They were doing it with sticks and wooden boards in 200 BC. and not just for two equations with two unknowns like we teach in high school. They were doing five equations and five unknowns, and they saw the pattern.

50:47They knew that they could do trillions. They just didn't have any reason to do trillions. But the ideas are very ancient. That's basically an analog computer, right? I mean, they build a physical device that implemented a mathematical algorithm. It's a computer because instead of moving electrons in a circuit, they're moving sticks on a board. Oh, my gosh. But it's the same thing. That's amazing. Conceptually. Yeah. I love that we have this connection with kids from thousands of years ago, that we've been torturing them with word problems for thousands of years. But let's take a break. And when we get back, we'll talk about, unfortunately, how we're torturing mathematicians with AI.

51:33Hey, it's Kelly Rowland. You may not know this, but I have eczema. So I get how it can steal your time. Why let eczema take over? Flare less with EbGliss. About 4 in 10 people taking EbGliss achieved clear or almost clear skin at 16 weeks, and most of these people maintained clearer skin at one year. EbGliss, Lubrikizumab LBKZ, a 250 milligram per 2 milliliter injection, is a prescription RX medicine for people age 12 and up who weigh at least 88 pounds with moderate to severe eczema, atopic dermatitis, that is not well controlled with RX topical therapies or who cannot use those therapies. Don't use if you are allergic to EpGliss.

52:09Allergic reactions can occur that can be severe. Eye problems can occur. Tell your doctor if you have new or worsening eye problems. You should not receive a live vaccine when treated with EpGliss. Before starting EpGliss, tell your doctor if you have a parasitic infection. Paid partnership with Lilly. Respect your time. Ask your doctor about EpGliss and visit epgliss.com. Or call 1-800-LILLY-RX or 1-800-545-5979. Morning, Carmen. Hey, Carmen. Morning. Quick question. Just one second. Sorry. Carmen's worked at this office for 20 years. She knows everything, which is why Carmen has to help with, well, everything.

52:42Hey, Carmen, what did Faye pitch in 2023? Who owns the Nordic account? Oh, uh, where's the printer? But Carmen's revenue report is due in five minutes, so she shows them work IQ. Hey, Copilot, who owns the Nordic account? And what did we pitch them in 2023? Get the context you need with Microsoft Copilot. Work lives on. Having peace of mind for your home and connected life is more important than ever. That's why Xfinity Shield has built-in Wi-Fi features like Xfinity Cybersecure, which provides protection against online threats and risky sites for every device. Plus, you can lock in one Wi-Fi price for five years.

53:18Guaranteed. It's added peace of mind so you can stream more of your favorite podcasts on iHeartRadio. Get it now at Xfinity.com slash Shield. Xfinity. Imagine that. Restrictions apply. Not available in all areas. Five-year guarantee with select plans only. Xfinity Gateway required. Buick knows confidence changes how you move through the world. That's why Buick designs SUVs with striking style and was named the most dependable mass market brand by J.D. Power. From the head-turning Buick and Vista to the seven-passenger enclave, experience the Buick SUV lineup at Buick.com. Buick, exceptional by design.

53:53Buick received the fewest reported problems in its segment in the J.D. Power 2025-2026 U.S. Vehicle Dependability Studies based on 2022-2023 models. For J.D. Power 2026 award information, visit jdpower.com slash awards.

54:14All right, we're back. And Kelly has projected her negative feelings about algebra onto people she's never met from thousands of years ago. How do you know, Kelly, they didn't love it? Maybe it was their favorite thing. Maybe they had the same trajectory I did, which is to say they found it very frustrating at first, but once it clicked, it was a ton of fun. It just, you needed to have that moment. But until you got that moment, it was, you know, maybe a little bit tear inducing. All right. So Stephen, we wanted to ask you your thoughts about how math is changing in the new world of AI. And I want to ask you about how math has sort of organized itself.

54:49We've seen AI become very powerful at one specific thing, proving theorems, and that sort of upset the math ecosystem. I've heard many mathematicians insist math wasn't about proving theorems actually, but about the insight and the tools along the way. And so is this an issue because math used the wrong metric, you know, rewarding theorem provers and giving them faculty positions? Why didn't math initially optimize for the thing they actually cared about and reward insight instead of theorem proving? That's a really interesting question about the sociology of math that I don't know that I have a great answer for you.

55:26To me, that argument is a historical anomaly. That is a very short-sighted argument because this reward system where we have these well-formulated, universally agreed upon open problems, the lighthouses that we all aspire to, or in the other metaphor, the mountains that we all want to climb, the destinations for the journey. That is a new thing. And it's not universal across all of math. Here's what I mean by it's new. Did Archimedes have a list of problems that everybody agreed on? Did Isaac Newton 2 ,000 years later than Archimedes, did he have a list? How about Euler after Newton? I mean, this is a very 20th century thing to say that it's around the time that pure math split off from applied math.

56:17Euler, as you know, was fantastic in fluid dynamics. I mentioned earlier rotational dynamics. That's Euler. Euler was a magnificent physicist as well as a magnificent mathematician. He was a scholar. We didn't make that distinction. Same thing with Isaac Newton. Same with Archimedes. You know, this is a very artificial 20th century split in math between pure and applied. So even though I've reinforced it by saying we have these two cultures, We didn't used to have them. We used to have people who like to look at the world mathematically. Sometimes they look inside at math. Sometimes they look outside at the world.

56:52We did both. Okay. So, but during the time of the split, the pure mathematicians, for some reason, since they weren't – oh, God, I'm going to get in trouble for saying it like this, but I'll just finish the sentence. Since the pure mathematicians had no impact on the world immediately, they created this fake system of rewards. Okay, I'm really in trouble now. Not fake. They tried to quantify who was doing important things. What were the targets? If it wasn't to do something in the world, it was to solve intrinsic math problems. So they all agreed on what was important. I have always not liked this system because I want to invent my own problem.

57:31I don't want to do a problem because somebody else told me this is the problem. It feels very authoritarian to me. I don't like it. I want to choose my own problem. I think that's part of the creativity. Okay, that's a personal thing. But yes, if we destroy that system, if the AI can solve all these agreed upon problems, is that a big loss? Personally, I don't think it's such a big loss. I think it might be good to get that done with. and now let's all think of our own problems and spend the rest of the time persuading each other that the thing I chose is a cool problem and here's why. And then AI will solve it.

58:07With your help, it's your AI, okay? You can point your AI at it, you solve it, and your job is to be, as I was saying to my friend Grant Sanderson on his podcast recently, maybe the value system in math will change to what the kids call Riz, charisma. your unique brand. Okay. So here again, I'm, this is an old man. I'm 67. My 24 year old and 26 year old daughter taught me about Riz. And I think I have misunderstood it because Riz apparently means Daniel, you have, you said, sounds like you have a young son. Like Riz is about how you get the opposite sex interested in you because you're so charming.

58:46Oh, I'm going to go Riz Sophie up right now. And then she's going to be all over me. That's what you're supposed. That's what Riz really means. And it usually doesn't involve math. Yeah, it's not about math normally. Math is not normally what your Riz is. Okay, but I want to generalize that in the style of a mathematician. I want mathematicians to show their own unique, okay, maybe Riz isn't the word, brand, their own taste, their own style. Like I think I want the humanity of the person to come to the fore, like what has mattered to them in their life. Sort of like if Vladimir Horowitz plays the Chopin Nocturnes and Arthur Rubinstein plays the Chopin Nocturnes, they don't play them the same way.

59:30They're both great maestros, great masters and artists, but they bring themselves something about their life to their playing. And I think that's what's going to come to the fore now that the machines can solve a lot of the problems for us. We get to put our unique individuality into the questions we choose and how we approach them, and also where we point the AI. I totally agree with you. And I think one productive response to this new world is to say, hey, we have new tools. And what new questions can we ask? What are interesting things to do now that we have these new tools? The way that, like, telescopes changed astronomy and digital computers changed mathematics and physics.

1:00:09I feel like that's sort of the future. But I was reading comments you made at the 2026 ICM, the Congress about math. And I thought I understood that you were suggesting that AI could become better than humans at asking interesting questions. Because a moment ago we were saying AI is a tool. It still resides within the human to be interesting, to be creative, to be clever, to have this riz, this research taste. But you're suggesting that AI could do that part as well. But why would AI ask questions, right? humans have curiosity. They have appeal of something, but isn't there always a human that's like programmed the AI, that's built it?

1:00:49I guess my question is, how does AI come to ask interesting questions? Thanks for calling me on this. You're bringing up something personal for me and very real, which is I change my mind about this every day. Amazing. Okay. So when I was recording from the International Congress of Mathematicians, whenever that was, Maybe it was July. I mean, that's like the dark ages in recent times. So much has changed since then. And today, my feeling is we humans have a window of opportunity here where we get to ask the good questions. Right now, AI is not good at asking questions. I do believe in a science fiction sense or maybe a two-month-from-now sense we'll find out that AI can be trained to ask good questions too.

1:01:38If I work with my graduate students, one of the hardest things for them that's so different from being an undergraduate is that now the problems are not given to them. As an undergraduate, you're told what to do. You have homework. You do your homework. You learn quickly that to be a researcher, nobody's giving you homework. You have to choose your path. And that's hard. That's a whole new game. So to be a good advisor, right, you guys get it, right? I mean, you're having to do that with your own students. And you did it yourself. your mentors did it with you. How do you decide what to work on?

1:02:10That's a very delicate, artistic, creative thing that's hard to, it's not in books. It's not easy to teach. But to the extent that we can teach people like our grad students how to do it, I don't see why we couldn't also teach machines to do it. And whether they would feel curious, that's a deeper question about consciousness or feelings, but they don't have to feel it. They might be able to do it without feeling it. So, but we don't know. Maybe there will be some unique refuge for humanity where, personally, if you really push me, I don't think I change from day to day on this. I do believe they will be better than us at every aspect of math soon.

1:02:51I really do. I think they're already better than us at problem solving. I think they will soon be better than us at exposition, at teaching, at asking questions, all of it. I do believe that, and I bet it will be done soon. All right. Well, you've perfectly set me up for the next question, because here's a test of whether AI can ask interesting questions. Okay. I have access to a frontier AI model, and I asked it what question it would like to ask you on the podcast. Oh, good one. Here's a question from AI about math. Yeah, which one? I like to know who I'm talking to. Yeah, so this is ChatGPT6 Pro.

1:03:29This is ChatGPT. Hey, my old friend. So it says, suppose I, ChatGPT, became extremely good at giving you explanations that make you say, aha. How would you test whether I had increased your understanding or merely learned what gives you the feeling of understanding? That's ChatGPT's question for you, Stephen. How would I be able to test whether the machine had given me understanding or merely made me have a kind of superficial aha moment on the spot? But that seems like a very easy question to me. Uh, because how do I test if anyone has understanding? I'm in the business of doing that. I do that.

1:04:08I can give oral exams. I could, I mean, chat GPT can give me an oral exam. If I understood something, if I really learned, you can test me. Why is this a hard question? Now we have to say, what's a good test? Well, I don't know. I mean, depends what you want to test. If you want to test my ability at problem solving, you can test that. If you want to test my ability to parrot it back to you in my own words. We can do it that way. You know, just think of all the ways that teachers currently test students or an employer tests their employee. I'm surprised that that's the best it could come up with.

1:04:43Now, that is a really interesting example, Daniel. To me, that shows how poor it is. You asked it to ask its best question. I think that's a very poor, easy question. What do you think? No, I agree with you. I mean, I thought it was trying to get a little philosophical and distinguished between the feeling of understanding and deep understanding. But I agree with you that I often have this experience with students. They think they understand something and I can just poke at it from another direction and see if their understanding withstands it or collapses. Exactly. And this for me is probably my biggest weakness as a teacher.

1:05:17If I would just tell the truth here and be honest, it's embarrassing to admit, but it's something that has troubled me for much of my career that I get teaching awards left and right. And yet, I think it's in this superficial sense that we want to avoid. The students are sitting there. They hear me doing my patter. They think this is fun. Oh, I want to learn. This is great. And even they may even have aha moments when I show them a beautiful proof at the board and I explain it clearly. Then I give them a test and like 80 % of the class doesn't understand anything. Right, right. Okay. And same thing, like Feynman famously wrote about this in his Feynman lecture series on physics.

1:06:00That guy gave you more aha moments than just about any physicist ever. I mean, I don't know if you felt that way, but when I read Feynman, I feel like this is so great. This is the right way to look at it. Over and over and over. And he says with some apparent confusion that the students didn't do well on the test. And I think it's because it's easy for a teacher to make the mistake that Feynman made and that I apparently make, which is that real learning happens when the student works. The student has to struggle. We know this. You can't learn by being a spectator and clapping for a beautiful proof.

1:06:36You have to get in there and suffer and make your own proofs. This is the part that where you, Kelly, said, you know, that sometimes math was hateful to you, but then sometimes you had the aha moment or you had a breakthrough, and that's when real learning clicked for you. Yeah, yeah. Yeah. Well, and those moments were worth it. Yeah. They made it worth the suffering. But that's the mystery of education. How can the teacher give, I mean, this is like a John Dewey way of, or a Montessori way of thinking about it, which has its own problems. Education is harder than math or physics. I mean, how do you set someone up to do the hard work of learning, really rewiring their brain or changing their perceptions so that they truly understand in a new way?

1:07:17It's not by lecturing at them. Even a very clear, beautiful lecture doesn't accomplish the right thing. So in contrast to an AI-generated question, we also asked Kelly's husband to ask a question because he has a whole podcast about AI. Oh, good. So Kelly, you want to read Zach's question? Sure. What is identity as a mathematician and do you think AI threatens it? I personally think this is the biggest source of strife and is similar to what's happened with other outmoded communities in the past. Outmoded. Sorry. I didn't say it. I didn't say it. Damn, tell us what you really think. He probably imagined I was going to soften it, but I accidentally just read it out loud.

1:08:01So, Stephen, what is identity as mathematician in the new AI world? That's such a personal question for each mathematician. For some people, it's about community. Even though they have their identity, I know a lot of mathematicians who love the social aspect of collaborating, explaining things to each other at a blackboard or being stuck on a problem together. I think about it with this book that I just had the pleasure of working with Alex Townsend on, that we had such a good time teaching each other, being confused together, laughing when we would figure something out or writing a passage that suddenly started to work.

1:08:39So for some of us, identity is about the communion with other minds and spirits that love the same thing that I love or that we love. Okay. So for some people, that's their identity. And if they weren't part of a community, like if nobody was doing math, no person was doing math anymore. You just had to sit and talk to a screen, sort of like the poor kids during COVID, you know, where they weren't in school, they were just talking to their screen and maybe the screen is explaining very well. it didn't work well for a lot of students. And, okay, so for some mathematicians, it's community. For many mathematicians, and this may be directly response to the question, for many of the most heartbroken mathematicians, math was always about competition.

1:09:27Math was about, I'm the smartest kid in the class, or I can do the hardest problem. Whatever the teacher throws at me, I can do it. I can answer every question in the book. These are the kids that like math team. They want to be on the international mathematical Olympiad team representing the U.S., whatever. It doesn't have to be that lofty. They could just try to be as good as they can be. Maybe they like solving puzzles. You know, like I say competition, like there's something wrong with it. I was, I loved competing. I was on math team. I mean, I liked being smarter than my friends, Bill and George, who always thought they were smarter than me.

1:10:05Dropping names. I hope they're listening. Yeah, Bill and George in seventh and eighth grade. Because George, you know why? Because George laughed at me one time when he said, what's six divided by zero? And I said, zero. I'd never thought about dividing by zero before. He said, you idiot, that can't be right. Zero times zero is not six. And I thought, oh, whoa, that's right. Wait, six divided by zero. What is that? So anyway, he was just laughing his head off that I didn't know about. You can't divide by zero. And so that irritated me as a 12-year-old. And actually, it was very motivating that I want to smoke George at the next math test.

1:10:42And here you are decades later. Okay, George. A new professor of math. Yeah, take it. Take it. George, if you're listening, please write in. No, I mean, I'm being silly. But let's, George, hopefully, I don't know if George will ever hear this, but if we're being realistic, I think it's probably a good dose of reality here that in all this talk about math is beautiful. It gives us joy. What is it that little kids like about math? Because some little kids do like it. When I think of my daughters, when they – they didn't particularly like it, but when they did, it was not because it was beautiful.

1:11:19It was because it was satisfying. You could get the answer to work. The teacher could teach you something. You could listen. You could do it. And it worked. And then you get it right. And it wasn't subjective. You know, I mean, there's something for many people that's very comforting about the black and white nature of math at the elementary school level. And it gets frustrating to some of those same people when math gets more ambiguous or like, how do I come up with the proof? What am I supposed to do? But anyway, for much of the world, math is this – to the extent that people like it is just because it works.

1:11:53It's helpful. It's very black and white. And you can be trained to do it. So that's part of it too. Anyway, but as far as identity, what's lost? We can't be heroes, some people have said. If we can't climb the mountains first because the AIs will get there first, those of us who fancied ourselves as mathematical heroes, we can't be heroes anymore. I don't know. Maybe. For a little quick context, so my husband and I, in the mornings, we talk about how different technologies have impacted the cultures of different communities of professionals. So, for example, like how did the loom impact the culture of weavers?

1:12:29Yes. And, you know, what happened to like the steel community when, you know, machines were able to take a lot of their jobs? And so I think that's where this question came from is like, how does technology impact the cultures of different professions? And he probably, yeah, as I said, didn't mean for me to say the word outmoded. And so I'm sorry about that. No, but that's OK. But I really like it what you and your husband are going for with this question because it's so common in these AI math interviews for mathematicians to think like this is – we don't think it. But it comes across like we think we're the first ones this is happening to.

1:13:02And I've frequently gotten letters from people, emails saying, you know, boo-hoo, it's happening to you. But how about those of us, like you mentioned, the weavers or the people with looms or anyone who works on a farm and now the tractor can do a lot of the work that a person used to do on a farm. So people who took pride in their labor, which a lot of people do feel a great deal of dignity to do a good job at the thing that they learned how to do, that they care about, that they have community of other people who do it with them. To have that taken away and then to be told, oh, well, you just need some retraining.

1:13:37You know, we're going to retrain you. You could you can have a different job. Learn to code. Oh, wait, no, not that. Well, yeah. OK, you could be retrained, but you maybe you love the thing you were doing. Right. And you became your identity and your culture. It was part of your identity. That's a serious, real thing that is should not be laughed at or sneezed at. Yeah. Well, in terms of identity and profession, the last question I want to ask you is a question we also asked our community of listeners, which is, do you think we'll have math professors in 10 years, in 50 years? What's your advice to students thinking about a career in math like you've had?

1:14:15It probably won't be a career like I've had, but I do believe we will still have teachers and students of math who are human beings. I do think so. and my reason for saying it is not super sentimental. I think it's just kind of, I mean, I want to believe it, first of all. I'm not convinced that it's true, but for it not to be true, it would mean that either people are learning math from AI or whatever the future version is, robots, or they're not learning it at all, right? Those are the only possibilities. The logical possibilities are kids continue to do math. They're either taught by machines or people, or they're not doing math.

1:14:58So which of these do we think is the real scenario? And I kind of picture it as people learning math from people. Now, why? What's the point if the machines are better than us at all aspects of it, as I think they will be? And I guess my answer would be today that we'll do it because we like to think and we like to think together. We still like to play games together. People have pointed out many times that we're still playing chess and getting pleasure from it and struggling and it's difficult and it hurts your head after a few hours. But we still do it partly because it's fun, partly because we're competitive and we want to beat the George in your life.

1:15:45And I mean, some people might do it because of tradition, like they love the history of chess. There is such a thing. I happen to love chess myself. I mean, I know about Paul Morphy and Jose Raul Capablanca and all the great masters, and I've studied their games. And so I would teach my kid chess. In fact, I have taught my girls chess. Now, they didn't take any great interest in it. But still, why does a person play chess anymore? Well, because they like it. It's something to do with someone else you feel like playing with. But it's a weird analogy because chess is not consequential in the way that math is.

1:16:24Math really has, as we've talked about over and over here, math has changed the world. It's helped build the civilization we're in. It also gives us, this I didn't go into yet, but let me try to go one more crazy place with you. Actually, I like where I'm going to go. Okay, see if you like it. Math gives us access to something transcendent. Math gives you a feeling of the sublime. Math is the only thing we know where you have absolute proof. We don't even have that in science. In math, you can come as close as a human being can come to certainty. And there's something comforting about that. There's something comforting about, one of my students recently when I was talking to her about why do you like math, she said, because it makes me feel safe.

1:17:07It's not a word I would have ever thought of. She said, I said, what do you mean safe? She said, a lot of things in her life when she was growing up didn't make sense to her. there were some very difficult things in her life that were unsafe. But in math, things make sense. And she found great solace in that. So there are things about math that go way beyond numbers and shapes that can make us feel at home in the universe and literally have access to some of the deepest things that physicists have taught us about the universe, right? I mean, as you know, you're nodding and smiling. You have that face, Daniel.

1:17:41I like it. That math is a route to the things that, like Einstein referred to as when he said, the eternal mystery of the universe is its comprehensibility. That we as primates can somehow, through our minuscule math and physics, we can actually learn how the planets move, how galaxies cluster. We can predict what the waves in space-time from colliding black holes will be like when they get here. And we're right. And it was with math that we, I mean, of course, our fantastic instruments, all of it. It's pretty good for this little ape on this middle-sized planet in a nothing special galaxy to be able to figure out how black holes send ripples through space-time.

1:18:27We figured that out. That's pretty good. Amen. Right? That is pretty good. Amen. So I think, you know, if you learn about math, even if you're not the best on the planet, this puts you in touch with something pretty close to the sublime. All right. Well, I think that's a great place to end it. Thank you very much for taking us there, Stephen. And thanks for all your candid answers to all of these crazy questions. Thanks, guys. Thank you, Daniel and Kelly. This was really fun. And everybody should check out Stephen Alex Townsend's new book, Big Math, The Hidden Codes That Run Our World.

1:19:05Thanks, everybody, for listening. Please go and do us a favor and rate the show on whatever podcast app you're using. It really helps people find us. Daniel and Kelly's Extraordinary Universe is edited by the amazing Matt Kesselman. He really is a wizard. You can also find us online on Blue Sky, Instagram, and XDNKUniverse. Come engage with us. You can email us at questions at danielandkelly.org. We really do want to hear from you. And you can find our website, www.danielandkelly.org, where you'll also find an invitation to join our Discord, where everybody comes and talks about the amazing universe.

1:19:45And we also have the most amazing moderators. This is an iHeart Podcast. Thanks for joining us. Morning, Carmen. Hey, Garb. Morning. Quick question. Just one second. Sorry. Carmen's worked at this office for 20 years. She knows everything, which is why Carmen has to help with, well, everything. Hey, Carmen, what did Faye pitch in 2023? Who owns the Nordic account? Oh, uh, where's the printer? But Carmen's revenue report is due in five minutes, so she shows them WorkIQ. Hey, Copilot, who owns the Nordic account? And what did we pitch them in 2023? Get the context you need with Microsoft Copilot.

1:20:22Work lives on. Hey, it's Kelly Rowland. You may not know this, but I have eczema, so I get how it can steal your time. Why let eczema take over? Flare less with EBCLIS. About 4 in 10 people taking EBCLIS achieved clear or almost clear skin at 16 weeks, and most of these people maintained clearer skin at one year. EBCLIS, Lubrikizumab LBKZ, a 250 milligram per 2 milliliter injection, is a prescription RX medicine for people age 12 and up who weigh at least 88 pounds with moderate to severe eczema, atopic dermatitis, that is not well controlled with Rx topical therapies or who cannot use those therapies.

1:20:59Don't use if you are allergic to Epglyss. Allergic reactions can occur that can be severe. Eye problems can occur. Tell your doctor if you have new or worsening eye problems. You should not receive a live vaccine when treated with Epglyss. Before starting Epglyss, tell your doctor if you have a parasitic infection. Paid partnership with Lilly. Respect your time. Ask your doctor about Epglyss and visit Epglyss.com or call 1-800-LILLY-RX or 1-800-545-5979. This war amongst us has got to end. The Hulu original phenomenon, All's Fair returns. You got one shot to make this right. I'm not saying I hate you, but I would unplug your life support to charge my phone.

1:21:38From executive producer Ryan Murphy. I will not stop, and I will win. Let's give these girls a show. This is going to be a good time. The Hulu original series, All's Fair. Now streaming on Hulu and Hulu on Disney Plus for bundle subscribers. Terms apply. With ChatGPT, you can do more than get answers. You can get things done. Whether you're leading a big project, planning an elaborate trip, or trying to balance work and life, ChatGPT can help. Give it a goal, plus things like files, notes, or project plans, develop a trip itinerary with daily updates, or create a prioritized schedule so you can spend less energy sorting out the details and more on bringing your biggest ideas to life.

1:22:19Get started at ChatGPT.com. This is an iHeart Podcast. Guaranteed human.

From the publisher

Daniel and Kelly talk to Dr. Steven Strogatz about his new book "Big Math", and how AI is changing mathematics.

See omnystudio.com/listener for privacy information.

More from Daniel and Kelly’s Extraordinary Universe

All episodes