← All guides

Languages

What happens to mathematics when artificial intelligence becomes a prophet?

From social media algorithms, package delivery, and weather forecasting to medical imaging and machine learning, the functioning of our modern world relies on mathematics. The rise of computers has bolstered the dominance of mathematics, enabling us to process increasingly vast quantities of data—numbers that would otherwise exceed the processing capacity of our limited human brains. Yet, with the emergence of large language models, AI-powered chatbots, and the advent of quantum computing, “big data” is beginning to completely outstrip the capabilities of its own creators. What this signifies, and how we might adapt, is the central focus of a new book by applied mathematicians Steven Strogatz and Alex Townsend, titled Big Data: The Hidden Code That Runs Our World.

Within his field, Strogatz is perhaps best known for a seminal 1998 paper, co-authored with Duncan Watts, on the collective dynamics of “small-world” networks. His research has also spanned the application of chaos theory in physics, the geometry of supercoiled DNA, and collective biological oscillators—such as swarms of fireflies flashing in unison or groups of crickets chirping in synchrony. In 2005, Strogatz extended his study of synchronization to the phenomenon of crowds causing the wobbling of London’s Millennium Bridge.

An award-winning communicator, Strogatz has authored columns on mathematics for The New York Times (appearing in 2010, 2012, and 2025) and hosted a podcast for Quanta Magazine from 2020 to 2021. He is the author of four bestselling books: Sync (2003), The Calculus of Friendship (2009), The Joy of x (2012), and the New York Times bestseller Infinite Powers (2019).

For this project, Strogatz has found an ideal collaborator in Alex Townsend—a younger colleague recently appointed at Cornell University, whose research focuses on numerical analysis and deep learning. As a senior professor, Strogatz was assigned to mentor Townsend, and the two began meeting regularly for lunch at Risley Hall. Townsend’s course on the 20 most significant algorithms of the 20th century showed great potential as the basis for a bestselling book; intrigued, Strogatz proposed a collaboration. This concept gradually evolved into a book about large-matrix computation, eventually developing into Big Math.

“There are many books on pure mathematics out there—as if computers didn’t exist,” Strogatz told Ars. “Alex and I are both applied mathematicians. We felt there should be a book about computers and mathematics—specifically regarding the kind of massive-scale problems involving billions of variables or parameters that are simply beyond the capacity of the human brain to solve. Many of the mathematical methods we’ve relied on for centuries are perfectly applicable to the computer age. Consider the abacus in relation to arithmetic, or the ‘counting board’ (a precursor to the matrix) used for systems of linear equations—a tool that could be manipulated by hand using rods or counters. In this sense, the practice of offloading mathematics from our minds into external tools is a tradition that stretches back two thousand years.”

In the era of artificial intelligence, mathematics faces unprecedented challenges, sparking intense debate over the role of human reasoning as machines drive most major breakthroughs. “There is a mathematician inside each of us, forever hungry to know more,” write Strogatz and Townsend. “Humanity must insist that knowledge serves a purpose greater than itself. This may be the mission of our century: to grow in power without losing our soul. As mathematics grows in scale, so must we.”

Ars spoke with Strogatz to learn more.

Ars Technica: Let’s start by defining “Big Math,” as it’s a very broad, umbrella term.

Steven Strogatz: Operationally speaking, “Big Math” refers to mathematical operations that are “massive” relative to the limits of human cognitive capacity. Something might appear small, yet be incredibly complex. We also want to extend the term to include what people typically think of as computational engineering—for instance, the construction of the Burj Khalifa, which falls under civil engineering or architectural design. It also applies to medical imaging and the moon landing. The lunar module was guided to the moon’s surface by a small computer that only had to account for twenty or thirty variables at a time—yet this is far beyond what a human could calculate mentally.

“Big Math” isn’t just about dealing with large numbers. The book explores the significance of linearity and sparsity. When problems are linear, they can scale to an immense size while remaining manageable. If a problem is non-linear, pure mathematics often hits a wall; you might be able to solve it using a computer, but for those who seek rigorous, provable theorems, entering the realm of non-linear, large-scale problems means stepping into the unknown.

Even simple-looking mathematical problems can be exceptionally difficult. Historically, the mathematical problems we encountered—such as those in arithmetic or algebra, the staples of high school curricula—typically involved only one or two variables. Even today, many problems in pure mathematics remain relatively “small.” They don’t involve a vast number of variables simultaneously, because doing so would render them completely unsolvable. While the scale of cryptography itself may be small, it remains highly complex due to its non-linear nature and other challenging algebraic problems. Thus, while a large problem scale is one manifestation of its difficulty, there are many other facets to what makes it difficult.

Today, even the largest language models often reach a parameter count in the trillions. Yet, in this context, infinity is not actually that large. One might assume that infinity is the grandest concept in mathematics, but that is not the case—in fact, calculus is remarkably well-suited to the human mind. Infinity is, in many ways, easier to grapple with than a billion.

Ars: One of the real-world examples of “big-picture” mathematics you cite is particularly relevant today: the relationship between mathematics and the issue of redistricting.

Strogatz: That’s a fascinating question: what constitutes a fair redistricting map? What makes for a reasonable redrawing of electoral districts? The legal system provides certain criteria, but they are quite vague. For instance, districts shouldn’t meander like a snake or resemble a salamander. The law uses the term “compact”—districts should be reasonably compact. But what does that mean mathematically? Districts should also be contiguous; you certainly wouldn’t want isolated areas counted as part of the same district. Redistricting should respect natural boundaries, such as rivers or other geographical features. Furthermore, each district should have roughly the same population.

So, how do you draw a map that meets all these criteria? As it turns out, there are many ways to do so. By using historical voting data, one can manipulate the drawing of district lines—this is the essence of gerrymandering: redrawing districts to favor the party in power. There has been intense debate over whether certain redistricting plans infringe upon citizens’ rights. The guiding principle should be “one person, one vote.” Any practice that diminishes someone’s power—effectively causing their vote to be “wasted”—is traditionally considered unfair, and potentially even illegal.

The Supreme Court has heard a wide variety of cases, yet it has shown significant resistance to the use of mathematical methods. I recall Chief Justice John Roberts’ remarks regarding the “efficiency gap” as a metric for gerrymandering. Referring to the efficiency gap using Greek letters, he remarked, “This is just sociological gobbledygook—it’s not something that should be used as a standard for legal consideration.”

Nevertheless, Moon Duchin and her colleagues have demonstrated a highly pragmatic approach: they can generate a vast number of reasonable redistricting maps that satisfy all legal requirements, thereby establishing a baseline against which other maps can be compared—much like a controlled experiment. The mathematical principle behind this approach is that it provides a principled method for randomly generating redistricting maps that comply with legal requirements; however, these maps are merely random samples of all plausible redistricting plans. How frequently does a given map yield the outcome anticipated by those who drew it? If it can be shown that such an outcome is an extreme outlier, that is significant. Does this prove malicious intent? No. But it does demonstrate that such an outcome is highly unusual.

In a voting rights case in Alabama, Chief Justice Roberts effectively sided with the plaintiffs, who argued that the state’s redistricting map caused actual harm to Black voters. The state claimed it was impossible to devise a reasonable map in which two districts would have a majority of Black voters. However, the work of Moon Duchin proved this assertion wrong; in fact, thousands of such maps could be generated. Thus, the state’s claim was simply untrue.

Just as the manuscript was being finalized for publication, the Supreme Court ruled that, in effect, partisan gerrymandering is permissible and does not violate the Voting Rights Act. Consequently, we pivoted to a more optimistic narrative.

Ars: From a philosophical perspective, how has mathematics evolved over time?

Strogatz: I think the most concise way to summarize this is through the role of the computer. For a long time, the computer served as a capable assistant to mathematicians, engineers, and scientists—philosophically, it was no more threatening than a microscope or a telescope. It extends our senses; it is a tool—a chisel, an anvil. It enables us to perform calculations more rapidly, handling tasks on a scale that would be difficult for humans to manage alone. We once relied on mechanical calculators; by the 1950s, with the invention of the transistor, these evolved into electronic computers—vastly faster, eventually leading to the supercomputers we have today. We write the programs, and they execute our instructions precisely. In that sense, they are our assistants.

The first shift of major philosophical significance occurred in the 1950s, when computers transitioned from mere tools to explorers. You could send a computer into the unknown, uncharted territories of the mathematical landscape. A prime example is the work of Enrico Fermi and his colleagues—Pasta, Ulam, and Tsingou—which we discuss in our chapter on simulations. It was Mary Tsingou who carried out the actual computations for what is now known as the Fermi-Pasta-Ulam numerical experiment.

Unlike previous applications of computers, Fermi and his colleagues did not know what the answer would be. Calculating the energy released by an atomic bomb explosion or the trajectory of an artillery shell is one thing—we may not know the exact numerical values, but we at least understand the underlying physical processes. In this experiment, they sought to simulate a simplified model of how atoms—to use the technical term—reach thermal equilibrium through the vibration of their chemical bonds. The goal was to observe the decay of entropy: as a system evolves naturally, how does it descend into an increasingly disordered state?

Fermi hypothesized that the non-linear nature of chemical bonds—meaning that stretching a bond to twice its length does not result in a restoring force exactly twice as strong (i.e., the relationship is not strictly proportional)—might explain the Second Law of Thermodynamics, or the law of increasing entropy. This problem could not be solved with pencil and paper, as it was impossible to calculate a sufficient number of non-linear interactions to observe the emergence of entropy.

When they ran the program on a computer, the results were startling: rather than dissipating, the energy exhibited a peculiar, recurring echo. Initially, all the energy was concentrated in a specific vibrational mode of the system; as the system evolved, this energy began to disperse—yet, after some time, 98% of it returned to the original mode. This phenomenon is known as “recurrence.” It paved the way for an entirely new branch of mathematics: soliton theory. This early experiment served as a warning that once one ventures into the “jungle” of non-linearity, human intuition ceases to be reliable. We could not enter that realm directly, so we sent out a scout—the computer—to explore this wilderness of thought; it was the only tool capable of performing the necessary calculations and reporting the findings back to us.

Ars: The rapid rise of artificial intelligence in mathematics has caused concern among many in the field—especially following OpenAI’s release of 722 AI-generated mathematical papers last month, an event jokingly dubbed the “math apocalypse.” While AI is, technically speaking, just another new tool, it feels fundamentally different from anything that came before.

Strogatz: Right—the transition from “servant” to “scout” represents something entirely new; the computer has become our partner. At first, it was all incredibly exciting—like having a brand-new Ferrari. You could race ahead and discover new things. That’s how Fermi and Ulam described it back in the day. It was truly magical; they had no idea what might emerge. Computers began to “teach” us by showing us phenomena, yet we were still able to keep pace—developing new frameworks, such as soliton theory, to explain what we were seeing.

What’s different today is that the technology is shifting from servant to scout, and potentially toward “prophet”—revealing things we don’t understand, yet remaining unable to provide a clear explanation for them. That seems accurate. A prime example is Google Translate: it excels at translating between languages, but it hasn’t uncovered any new linguistic principles—or, if it has, it cannot articulate what they are.

Similarly, AlphaFold—the protein-folding algorithm that contributed to a Nobel Prize in Chemistry a few years ago—allows us to resolve protein structures at a scale a thousand times greater than before. Yet, this breakthrough didn’t stem from a fundamental new understanding of physics or chemistry; the system simply identified patterns in protein folding without explaining the underlying principles. That is where we find ourselves today, and the situation is likely to intensify.

Ars: In your book, you raise the question of whether human insight still matters, suggesting that Stephen Wolfram might argue it does not.

Strogatz: Wolfram’s position is nuanced. He emphasizes the concept of “computational irreducibility.” I personally favor terms like “clarity” and “acuity.” Systems can be transparent—you can see right through them. For millennia, science has focused on such transparent systems—systems that yield to insight precisely because you possess the keenness to perceive them. But Wolfram might ask: are we like the proverbial drunkard who searches for his lost keys under a streetlight simply because it’s the only place where there’s any light? In other words, perhaps the keys simply cannot be found anywhere else.

So, have we artificially limited ourselves to systems transparent enough to yield to the kind of science that relies on human insight? I suspect Wolfram would argue that most natural phenomena are computationally irreducible—meaning there is no theory capable of providing a simple, comprehensive explanation of how they function. The best you can do is run a simulation and observe the outcome—much like letting nature take its course and watching what actually unfolds in reality.

In today’s parlance, you shouldn’t expect to find a neat, concise description—a simplified model that explains the phenomena you’re witnessing. We’ve seen such models before; Newton, for instance, could explain planetary motion using just a few simple principles, and it worked perfectly. But from Wolfram’s perspective, I believe he would view that as an artificial simplification. Most of science, ultimately, won’t work that way. Whether Wolfram is entirely correct remains an open question. While it is clearly true that many computer simulations are computationally irreducible, it does not necessarily follow that the natural world is, too.

Ars: If you can’t see what’s happening inside a model—and the model itself can’t explain its own operations—how can you know it’s correct?

Strogatz: Let’s assume, for the sake of argument, that we don’t understand these systems, and they aren’t currently capable of explaining themselves to us. We should perhaps add an asterisk here, though—these systems might eventually prove to be excellent teachers. It’s possible that while we don’t currently grasp how tools like Google Translate or AlphaFold work, we might fully understand them a decade from now. I believe that, given the inherent limitations of human beings—our finite processing speed, intelligence, and memory—we will inevitably encounter systems that are truly opaque. Artificial intelligence may reveal things to us that, even with the aid of AI itself, we remain fundamentally unable to comprehend. What happens then?

At that point, I think mathematics begins to resemble fields like science, medicine, or engineering. Consider the FDA: for a new drug to be approved, it must undergo extensive controlled studies and randomized trials to demonstrate its safety. The medical field has developed a robust methodology for dealing with the “opacity” of drugs—after all, a physician doesn’t necessarily need to understand the precise mechanism by which a medication works. You do not need to know how aspirin works. Nor do you need to know how electroconvulsive therapy treats treatment-resistant depression. Even without a full understanding, we can still utilize these techniques or medications. You do not need to grasp the underlying principles. If artificial intelligence functions as an “oracle,” you can still verify it by observing its accuracy rate.

Ars: What do we lose in this process?

Strogatz: It is indeed a loss. I don’t want to downplay the seriousness of this—to use a technical philosophical term—epistemic loss. For a long time, one of the things we in pure mathematics have taken the greatest pride in is that we are the only discipline that possesses absolute proof. We have certainty. We have come as close to certainty as is humanly possible. We can construct proofs. A proof by Euclid, written in 300 BCE, remains correct to this day. Scientific and medical theories from 300 BCE are no longer accepted, but the mathematical derivations from that era remain valid.

Thus, mathematics carries a certain eternal and unique significance. Even though we are merely primates, we seem to touch upon something profoundly essential. Consequently, the idea that “science and engineering are models of mathematics” represents a massive sacrifice for pure mathematicians—they are loath to relinquish the certainty that has long been our greatest source of pride.

Ars: Many mathematicians feel a sense of sadness—not just because of these epistemological issues, but because the very purpose that drew us to mathematics in the first place is being compromised. You’ve written about those “Aha!” moments—those flashes of insight when, after struggling with a problem, you suddenly discern the underlying pattern. Are we losing these moments of epiphany?

Strogatz: There are many reasons for mathematicians to feel sad—and those of us who do feel it, feel it deeply. I’ve gone through all five stages of grief—though I’m only joking, of course. Young students, however, don’t find it funny. I’m nearing the end of my career, so it’s one thing for me to view this through a philosophical lens. But imagine being 25 years old, passionate about mathematics, and dreaming of becoming a mathematician—only to be told that mathematics is now obsolete. Some even take a certain glee in this; there are people in the tech world who are quite callous about the current state of mathematics.

We’ve discussed the epistemic dimension, but there is also the spirit of fruitful inquiry. Pure mathematicians like Terence Tao have consistently emphasized the importance of understanding. This is how we understand insight: when you strive to understand something—even if you don’t ultimately solve it—you often come up with various strategies or simplified approaches to the problem. You learn a great deal throughout the process of exploration. You also gain friendships and the joy that comes from being with like-minded people. This is not unique to mathematics. Consider those who love art or knitting—or my colleagues in creative writing, who are deeply disappointed by the AI-generated work their students are now submitting. What they seek is work that possesses a soul and a distinct personality.

Ars: This is the exact opposite of the approach taken by artificial intelligence. I worry that it might have the same consequence for mathematics. Mathematics inherently possesses a sense of beauty—an aesthetic dimension. If you focus solely on the answer while disregarding the process, you suffer a qualitative loss. Furthermore, individual mathematicians bring their own unique perspectives.

Strogatz: Exactly. Some people have a very distinct style—evident in the questions they choose to ask, how they approach those problems, and how they explain their reasoning. In several interviews over the past few months, I’ve found myself moved to tears on camera, fearing that much of this might be slipping away. But I must emphasize that we don’t truly know yet. For instance, what if AI turns out to be exceptionally good at teaching us? In my own interactions with AI, I’ve found it to be a remarkably effective explainer.

Why should we feel sad? Perhaps, with the help of these machines, we might come to understand mathematics better than ever before. We may no longer be the ones to make the greatest breakthroughs—as the machines will likely have already made them—but we can still make smaller discoveries of our own. I have never made any major, groundbreaking discoveries myself; I’m certainly not at the level of a Fields Medalist. Yet, I still find great joy in mathematics—as do many others whose mathematical abilities may be far more modest. There are countless amateur mathematicians who take pleasure in exploration, regardless of what they might independently discover. It’s much like solving a puzzle—the jigsaw or the Sudoku grid itself isn’t what matters; it’s the joy of the thinking process.

Some people enjoy physical exercise, others thrive on competition—there’s no need to pretend otherwise. Some are driven by the desire to be the first to make a major discovery. That can be an alluring goal, though it often comes with significant hardship.

Ars: Is there still hope?

Strogatz: At the moment, AI isn’t particularly good at asking questions. However, I believe it could be trained to pose meaningful ones. Graduate students, when they first start out, are often not very good at asking questions either—largely because universities tend to teach how to provide answers rather than how to formulate questions. In fact, one of the most challenging aspects of mentoring new PhDs or postdocs is helping them learn how to ask the right questions. I believe we could train AI to do the same.

It is true that one might feel a sense of loss, but I believe we should not overlook the excitement. Fields such as energy technology, climate science, medicine, theoretical physics, and pure mathematics all hold the potential for significant, tangible benefits. The fact is, we are making astonishing discoveries through this machine. For many, therefore, this is a moment of great exhilaration—yet it is also deeply bewildering. To say anything else would be dishonest.

This book concludes with a reference to Faust. A “Faustian bargain” is commonly understood as “selling your soul to the devil, for which you must now pay the price.” Goethe’s version, however, is more ambiguous: perhaps we might find redemption if we strive, unceasingly, to create something that transcends our own limitations. Even so, it remains an exceptionally dangerous game—one that we have chosen to play.