Latest news with #QuantaMagazine
Scientific American
15 hours ago
- Science
- Scientific American
Science Quiz: Event Horizons and Flesh-Eating Parasites
Allison Parshall is an associate editor at Scientific American covering mind and brain. She writes the magazine's Contributors column and weekly online Science Quizzes. As a multimedia journalist, she contributes to Scientific American 's podcast Science Quickly. Parshall's work has also appeared in Quanta Magazine and Inverse. She graduated from New York University's Arthur L. Carter Journalism Institute with a master's degree in science, health and environmental reporting. She has a bachelor's degree in psychology from Georgetown University. Follow Parshall on X (formerly Twitter) @parshallison
Scientific American
13-06-2025
- Science
- Scientific American
Science Quiz: Immortal Animals and Analemmas
Allison Parshall is an associate editor at Scientific American covering mind and brain. She writes the magazine's Contributors column and weekly online Science Quizzes. As a multimedia journalist, she contributes to Scientific American 's podcast Science Quickly. Parshall's work has also appeared in Quanta Magazine and Inverse. She graduated from New York University's Arthur L. Carter Journalism Institute with a master's degree in science, health and environmental reporting. She has a bachelor's degree in psychology from Georgetown University. Follow Parshall on X (formerly Twitter) @parshallison
WIRED
08-06-2025
- Science
- WIRED
A New Law of Nature Attempts to Explain the Complexity of the Universe
Jun 8, 2025 7:00 AM A novel suggestion that complexity increases over time, not just in living organisms but in the nonliving world, promises to rewrite notions of time and evolution. Illustration: Irene Pérez for Quanta Magazine The original version of this story appeared in Quanta Magazine. In 1950 the Italian physicist Enrico Fermi was discussing the possibility of intelligent alien life with his colleagues. If alien civilizations exist, he said, some should surely have had enough time to expand throughout the cosmos. So where are they? Many answers to Fermi's 'paradox' have been proposed: Maybe alien civilizations burn out or destroy themselves before they can become interstellar wanderers. But perhaps the simplest answer is that such civilizations don't appear in the first place: Intelligent life is extremely unlikely, and we pose the question only because we are the supremely rare exception. A new proposal by an interdisciplinary team of researchers challenges that bleak conclusion. They have proposed nothing less than a new law of nature, according to which the complexity of entities in the universe increases over time with an inexorability comparable to the second law of thermodynamics—the law that dictates an inevitable rise in entropy, a measure of disorder. If they're right, complex and intelligent life should be widespread. In this new view, biological evolution appears not as a unique process that gave rise to a qualitatively distinct form of matter—living organisms. Instead, evolution is a special (and perhaps inevitable) case of a more general principle that governs the universe. According to this principle, entities are selected because they are richer in a kind of information that enables them to perform some kind of function. This hypothesis, formulated by the mineralogist Robert Hazen and the astrobiologist Michael Wong of the Carnegie Institution in Washington, DC, along with a team of others, has provoked intense debate. Some researchers have welcomed the idea as part of a grand narrative about fundamental laws of nature. They argue that the basic laws of physics are not 'complete' in the sense of supplying all we need to comprehend natural phenomena; rather, evolution—biological or otherwise—introduces functions and novelties that could not even in principle be predicted from physics alone. 'I'm so glad they've done what they've done,' said Stuart Kauffman, an emeritus complexity theorist at the University of Pennsylvania. 'They've made these questions legitimate.' Michael Wong, an astrobiologist at the Carnegie Institution in Washington, DC. Photograph: Katherine Cain/Carnegie Science Others argue that extending evolutionary ideas about function to non-living systems is an overreach. The quantitative value that measures information in this new approach is not only relative—it changes depending on context—it's impossible to calculate. For this and other reasons, critics have charged that the new theory cannot be tested, and therefore is of little use. The work taps into an expanding debate about how biological evolution fits within the normal framework of science. The theory of Darwinian evolution by natural selection helps us to understand how living things have changed in the past. But unlike most scientific theories, it can't predict much about what is to come. Might embedding it within a meta-law of increasing complexity let us glimpse what the future holds? Making Meaning The story begins in 2003, when the biologist Jack Szostak published a short article in Nature proposing the concept of functional information. Szostak—who six years later would get a Nobel Prize for unrelated work—wanted to quantify the amount of information or complexity that biological molecules like proteins or DNA strands embody. Classical information theory, developed by the telecommunications researcher Claude Shannon in the 1940s and later elaborated by the Russian mathematician Andrey Kolmogorov, offers one answer. Per Kolmogorov, the complexity of a string of symbols (such as binary 1s and 0s) depends on how concisely one can specify that sequence uniquely. For example, consider DNA, which is a chain of four different building blocks called nucleotides. Α strand composed only of one nucleotide, repeating again and again, has much less complexity—and, by extension, encodes less information—than one composed of all four nucleotides in which the sequence seems random (as is more typical in the genome). Jack Szostak proposed a way to quantify information in biological systems. Photograph: HHMI But Szostak pointed out that Kolmogorov's measure of complexity neglects an issue crucial to biology: how biological molecules function. In biology, sometimes many different molecules can do the same job. Consider RNA molecules, some of which have biochemical functions that can easily be defined and measured. (Like DNA, RNA is made up of sequences of nucleotides.) In particular, short strands of RNA called aptamers securely bind to other molecules. Let's say you want to find an RNA aptamer that binds to a particular target molecule. Can lots of aptamers do it, or just one? If only a single aptamer can do the job, then it's unique, just as a long, seemingly random sequence of letters is unique. Szostak said that this aptamer would have a lot of what he called 'functional information.' Illustration: Irene Pérez for Quanta Magazine If many different aptamers can perform the same task, the functional information is much smaller. So we can calculate the functional information of a molecule by asking how many other molecules of the same size can do the same task just as well. Szostak went on to show that in a case like this, functional information can be measured experimentally. He made a bunch of RNA aptamers and used chemical methods to identify and isolate the ones that would bind to a chosen target molecule. He then mutated the winners a little to seek even better binders and repeated the process. The better an aptamer gets at binding, the less likely it is that another RNA molecule chosen at random will do just as well: The functional information of the winners in each round should rise. Szostak found that the functional information of the best-performing aptamers got ever closer to the maximum value predicted theoretically. Selected for Function Hazen came across Szostak's idea while thinking about the origin of life—an issue that drew him in as a mineralogist, because chemical reactions taking place on minerals have long been suspected to have played a key role in getting life started. 'I concluded that talking about life versus nonlife is a false dichotomy,' Hazen said. 'I felt there had to be some kind of continuum—there has to be something that's driving this process from simpler to more complex systems.' Functional information, he thought, promised a way to get at the 'increasing complexity of all kinds of evolving systems.' In 2007 Hazen collaborated with Szostak to write a computer simulation involving algorithms that evolve via mutations. Their function, in this case, was not to bind to a target molecule, but to carry out computations. Again they found that the functional information increased spontaneously over time as the system evolved. There the idea languished for years. Hazen could not see how to take it any further until Wong accepted a fellowship at the Carnegie Institution in 2021. Wong had a background in planetary atmospheres, but he and Hazen discovered they were thinking about the same questions. 'From the very first moment that we sat down and talked about ideas, it was unbelievable,' Hazen said. Robert Hazen, a mineralogist at the Carnegie Institution in Washington, DC. Photograph: Courtesy of Robert Hazen 'I had got disillusioned with the state of the art of looking for life on other worlds,' Wong said. 'I thought it was too narrowly constrained to life as we know it here on Earth, but life elsewhere may take a completely different evolutionary trajectory. So how do we abstract far enough away from life on Earth that we'd be able to notice life elsewhere even if it had different chemical specifics, but not so far that we'd be including all kinds of self-organizing structures like hurricanes?' The pair soon realized that they needed expertise from a whole other set of disciplines. 'We needed people who came at this problem from very different points of view, so that we all had checks and balances on each other's prejudices,' Hazen said. 'This is not a mineralogical problem; it's not a physics problem, or a philosophical problem. It's all of those things.' They suspected that functional information was the key to understanding how complex systems like living organisms arise through evolutionary processes happening over time. 'We all assumed the second law of thermodynamics supplies the arrow of time,' Hazen said. 'But it seems like there's a much more idiosyncratic pathway that the universe takes. We think it's because of selection for function—a very orderly process that leads to ordered states. That's not part of the second law, although it's not inconsistent with it either.' Looked at this way, the concept of functional information allowed the team to think about the development of complex systems that don't seem related to life at all. At first glance, it doesn't seem a promising idea. In biology, function makes sense. But what does 'function' mean for a rock? All it really implies, Hazen said, is that some selective process favors one entity over lots of other potential combinations. A huge number of different minerals can form from silicon, oxygen, aluminum, calcium, and so on. But only a few are found in any given environment. The most stable minerals turn out to be the most common. But sometimes less stable minerals persist because there isn't enough energy available to convert them to more stable phases. 'Information itself might be a vital parameter of the cosmos, similar to mass, charge, and energy.' This might seem trivial, like saying that some objects exist while other ones don't, even if they could in theory. But Hazen and Wong have shown that, even for minerals, functional information has increased over the course of Earth's history. Minerals evolve toward greater complexity (though not in the Darwinian sense). Hazen and colleagues speculate that complex forms of carbon such as graphene might form in the hydrocarbon-rich environment of Saturn's moon Titan—another example of an increase in functional information that doesn't involve life. It's the same with chemical elements. The first moments after the Big Bang were filled with undifferentiated energy. As things cooled, quarks formed and then condensed into protons and neutrons. These gathered into the nuclei of hydrogen, helium, and lithium atoms. Only once stars formed and nuclear fusion happened within them did more complex elements like carbon and oxygen form. And only when some stars had exhausted their fusion fuel did their collapse and explosion in supernovas create heavier elements such as heavy metals. Steadily, the elements increased in nuclear complexity. Wong said their work implies three main conclusions. First, biology is just one example of evolution. 'There is a more universal description that drives the evolution of complex systems.' Illustration: Irene Pérez for Quanta Magazine Second, he said, there might be 'an arrow in time that describes this increasing complexity,' similar to the way the second law of thermodynamics, which describes the increase in entropy, is thought to create a preferred direction of time. Finally, Wong said, 'information itself might be a vital parameter of the cosmos, similar to mass, charge and energy.' In the work Hazen and Szostak conducted on evolution using artificial-life algorithms, the increase in functional information was not always gradual. Sometimes it would happen in sudden jumps. That echoes what is seen in biological evolution. Biologists have long recognized transitions where the complexity of organisms increases abruptly. One such transition was the appearance of organisms with cellular nuclei (around 1.8 billion to 2.7 billion years ago). Then there was the transition to multicellular organisms (around 2 billion to 1.6 billion years ago), the abrupt diversification of body forms in the Cambrian explosion (540 million years ago), and the appearance of central nervous systems (around 600 million to 520 million years ago). The arrival of humans was arguably another major and rapid evolutionary transition. Evolutionary biologists have tended to view each of these transitions as a contingent event. But within the functional-information framework, it seems possible that such jumps in evolutionary processes (whether biological or not) are inevitable. In these jumps, Wong pictures the evolving objects as accessing an entirely new landscape of possibilities and ways to become organized, as if penetrating to the 'next floor up.' Crucially, what matters—the criteria for selection, on which continued evolution depends—also changes, plotting a wholly novel course. On the next floor up, possibilities await that could not have been guessed before you reached it. For example, during the origin of life it might initially have mattered that proto-biological molecules would persist for a long time—that they'd be stable. But once such molecules became organized into groups that could catalyze one another's formation—what Kauffman has called autocatalytic cycles—the molecules themselves could be short-lived, so long as the cycles persisted. Now it was dynamical, not thermodynamic, stability that mattered. Ricard Solé of the Santa Fe Institute thinks such jumps might be equivalent to phase transitions in physics, such as the freezing of water or the magnetization of iron: They are collective processes with universal features, and they mean that everything changes, everywhere, all at once. In other words, in this view there's a kind of physics of evolution—and it's a kind of physics we know about already. The Biosphere Creates Its Own Possibilities The tricky thing about functional information is that, unlike a measure such as size or mass, it is contextual: It depends on what we want the object to do, and what environment it is in. For instance, the functional information for an RNA aptamer binding to a particular molecule will generally be quite different from the information for binding to a different molecule. Yet finding new uses for existing components is precisely what evolution does. Feathers did not evolve for flight, for example. This repurposing reflects how biological evolution is jerry-rigged, making use of what's available. Kauffman argues that biological evolution is thus constantly creating not just new types of organisms but new possibilities for organisms, ones that not only did not exist at an earlier stage of evolution but could not possibly have existed. From the soup of single-celled organisms that constituted life on Earth 3 billion years ago, no elephant could have suddenly emerged—this required a whole host of preceding, contingent but specific innovations. However, there is no theoretical limit to the number of uses an object has. This means that the appearance of new functions in evolution can't be predicted—and yet some new functions can dictate the very rules of how the system evolves subsequently. 'The biosphere is creating its own possibilities,' Kauffman said. 'Not only do we not know what will happen, we don't even know what can happen.' Photosynthesis was such a profound development; so were eukaryotes, nervous systems and language. As the microbiologist Carl Woese and the physicist Nigel Goldenfeld put it in 2011, 'We need an additional set of rules describing the evolution of the original rules. But this upper level of rules itself needs to evolve. Thus, we end up with an infinite hierarchy.' The physicist Paul Davies of Arizona State University agrees that biological evolution 'generates its own extended possibility space which cannot be reliably predicted or captured via any deterministic process from prior states. So life evolves partly into the unknown.' 'An increase in complexity provides the future potential to find new strategies unavailable to simpler organisms.' Mathematically, a 'phase space' is a way of describing all possible configurations of a physical system, whether it's as comparatively simple as an idealized pendulum or as complicated as all the atoms comprising the Earth. Davies and his co-workers have recently suggested that evolution in an expanding accessible phase space might be formally equivalent to the 'incompleteness theorems' devised by the mathematician Kurt Gödel. Gödel showed that any system of axioms in mathematics permits the formulation of statements that can't be shown to be true or false. We can only decide such statements by adding new axioms. Davies and colleagues say that, as with Gödel's theorem, the key factor that makes biological evolution open-ended and prevents us from being able to express it in a self-contained and all-encompassing phase space is that it is self-referential: The appearance of new actors in the space feeds back on those already there to create new possibilities for action. This isn't the case for physical systems, which, even if they have, say, millions of stars in a galaxy, are not self-referential. 'An increase in complexity provides the future potential to find new strategies unavailable to simpler organisms,' said Marcus Heisler, a plant developmental biologist at the University of Sydney and co-author of the incompleteness paper. This connection between biological evolution and the issue of noncomputability, Davies said, 'goes right to the heart of what makes life so magical.' Is biology special, then, among evolutionary processes in having an open-endedness generated by self-reference? Hazen thinks that in fact once complex cognition is added to the mix—once the components of the system can reason, choose, and run experiments 'in their heads'—the potential for macro-micro feedback and open-ended growth is even greater. 'Technological applications take us way beyond Darwinism,' he said. A watch gets made faster if the watchmaker is not blind. Back to the Bench If Hazen and colleagues are right that evolution involving any kind of selection inevitably increases functional information—in effect, complexity—does this mean that life itself, and perhaps consciousness and higher intelligence, is inevitable in the universe? That would run counter to what some biologists have thought. The eminent evolutionary biologist Ernst Mayr believed that the search for extraterrestrial intelligence was doomed because the appearance of humanlike intelligence is 'utterly improbable.' After all, he said, if intelligence at a level that leads to cultures and civilizations were so adaptively useful in Darwinian evolution, how come it only arose once across the entire tree of life? Mayr's evolutionary point possibly vanishes in the jump to humanlike complexity and intelligence, whereupon the whole playing field is utterly transformed. Humans attained planetary dominance so rapidly (for better or worse) that the question of when it will happen again becomes moot. Illustration: Irene Pérez for Quanta Magazine But what about the chances of such a jump happening in the first place? If the new 'law of increasing functional information' is right, it looks as though life, once it exists, is bound to get more complex by leaps and bounds. It doesn't have to rely on some highly improbable chance event. What's more, such an increase in complexity seems to imply the appearance of new causal laws in nature that, while not incompatible with the fundamental laws of physics governing the smallest component parts, effectively take over from them in determining what happens next. Arguably we see this already in biology: Galileo's (apocryphal) experiment of dropping two masses from the Leaning Tower of Pisa no longer has predictive power when the masses are not cannonballs but living birds. Together with the chemist Lee Cronin of the University of Glasgow, Sara Walker of Arizona State University has devised an alternative set of ideas to describe how complexity arises, called assembly theory. In place of functional information, assembly theory relies on a number called the assembly index, which measures the minimum number of steps required to make an object from its constituent ingredients. 'Laws for living systems must be somewhat different than what we have in physics now,' Walker said, 'but that does not mean that there are no laws.' But she doubts that the putative law of functional information can be rigorously tested in the lab. 'I am not sure how one could say [the theory] is right or wrong, since there is no way to test it objectively,' she said. 'What would the experiment look for? How would it be controlled? I would love to see an example, but I remain skeptical until some metrology is done in this area.' Hazen acknowledges that, for most physical objects, it is impossible to calculate functional information even in principle. Even for a single living cell, he admits, there's no way of quantifying it. But he argues that this is not a sticking point, because we can still understand it conceptually and get an approximate quantitative sense of it. Similarly, we can't calculate the exact dynamics of the asteroid belt because the gravitational problem is too complicated—but we can still describe it approximately enough to navigate spacecraft through it. Wong sees a potential application of their ideas in astrobiology. One of the curious aspects of living organisms on Earth is that they tend to make a far smaller subset of organic molecules than they could make given the basic ingredients. That's because natural selection has picked out some favored compounds. There's much more glucose in living cells, for example, than you'd expect if molecules were simply being made either randomly or according to their thermodynamic stability. So one potential signature of lifelike entities on other worlds might be similar signs of selection outside what chemical thermodynamics or kinetics alone would generate. (Assembly theory similarly predicts complexity-based biosignatures.) There might be other ways of putting the ideas to the test. Wong said there is more work still to be done on mineral evolution, and they hope to look at nucleosynthesis and computational 'artificial life.' Hazen also sees possible applications in oncology, soil science and language evolution. For example, the evolutionary biologist Frédéric Thomas of the University of Montpellier in France and colleagues have argued that the selective principles governing the way cancer cells change over time in tumors are not like those of Darwinian evolution, in which the selection criterion is fitness, but more closely resemble the idea of selection for function from Hazen and colleagues. Hazen's team has been fielding queries from researchers ranging from economists to neuroscientists, who are keen to see if the approach can help. 'People are approaching us because they are desperate to find a model to explain their system,' Hazen said. But whether or not functional information turns out to be the right tool for thinking about these questions, many researchers seem to be converging on similar questions about complexity, information, evolution (both biological and cosmic), function and purpose, and the directionality of time. It's hard not to suspect that something big is afoot. There are echoes of the early days of thermodynamics, which began with humble questions about how machines work and ended up speaking to the arrow of time, the peculiarities of living matter, and the fate of the universe. Original story reprinted with permission from Quanta Magazine, an editorially independent publication of the Simons Foundation whose mission is to enhance public understanding of science by covering research developments and trends in mathematics and the physical and life sciences.
Scientific American
06-06-2025
- Science
- Scientific American
Science Quiz: Genes, Drugs and Nematodes
Allison Parshall is an associate editor at Scientific American covering mind and brain. She writes the magazine's Contributors column and weekly online Science Quizzes. As a multimedia journalist, she contributes to Scientific American 's podcast Science Quickly. Parshall's work has also appeared in Quanta Magazine and Inverse. She graduated from New York University's Arthur L. Carter Journalism Institute with a master's degree in science, health and environmental reporting. She has a bachelor's degree in psychology from Georgetown University. Follow Parshall on X (formerly Twitter) @parshallison
WIRED
01-06-2025
- Health
- WIRED
How to Make AI Faster and Smarter—With a Little Help from Physics
Jun 1, 2025 7:00 AM Rose Yu has drawn on the principles of fluid dynamics to improve deep learning systems that predict traffic, model the climate, and stabilize drones during flight. Photograph: Peggy Peattie for Quanta Magazine The original version of this story appeared in Quanta Magazine. When she was 10 years old, Rose Yu got a birthday present that would change her life—and, potentially, the way we study physics. Her uncle got her a computer. That was a rare commodity in China 25 years ago, and the gift did not go unused. At first, Yu mainly played computer games, but in middle school she won an award for web design. It was the first of many computer-related honors. Yu went on to major in computer science at Zhejiang University, where she won a prize for innovative research. For her graduate studies, she chose the University of Southern California (USC), partly because the same uncle—who was the only person she knew in the United States—was then working at the Jet Propulsion Laboratory in nearby Pasadena. Yu earned her doctorate in 2017 with an award for best dissertation. Her most recent honor came in January, when President Joe Biden, in his last week in office, gave her a Presidential Early Career Award. Yu, now an associate professor at the University of California, San Diego (UCSD), is a leader in a field known as 'physics-guided deep learning,' having spent years incorporating our knowledge of physics into artificial neural networks. The work has not only introduced novel techniques for building and training these systems, but it's also allowed her to make progress on several real-world applications. She has drawn on principles of fluid dynamics to improve traffic predictions, sped up simulations of turbulence to enhance our understanding of hurricanes, and devised tools that helped predict the spread of Covid-19. This work has brought Yu closer to her grand dream—deploying a suite of digital lab assistants that she calls AI Scientist. She now envisions what she calls a 'partnership' between human researchers and AI tools, fully based on the tenets of physics and thus capable of yielding new scientific insights. Combining inputs from a team of such assistants, in her opinion, may be the best way to boost the discovery process. Quanta spoke with Yu about turbulence in its many guises, how to get more out of AI, and how it might get us out of urban gridlock. The interview has been condensed and edited for clarity. Yu on the UCSD campus, where she is an associate professor. Photograph: Peggy Peattie for Quanta Magazine When did you first try to combine physics with deep learning? Rose Yu: It started with traffic. I was a grad student at USC, and the campus is right near the intersection of I-10 and I-110. To get anywhere, you have to go through a lot of traffic, which can be very annoying. In 2016, I began to wonder whether I could do anything about this. Deep learning—which uses multilayered neural networks to elicit patterns from data—was getting really hot back then. There was already a lot of excitement about applications in image classification, but images are just static things. I wondered whether deep learning could help with problems where things are constantly changing. I wasn't the first person to consider this, but my colleagues and I did find a novel way of framing the problem. What was your new approach? First, we thought of traffic in terms of the physical process of diffusion. In our model, the flow of traffic over a network of roads is analogous to the flow of fluids over a surface—motions that are governed by the laws of fluid dynamics. But our main innovation was to think of traffic as a graph, from the mathematical field of graph theory. Sensors, which monitor traffic on highways and other roads, serve as the nodes of this graph. And the edges of the graph represent the roads (and distances) between those sensors. Yu's interest in computers began with a gift for her 10th birthday. Photograph: Peggy Peattie for Quanta Magazine A graph provides a snapshot of the entire road network at a given time, telling you the average velocity of cars at every point on the graph. When you put together a series of these snapshots, spaced every five minutes apart, you get a good picture of how traffic is evolving. From there, you can try to predict what will happen in the future. The big challenge in deep learning is that you need a lot of data to train the neural network. Fortunately, one of my advisers, Cyrus Shahabi, had worked for many years on the problem of traffic forecasting, and he'd accumulated a vast amount of LA traffic data that I had access to. So how good were your predictions? Prior to our work, people could only make traffic forecasts that were reliable for about 15 minutes. Our forecasts were valid for one hour—a big improvement. Our code was deployed by Google Maps in 2018. A bit later, Google invited me to become a visiting researcher. That's about when you began working on climate modeling, right? Yes, that started in 2018, when I gave a talk at the Lawrence Berkeley National Laboratory. Afterward, I spoke with scientists there, and we looked for a problem that would be a good testbed for physics-guided deep learning. We settled on predicting the evolution of turbulent flow, which is a key factor in climate models, as well as an area of major uncertainty. Familiar examples of turbulence are the swirling patterns you see after pouring milk into a cup of coffee and giving it a stir. In the oceans, swirls like this can span thousands of miles. Predictions of turbulent behavior that are based on solving the Navier-Stokes equation, which describes the flow of fluids, are considered the gold standard in this field. But the required calculations are very slow, which is why we don't have good models for predicting hurricanes and tropical cyclones. The heavy congestion of Los Angeles first inspired Yu to model highway traffic as the flow of fluids. Photograph: Peggy Peattie for Quanta Magazine And deep learning can help? The basic idea is that deep neural networks that are trained on our best numerical simulations can learn to imitate—or as we say, 'emulate'—those simulations. They do that by recognizing properties and patterns buried within the data. They don't have to go through time-consuming, brute-force calculations to find approximate solutions. Our models sped up predictions by a factor of 20 in two-dimensional settings and by a factor of 1,000 in three-dimensional settings. Something like our turbulence prediction module might someday be inserted into bigger climate models that can do better at predicting things like hurricanes. Where else does turbulence show up? It's pretty much everywhere. Turbulence in blood flow, for instance, can lead to strokes or heart attacks. And when I was a postdoc at Caltech, I coauthored a paper that looked into stabilizing drones. Propellor-generated airflows interact with the ground to create turbulence. That, in turn, can cause the drone to wobble. We used a neural network to model the turbulence, and that led to better control of drones during takeoffs and landings. I'm currently working with scientists at UCSD and General Atomics on fusion power. One of the keys to success is learning how to control the plasma, which is a hot, ionized phase of matter. At temperatures of about 100 million degrees, different kinds of turbulence arise within the plasma, and physics-based numerical models that characterize that behavior are very slow. We're developing a deep learning model that should be able to predict the plasma's behavior in a split second, but this is still a work in progress. Yu and doctoral student Jianke Yang in her office at UCSD. Photograph: Peggy Peattie for Quanta Magazine Where did your AI Scientist idea come from? In the past couple of years, my group has developed AI algorithms that can automatically discover symmetry principles from data. For example, our algorithm identified the Lorentz symmetry, which has to do with the constancy of the speed of light. Our algorithm also identified rotational symmetry—the fact, for example, that a sphere doesn't look any different regardless of how you rotate it—which is something it was not specifically trained to know about. While these are well-known properties, our tools also have the capability to discover new symmetries presently unknown to physics, which would constitute a huge breakthrough. It then occurred to me that if our tools can discover symmetries from raw data, why don't we try to generalize this? These tools could also generate research ideas or new hypotheses in science. That was the genesis of AI Scientist. What exactly is AI Scientist—just a fancy kind of neural net? It's not a single neural network, but rather an ensemble of computer programs that can help scientists make new discoveries. My group has already developed algorithms that can help with individual tasks, such as weather forecasting, identifying the drivers of global temperature rise, or trying to discover causal relationships like the effects of vaccination policies on disease transmission. We're now building a broader 'foundation' model that's versatile enough to handle multiple tasks. Scientists gather data from all types of instruments, and we want our model to include a variety of data types—numbers, text, images, and videos. We have an early prototype, but we want to make our model more comprehensive, more intelligent and better trained before we release it. That could happen within a couple of years. What do you imagine it could do? AI can assist in practically every step of the scientific discovery process. When I say 'AI Scientist,' I really mean an AI scientific assistant. The literature survey stage in an experiment, for example, typically requires a massive data-gathering and organization effort. But now, a large language model can read and summarize thousands of books during a single lunch break. What AI is not good at is judging scientific validity. In this case, it can't compete with an experienced researcher. While AI could help with hypothesis generation, the design of experiments and data analysis, it still cannot carry out sophisticated experiments. How far would you like to see the concept go? As I picture it, an AI Scientist could relieve researchers of some of the drudgery while letting people handle the creative aspects of science. That's something we're particularly good at. Rest assured, the goal is not to replace human scientists. I don't envision—nor would I ever want to see—a machine substituting for, or interfering with, human creativity. Original story reprinted with permission from Quanta Magazine, an editorially independent publication of the Simons Foundation whose mission is to enhance public understanding of science by covering research developments and trends in mathematics and the physical and life sciences.
