1THE TWIN HARD PROBLEMS
I GO TO A LOT of physics conferences, where I learn the latest about black holes, the Higgs boson, dark matter, and the deepest workings of the natural world. But about ten years ago, I started seeing an unexpected topic on the conference agendas: the mind. In the evening, when attendees gathered for drinks or dinner, I wouldn’t have to wait long before physicists brought up the topic of consciousness. This was new: physics has sought for ages to get our minds out of the picture—to transcend our everyday experience and reveal how puny humans are compared to the vastness of the cosmos.
People sometimes told me that an intensely personal experience had awakened their interest. The young Italian physicist Giovanni Rabuffo got his PhD in 2018 in quantum gravity—the branch of theoretical physics that seeks to comprehend the nature of space and time and, with them, the origins of the universe. He was first drawn to physics as a teenager growing up in a hill town southeast of Rome. “It’s so abstract, it’s so precise, and it really goes down to deep things,” he said. “It’s reshaping philosophy. It’s saying things about nature you could not discover using normal reasoning.”
In 2013, when he was twenty-three and studying for his master’s degree at the University of Pisa, Rabuffo heard about lucid dreaming: dreams when you know you’re dreaming. Curious to experience this phenomenon himself, he found a how-to guide that advised him to start practicing meditation, and over time he learned to quiet his thoughts. Although Rabuffo never hid his side interest from his classmates, neither did he go around advertising it. “In the field of physics, sometimes I found it hard to transmit this passion, this curiosity,” he said. “I found, in some people, a wall to these arguments. Not everyone: the community, in my experience, is really split.” His girlfriend couldn’t understand why anyone would be interested in lucid dreaming, he recalled: “She was like, ‘So what?’ So we broke up.”
One day while he was lying in bed, Rabuffo realized he was dreaming. Excited that he might finally be having a lucid dream, he imagined thrusting his arms out from his body—whereupon he found himself outside his body. This didn’t feel like a dream at all. “It was extremely, extremely realistic, like I was really there, awake,” he told me. Rabuffo saw that his room was dark, but dimly lit by a blue light. As he began to navigate his environment, he found the source of the light: “I moved toward the mirror, and I saw in the mirror, there was not me, there was this light moving,” he said. “As I was moving, the light was approaching the mirror, so I understood that that light was me.” He went to the door, tried to open it, and heard the handle make a rusty sound, before feeling himself pulled back to the bed and into his body. The experience lasted maybe a minute.
Rabuffo continued with his studies and eventually relocated to Marseille, France, to finish his PhD. But he couldn’t shake his fascination with what he’d experienced. He began knocking on neuroscientists’ doors until a leader of the European Human Brain Project offered him a postdoctoral position, where he has found that neuroscientists value his physics expertise in coping with a flood of data. “They need that mathematics,” he said. At the same time, he isn’t presumptuous: “The best quality you can have as a physicist is an open mind. You don’t pretend to know everything.” Rabuffo hopes one day to be able to study why the brain produces sensations such as his. “It is surprising how frequent are these unconventional experiences—and how much they are disregarded,” he said.
* * *
RABUFFO’S OUT-OF-BODY EXPERIENCE may be unusual, but he is hardly the only physicist in recent years to have pivoted from studying the workings of the cosmos to contemplating the intricacies of brains—both natural and artificial. Lenka Zdeborová, a Czech-born theorist who is now at the Swiss Federal Institute of Technology in Lausanne, specializes in statistical physics, which looks at the behavior of large groups of particles—billions of particles or more. It comes as no surprise that these vast hordes are complicated. What is much stranger is that their complexity begets simplicity. Through a marvelous and only hazily understood propensity for self-organization, particles spontaneously arrange themselves into collectives: crystals, gases, glasses, and other forms that are being discovered all the time.
In 2015, when Zdeborová was working on her application for a grant that required recipients to take their careers in an entirely new direction, she read about the renaissance of artificial intelligence (AI) after decades of dashed hopes. She thought back to playing chess as a teenager in the ’90s and watching IBM’s Deep Blue chess computer beat the human champion Garry Kasparov. Deep Blue, which relied on rules that had been painstakingly programmed into it, was the crowning achievement of traditional AI methods.1 But in a strange way it was a letdown. The machine worked largely by brute force, and its victory, though certainly impressive, was no more surprising than a computer calculating π to a trillion places. For real insight into the nature of intelligence, most researchers in the ’90s and early 2000s thought they were better off looking to the ancient Chinese strategy game of Go—now that’s a game you couldn’t mechanize so easily. It has too many possible moves. Playing it takes creativity and high-level thought, which seem quintessentially human. The idea of a computer “beating people in Go was for me like saying one day we will have power plants with nuclear fusion: that’s always fifty years ahead,” she said.
But as Zdeborová was preparing to give her oral presentation for the grant, Google DeepMind’s AlphaGo program thrashed some of the world’s top human Go players.2 These victories marked the coming of age of an AI technology based on neural networks, which learned by doing—as a human does—rather than being programmed with rules as Deep Blue was. What gives these systems their almost humanlike (if still narrow) abilities? Raw computing power is part of the answer, but only part. “Fundamentally, we still don’t know,” Zdeborová said. “And so this makes it very interesting. It’s the kind of puzzle that physics loves.” She made such a strong case for crossing disciplines in her presentation that she convinced not only the grant agency but even herself. “Some of the criteria that you have in grant writing are purely annoying. You write something around them, and you’re not really honest. But sometimes it actually helps to find the right direction for you,” she said.
Like Rabuffo, she was acutely aware of physicists’ reputation for barging into other fields, assuming they already know it all. “Physicists are infamously known to do that. Sometimes we are too arrogant!” she told me. So she was careful to spell out exactly what she thought physics has to offer neural network research.
Neural networks can contain billions of computing units—their “neurons,” so to speak. Billions of neurons, billions of particles: they’re not so different. Both neurons and particles interact with others of their kind. Particles attract or repel one another magnetically or electrically; artificial neurons fire off signals through wires, natural ones through axons. One particle might flip another upside down; one neuron might trigger another to start firing. The physical details of these interactions differ, but at an abstract level these particles and neurons are doing exactly the same thing: organizing and reorganizing en masse. “The way we train these neural networks, they really are systems of particles,” Zdeborová said.
Systems of particles and systems of neurons are also alike in their inscrutability. It is hopeless to try to track each and every molecule in a roomful of air (hence the “statistical” in statistical physics, which describes the behavior of particles in terms of probabilities). Neural networks, too, are so enormous that you can’t predict with absolute certainty what they’ll do. This makes them like humans—and this isn’t necessarily what their users want. Because they are too complex to program in the traditional way and must instead be taught, they can be misled. Set loose to learn from information on the internet, for instance, they absorb all its racism and sexism.3 They have been exposed to enough human psychology to be capable of manipulating us, as we are now beginning to see with AI-powered chatbots.4 For these reasons, achieving some deeper understanding of neural networks is essential to designing and using them wisely. “Without really new ideas, just with pure engineering power, there are things we will not overcome,” Zdeborová said.
As chapter 2 will explore, the methods of physics readily carry over to neural networks. You can conduct experiments on networks as if they were gases or crystals and discover the laws governing their behavior. “It’s so complicated, nobody understands,” Zdeborová said. “So it’s the same as if it were real nature. So it’s an object for physics. It’s too complicated to understand elementarily. We really need to look at [a neural network] as a product of nature and treat it as if it was a physical experimental system.” She and others seek a general theory of intelligence that will apply not just to artificial brains, but to ours as well.
Lots of young physicists are following in her footsteps. “I’m very overbooked with student interest,” Zdeborová said. “They just tell me, ‘Oh, I fell in love with the subject. I don’t know really why.’” It doesn’t hurt that students see AI as a brighter career path. The physics life is hard: jobs are scarce, hours are long, progress takes decades. There’s a long history of physicists colonizing other fields for lack of opportunities in their own.5
One of the highest-profile physicists to change the direction of his career is Max Tegmark, a cosmologist at the Massachusetts Institute of Technology (MIT). I got to know him in 1998, when he was a postdoc analyzing measurements of the primordial universe. Later we worked together on an article in Scientific American arguing that our universe is only one among many—that we live in a vast, perhaps infinite, multiverse.6
During an afternoon coffee break a few years ago, Tegmark told me: “When I was a teenager and I realized I loved mysteries—the bigger they were, the more I really loved them—it felt to me that the two greatest mysteries of all were our universe out there and the universe in here.” He tapped his forehead. Tegmark had devoted the first twenty-five years of his career to the former universe because it seemed premature to tackle the latter. Now, consciousness—specifically what philosophers call phenomenal consciousness, the nature of our subjective experience—strikes him as ripe for progress. “It doesn’t feel premature anymore,” he said. “It feels like cosmology’s peaked in some ways.”
Tegmark, who has been applying cosmological data-analysis techniques to brain imaging, thinks scientists can begin to gain traction on consciousness through integrated information theory, which contends that the brain is conscious to the extent that its parts act together in harmony. (Chapter 3 will delve into this theory.) The cerebrum’s vast network of neurons works as a unified whole, fusing sights, sounds, and memories into a seamless field of experience. Its cohesiveness is not unlike the collective order that Zdeborová finds in particle systems, and Tegmark thinks the same mathematical methods that describe those systems should apply to the brain, too. “There’s probably some equation: if information processing obeys this, there’s an experience; otherwise not,” he said.
Lest you think he has neglected the big picture, Tegmark thinks of intelligence as a cosmic phenomenon. Like the late theoretical physicist Freeman Dyson, he believes that in the far future, our distant descendants may be an astrophysical force on a par with natural ones.7 So, cosmologists who make predictions about the fate of the universe need to consider the goals and abilities of intelligent beings. But Tegmark is worried this may be a moot point if humanity can’t survive the many existential perils it has created for itself, such as nuclear war and superintelligent robots. (He’s also concerned that we may not have seen any extraterrestrial civilizations because they all blew themselves up—not a promising precedent for us.) Tegmark notes that scientists have expertise on these threats and, to be frank, have played a role in creating them. “We therefore have a special responsibility to weigh in to counterbalance ‘fake news’ and ‘alternative facts,’” he said. In 2014 he cofounded the Future of Life Institute to that end.
These physicists still consider themselves physicists. They don’t feel that they have left the field, just that they are pursuing it by other means. Not only do they think they might be able to help neuroscientists and philosophers of mind, they also think researchers in other disciplines can help them. As we’ll see, the latest advances in physics present scientists with a paradox: We can’t understand the measurable, material universe beyond our minds without first understanding our minds. Physics seeks objective reality, but can’t get away from the subjective element. As Tegmark put it: “If you look at the problems that we’re still stumped on in foundational physics, pretty much all of them trace back to consciousness.”
A “TICKING TIME BOMB”
Physics is the science of hard, elemental stuff, while the mind is messy and mushy—it can’t be captured in an elegant equation or graphed using x and y axes. You can write the equations describing the behavior of light in a few lines and derive all the principles of optics from them; at the same time, you can read a thousand-page novel and still not feel you understand the characters, or spend your whole life in therapy and still not fully know yourself. Physicists have traditionally left subjective experience to psychologists, poets, and pastors, and those specialists, in turn, have had little use for physics.
This disciplinary divide isn’t temporary or easily overcome; it is inherent in the four-hundred-year-old trade-off that opened the way for modern science. Physics was born from the split between mind and matter. The scientific revolutionaries of seventeenth-century Europe, notably Galileo Galilei and René Descartes, defined their domain as what is externally observable and quantifiable.8 Basically, that meant motion: the paths of cannonballs, planets, pendulums, and so on. You can measure them, plot them, mathematize them. The study of motion, known as mechanics, is still the first thing that physics students engage with. When I was in college, I actively resented that. I wanted to get to relativity theory, quantum fields, the big bang—the juicy stuff. But with age I have come to appreciate the simple things. Swing a pendulum or throw a stone in a pond, and you are already demonstrating essential concepts of advanced physics: oscillation, momentum, energy. The genius (or luck) of Galileo and Descartes was that almost the entire physical universe can be analyzed in terms of movement.
These early scientists weren’t uninterested in the functioning of the mind—Descartes was as much the founder of cognitive science as of physics9—but they recognized it would be harder to explain, and so effectively bracketed it. This split enabled the divide-and-conquer strategy that made science so successful. But the split between subjective and objective was also, in the words of the University of Toronto philosopher William Seager, a “ticking time bomb,” because some things just can’t be conquered by dividing them.10 Clearly, any investigation of the brain requires scientists to integrate mind with matter. Less obviously, so do the problems in foundational physics Tegmark was alluding to, in which our understanding of matter hinges on the nature of consciousness. Scientists could kick these interdisciplinary problems down the road for only so long.
For today’s physicists, nothing forces the issue more than the puzzles of quantum mechanics. Quantum theory underpins our modern description of matter. It governs everything from DNA mutations to supernova explosions and enables technologies from transistors to lasers. No exception to quantum mechanics has ever been found. But there is a troubling superficiality to its success: scratch the surface, and the theory makes no sense. Put three physicists in a room, and you will get four ideas about what it means. I know, because I’ve been in many such rooms. Once I was at a formal dinner with a panel of Nobel laureates and other distinguished physicists and philosophers who had gathered to debate the meaning of quantum mechanics; some had flown halfway around the world for the occasion. But they were so divided that they ended up talking mostly about international economic development instead. Consider how divisive quantum mechanics must be in order for politics to be the more polite subject!
Copyright © 2023 by George Musser