The Genius Neuroscientist Who Might Hold the Key to True AI By Shaun Raviv (@ShaunRaviv) (snipped) https://www.wired.com/story/karl-friston-free-energy-principle-artificial-intelligence/
Karl Friston’s free energy principle might be the most all-encompassing idea since Charles Darwin’s theory of natural selection. But to understand it, you need to peer inside the mind of Friston himself. When Friston was inducted into the Royal Society of Fellows in 2006, the academy described his impact on studies of the brain as “revolutionary” and said that more than 90 percent of papers published in brain imaging used his methods. Two years ago, the Allen Institute for Artificial Intelligence, a research group led by AI pioneer Oren Etzioni, calculated that Friston is the world’s most frequently cited neuroscientist. He has an h-index—a metric used to measure the impact of a researcher’s publications—nearly twice the size of Albert Einstein’s. Last year Clarivate Analytics, which over more than two decades has successfully predicted 46 Nobel Prize winners in the sciences, ranked Friston among the three most likely winners in the physiology or medicine category. What’s remarkable, however, is that few of the researchers who make the pilgrimage to see Friston these days have come to talk about brain imaging at all. Over a 10-day period this summer, Friston advised an astrophysicist, several philosophers, a computer engineer working on a more personable competitor to the Amazon Echo, the head of artificial intelligence for one of the world’s largest insurance companies, a neuroscientist seeking to build better hearing aids, and a psychiatrist with a startup that applies machine learning to help treat depression. And most of them had come because they were desperate to understand something else entirely. For the past decade or so, Friston has devoted much of his time and effort to developing an idea he calls the free energy principle. With this idea, Friston believes he has identified nothing less than the organizing principle of all life, and all intelligence as well. “If you are alive,” he sets out to answer, “what sorts of behaviors must you show?” First the bad news: The free energy principle is maddeningly difficult to understand. So difficult in fact, that entire rooms of very, very smart people have tried and failed to grasp it. But often those same people hastened to add that the free energy principle, at its heart, tells a simple story and solves a basic puzzle. The second law of thermodynamics tells us that the universe tends toward entropy, toward dissolution; but living things fiercely resist it. We wake up every morning nearly the same person we were the day before, with clear separations between our cells and organs, and between us and the world without. How? Friston’s free energy principle says that all life, at every scale of organization—from single cells to the human brain, with its billions of neurons—is driven by the same universal imperative, which can be reduced to a mathematical function. To be alive, he says, is to act in ways that reduce the gulf between your expectations and your sensory inputs. Or, in Fristonian terms, it is to minimize free energy. Friston became convinced that the best way to think of the brain was as a Bayesian probability machine. The idea, which goes back to the 19th century and the work of Hermann von Helmholtz, is that brains compute and perceive in a probabilistic manner, constantly making predictions and adjusting beliefs based on what the senses contribute. According to the most popular modern Bayesian account, the brain is an “inference engine” that seeks to minimize “prediction error.” Even Friston has a hard time deciding where to start when he describes the free energy principle. He often sends people to its Wikipedia page. But, it seems apt to begin with a Russian mathematician named Andrei Andreyevich Markov, who died in 1922. Markov is the eponym of a concept called a Markov blanket, which in machine learning is essentially simply a shield that separates one set of variables from others in a layered, hierarchical system. The psychologist Christopher Frith—who has an h-index on par with Friston’s—once described a Markov blanket as “a cognitive version of a cell membrane, shielding states inside the blanket from states outside.” In Friston’s mind, the universe is made up of Markov blankets inside of Markov blankets. Each of us has a Markov blanket that keeps us apart from what is not us. And within us are blankets separating organs, which contain blankets separating cells, which contain blankets separating their organelles. The blankets define how biological things exist over time and behave distinctly from one another. Without them, we’re just hot gas dissipating into the ether. Invisible cloaks around everyone, and underneath each one a different living system that minimizes its own free energy. The concept of free energy itself comes from physics, which means it’s a measurable quantity that can be modeled. Free energy is the difference between the states you expect to be in and the states your sensors tell you that you are in. Or, to put it another way, when you are minimizing free energy, you are minimizing surprise. According to Friston, any biological system that resists a tendency to disorder and dissolution will adhere to the free energy principle—whether it’s a protozoan or a pro basketball team. A single-celled organism has the same imperative to reduce surprise that a brain does. The only difference is that, as self-organizing biological systems go, the human brain is inordinately complex: It soaks in information from billions of sense receptors, and it needs to organize that information efficiently into an accurate model of the world. “It’s literally a fantastic organ in the sense that it generates hypotheses or fantasies that are appropriate for trying to explain these myriad patterns, this flux of sensory information that it is in receipt of,” Friston says. In seeking to predict what the next wave of sensations is going to tell it—and the next, and the next—the brain is constantly making inferences and updating its beliefs based on what the senses relay back, and trying to minimize prediction-error signals. So far, as you might have noticed, this sounds a lot like the Bayesian idea of the brain as an “inference engine” that Hinton told Friston about in the 1990s. And indeed, Friston regards the Bayesian model as a foundation of the free energy principle (“free energy” is even a rough synonym for “prediction error”). But the limitation of the Bayesian model, for Friston, is that it only accounts for the interaction between beliefs and perceptions; it has nothing to say about the body or action. It can’t get you out of your chair. This isn’t enough for Friston, who uses the term “active inference” to describe the way organisms minimize surprise while moving about the world. When the brain makes a prediction that isn’t immediately borne out by what the senses relay back, Friston believes, it can minimize free energy in one of two ways: It can revise its prediction—absorb the surprise, concede the error, update its model of the world—or it can act to make the prediction true. If I infer that I am touching my nose with my left index finger, but my proprioceptors tell me my arm is hanging at my side, I can minimize my brain’s raging prediction-error signals by raising that arm up and pressing a digit to the middle of my face. And in fact, this is how the free energy principle accounts for everything we do: perception, action, planning, problem solving. When I get into the car to run an errand, I am minimizing free energy by confirming my hypothesis—my fantasy—through action. For Friston, folding action and movement into the equation is immensely important. Even perception itself, he says, is “enslaved by action”: To gather information, the eye darts, the diaphragm draws air into the nose, the fingers generate friction against a surface. And all of this fine motor movement exists on a continuum with bigger plans, explorations, and actions. “We sample the world,” Friston writes, “to ensure our predictions become a self-fulfilling prophecy.” So what happens when our prophecies are not self-fulfilling? What does it look like for a system to be overwhelmed by surprise? The free energy principle, it turns out, isn’t just a unified theory of action, perception, and planning; it’s also a theory of mental illness. When the brain assigns too little or too much weight to evidence pouring in from the senses, trouble occurs. Someone with schizophrenia, for example, may fail to update their model of the world to account for sensory input from the eyes. Where one person might see a friendly neighbor, others might see a giant, evil crow. “If you think about psychiatric conditions, and indeed most neurological conditions, they are just broken beliefs or false inference—hallucinations and delusions,” Friston says. Over the past few years, Friston and a few other scientists have used the free energy principle to help explain anxiety, depression, and psychosis, along with certain symptoms of autism, Parkinson’s disease, and psychopathy. In many cases, scientists already know—thanks to Friston’s neuroimaging methods—which regions of the brain tend to malfunction in different disorders and which signals tend to be disrupted. But that alone isn’t enough to go on. “It’s not sufficient to understand which synapses, which brain connections, are working improperly,” he says. “You need to have a calculus that talks about beliefs.” So: The free energy principle offers a unifying explanation for how the mind works and a unifying explanation for how the mind malfunctions. It stands to reason, then, that it might also put us on a path toward building a mind from scratch. According to the free energy principle, the fundamental drive of human thought isn’t to seek some arbitrary external reward. It’s to minimize prediction error. Clearly, neural networks ought to do the same. It helps that the Bayesian formulas behind the free energy principle—the ones that are so difficult to translate into English—are already written in the native language of machine learning. Julie Pitt, head of machine-learning infrastructure at Netflix, discovered Friston and the free energy principle in 2014, and it transformed her thinking. (Pitt’s Twitter bio reads, “I infer my own actions by way of Active Inference.”) Outside of her work at Netflix, she’s been exploring applications of the principle in a side project called Order of Magnitude Labs. Pitt says that the beauty of the free energy model is that it allows an artificial agent to act in any environment, even one that’s new and unknown. Under the old reinforcement-learning model, you’d have to keep stipulating new rules and sub-rewards to get your agent to cope with a complex world. But a free energy agent always generates its own intrinsic reward: the minimization of surprise. And that reward, Pitt says, includes an imperative to go out and explore. In late 2017, a group led by Rosalyn Moran, a neuroscientist and engineer at King’s College London, pitted two AI players against one another in a version of the 3D shooter game Doom. The goal was to compare an agent driven by active inference to one driven by reward-maximization. The reward-based agent’s goal was to kill a monster inside the game, but the free-energy-driven agent only had to minimize surprise. The Fristonian agent started off slowly. But eventually it started to behave as if it had a model of the game, seeming to realize, for instance, that when the agent moved left the monster tended to move to the right. After a while it became clear that, even in the toy environment of the game, the reward-maximizing agent was “demonstrably less robust”; the free energy agent had learned its environment better. “It outperformed the reinforcement-learning agent because it was exploring,” Moran says. In another simulation that pitted the free-energy-minimizing agent against real human players, the story was similar. The Fristonian agent started slowly, actively exploring options—epistemically foraging, Friston would say—before quickly attaining humanlike performance. Active inference is starting to spread into mainstream deep-learning research, albeit slowly. Some of Friston’s students have gone on to work at DeepMind and Google Brain, and one of them founded Huawei’s Artificial Intelligence Theory lab. “It’s moving, but it’s still not nearly as common as reinforcement learning, which even undergraduates learn. “You don’t teach undergraduates the free energy principle—yet.” The first time I asked Friston about the connection between the free energy principle and artificial intelligence, he predicted that within five to 10 years, most machine learning would incorporate free energy minimization. The second time, his response was droll. “Think about why it’s called active inference,” he said. His straight, sparkly white teeth showed through his smile as he waited for me to follow his wordplay. “Well, it’s AI,” Friston said. “So is active inference the new AI? Yes, it’s the acronym.” Not for the first time, a Fristonian joke had passed me by. In 2017, Friston coauthored a paper, with Canadian philosopher Maxwell Ramstead and Paul Badcock of the University of Melbourne, in which they described all life in terms of Markov blankets. Just as a cell is a Markov-blanketed system that minimizes free energy in order to exist, so are tribes and religions and species. After the publication of the Ramstead paper, Micah Allen, a cognitive neuroscientist then at the FIL, wrote that the free energy principle had evolved into a real-life version of Isaac Asimov’s psychohistory, a fictional system that reduced all of psychology, history, and physics down to a statistical science. Friston says his work has two primary motivations. Sure, it would be nice to see the free energy principle lead to true artificial consciousness someday, he says, but that’s not one of his top priorities. Rather, his first big desire is to advance schizophrenia research, to help repair the brains of patients like the ones he once treated. And his second main motivation, he says, is “much more selfish.” It goes back to one evening in his bedroom, as a teenager, wondering, “Can I sort this all out in the simplest way possible?” “And that is a very self-indulgent thing. It has no altruistic clinical compassion behind it. It is just the selfish desire to try and understand things as completely and as rigorously and as simply as possible,” he says. == Cheers, Stephen _______________________________________________ Link mailing list [email protected] http://mailman.anu.edu.au/mailman/listinfo/link
