The Brain That Never Touches the World

Lisa Feldman Barrett is a professor of psychology at Northeastern University, author of *How Emotions Are Made*, and, at the time of this recording, about to publish *Seven and a Half Lessons About the Brain*. Lex Fridman, who once invited her to an AGI series at MIT and ended up in conversation instead of a lecture, opens with the wildest question he can think of: is there intelligent life elsewhere? Barrett says the probabilities suggest yes, and she privately hopes so. Being alone in the universe would seem, to her, profoundly sad.

She will not call the human brain the fanciest one we have. An octopus brain is different and, in its own way, spectacular: it can grow back a limb, change color and texture, and squeeze itself into a crevice. Those are superhero abilities people invent in stories and cannot do. Intelligence, she argues, is not one form and does not require one architecture. If evolution ran again, the menagerie would look different and still be interesting.

What humans have is not a special size. Grow a chimpanzee to human scale and you would get a human-sized brain; the embryonic blueprint that builds us builds other mammalian brains, and maybe vertebrate brains generally. The cerebral cortex is larger, and people mistakenly treat it as the seat of rationality, even though plenty of clever behavior happens in animals that do not have one. That extra cortex, Barrett says, is what lets us do more than use tools. We draw a line in the sand and make countries. We invent citizens and immigrants — ideas imposed on reality that then have serious physical effects on actual people.

Plenty of animals regulate one another's nervous systems, mostly by chemistry, scent, sound, vision, or touch. Humans, as far as she knows, are the only species that do it with ideas and words. She can text someone halfway around the world, without voice or face, and change that person's heart rate and breathing. Words, in that sense, are a way of doing mental telepathy. The network of brains is not a canvas on which memes breed by themselves. Ideas are invented and propagated by human brains, and they land on other bodies.

Plato described the psyche as two horses and a charioteer. One horse stood for instincts — feeding, fighting, fleeing, reproduction. The other stood for emotions. The charioteer was rationality, tasked with controlling the beasts. Fast-forward to the middle of the twentieth century and the metaphor was recoded as neuroscience: a reptilian core for instincts, a mammalian limbic system that bestowed emotion, and, wrapped on top, a cerebral cortex that gave primates, and especially humans, reason.

Barrett's objection is not a private taste. As far back as the 1960s, and clearly by the 1970s, molecular genetics showed that brains did not evolve by stacking new layers on an inner lizard. The irony is almost too neat. Carl Sagan popularized the three-layer story in *The Dragons of Eden*, which won a Pulitzer Prize in 1977 — a year when evolutionary neuroscientists already regarded the whole narrative as a myth. The inner beast that 'hijacks' you was already known to be a story, and it became the story anyway.

Fridman asks whether the lizard brain is at least a useful conversational tool, the way people blame Twitter outbursts on their inner reptile. Barrett does not think a false story is useful. Philosophy and science, as she uses them, are tools for living: they have to help you make good decisions. The triune brain is the foundation of the law in most Western countries and a foundation of economic theory. It also does two pieces of cultural work she wants undone. It lets people off the hook for nasty behavior. And it implies that emotions cannot be a source of wisdom, when in fact you would not want a person without emotions deciding your outcome. That person, she says, is a psychopath.

Even the backup concepts are shaky. Look at animals in their ecological context and so-called instincts are highly sensitive to circumstance. A physician tapping the patellar tendon produces a kick, but the force of that kick is influenced by all kinds of things. A reflex is not a robotic response. The better picture, she says, is the one that matches the science, and it is counterintuitive enough that she found it independently in neuroanatomy and in electrical engineering — literatures that do not speak to each other and were pointing the same way.

It feels as if the eyes are windows and the world arrives as a stimulus. In psychology that is the stimulus-response picture: a face, a voice, then a reaction, sometimes automatic, sometimes checked by something like System 2. Barrett says that is not how a brain works. It is predicting all the time, talking to itself, talking to the body, guessing what will happen next inside the skin and outside it. Information from the body and the world mostly confirms or corrects those guesses.

The romantic version is also the accurate one. "Your brain is trapped in a dark, silent box" — the skull. It never meets the world. It receives a continuous stream of sense data, including interoceptive signals from the body that you are largely unaware of, and those signals are only the *effects* of causes. A flash of light, a change in air pressure, a tug somewhere in the viscera: the brain does not know what produced them. Philosophers call this an inverse inference problem. From effects alone, how do you know the cause, and what to do next to stay alive?

The brain has one other source of information: past experience. It can reconstitute old episodes in its wiring and combine them in new ways. Psychology has many names for that trick — memory, perceptual inference, simulation, concepts, prediction — but the job is the same. Stop the world at any instant and the brain is in a state, representing what it believes is going on, and preparing both the next action and the next experience. When the data arrive, they either confirm the plan or surprise it. Surprise, encoded, is learning. The internal model gets an update so the next prediction can be cheaper.

Predicting and correcting is metabolically efficient. Constant reacting would mean living inside overwhelming uncertainty, which a nervous system cannot afford. On this picture, a reflex is not a dumb circuit firing. It is a prediction the brain executes *without* checking the sensory data, because the potential cost of waiting — a threat to life, for example — is too high. Perception, Fridman suggests, starts to look like a feature riding on top of a giant predictor. Barrett is happy to take that implication seriously.

One way to size the predictor against the perceiver is to look at evolution. About 550 million years ago, Barrett says, the world was populated by creatures without brains — a biological claim, she jokes, not a political one. The modern lancelet, amphioxus, is thought to resemble a common ancestor shared with invertebrates. That animal had no brain, no head, no eyes, and no ears. It had an eye spot not for seeing but for entraining a circadian rhythm to light and dark, a bit of touch, and a vestibular cell so it could stay upright in the water.

When a backbone, a head, sense organs, and a full brain evolved, something had made eyes and ears worth the cost. The pressure commonly entertained now is predation. Once an animal began deliberately eating another animal, an arms race followed, and it became useful to look ahead and ask a brutal first social question: should I eat that, or will it eat me? Sense data, on this view, did not evolve for consciousness or for the luxury of experiencing the world. They evolved in the service of motor control.

Amphioxus itself is almost a parable of life without intention. It plants itself in the sand like a living blade of grass and filters whatever drifts into its mouth. When the concentration of food drops, it wriggles at random to another spot that will, probabilistically, be richer. Barrett would not call that intentional action in the usual sense. Perception was never a gift for sitting still and taking in the scenery. It was a way to move in time.

Asked about free will, Barrett does not pretend to have a laboratory answer. It is a hard question and scientists do not know. She has taken a side — she thinks she has free will — but she puts little stock in anyone's intuitions about causes, including her own. One thing we know, she says, is that a brain creates experiences that lure you into believing those experiences reveal how the brain works. They do not. Dan Dennett's reminder still stands: people are obviously making choices; we are not amphioxus.

The version of free will she thinks is worth arguing about is not a ghost in the machine. You were born into a physical and social surround, and your brain wired itself to that surround. You were handed an internal model. You are not responsible for the model your caregivers cultivated. You are responsible for the one you have now. As you grow, you gain some control over where you are and what you do. You can cultivate new experiences, and you can practice them until they become automatic — until the brain can make them again more cheaply. Not everyone has choice over everything, but everybody has a little.

A colleague put the same thought in a sentence Barrett calls poetic: you are continually cultivating your past as a means of controlling your future. You cannot reach back and rewrite what happened, even if psychotherapy tries. You can change the present, which becomes the past the predictor will use tomorrow. Fridman hears in this a constant rewriting of one's narrative. Barrett does not flinch from that awe; she also refuses the sad picture of a clockwork with no magic in it.

There is still plenty we do not understand. Neurons are not purely digital; there is analog communication as well as axonal firing, and there is noise, there on purpose. The more variability a system has, the more information it can bear, and some animals even have clusters of cells whose only job is to inject noise into neural patterns. Stochasticity is one candidate source of the freedom worth wanting. Another is conceptual combination: the brain can take bits and pieces of the past and assemble them so you make sense of something you have never met. In that sense a brain does not merely store information. It is, she says, information gaining meaning. It can generate new information.

If the brain only ever sees effects, is anything real? Barrett will not take the question without a joke about passing a joint, but her answer is empirical. The information in the world and in the body is not random. It has spatial and temporal structure, and that structure wires the brain. An infant brain is not a miniature adult brain. It is waiting for wiring instructions. The retina, which is part of the brain, must be stimulated with photons; without that, the baby will not see in a neurotypical way. The same is true of hearing and of the other senses. As she writes in the book, we have the kind of nature that requires nurture.

In humans the required input is not only physical. A typical infant brain needs eye contact, touch, certain smells, cuddling. Cultures differ in how they care for babies, and that variety is an advantage for the species even when it does not feel like one. Strip the social extras away and merely feed and water a child, and what you get is a profoundly impaired human. Change the statistical regularities later in life and the brain can still learn, though learning is expensive and slow to automate. Restrict the diet of input — stay home, solitary confinement, an impoverished nursing home, the loss of a sense — and you take a risk. Human brains thrive on variety. The only diversity we consistently mishandle, she says, is one another.

We are also, obviously, not perceiving everything there is. That is why scientific tools exist: to get past the limited slice our sense organs allow. Whatever we do experience, humans tend to treat as natural and obvious, and to dismiss the rest. Many of the things we call natural we created. They are real, they have real effects, and we made those effects. Outside awareness, plenty more is still steering what we feel and do.

The executive control network that helps select among possible predictions runs all the time, whether or not you feel in control, and it relaxes in sleep. When you sleep, the external world recedes, the tethers loosen, and the internal model runs with fewer constraints from the immediate scene. That is why you can fly in a dream. Barrett reports a personal detail she cannot explain: she is much faster flying on her back than on her front. Psychedelics, she guesses from what she knows rather than from a study of her own, cut the tethers more completely. That is why a guide helps. The guide is a thin cable of sense data, so the unleashed model does not go entirely off the rails.

The Western caricature is that you are born with an inner beast: circuits for anger, sadness, and fear, interestingly all with English names, baked in and waiting to be triggered. When the fear circuit fires, you widen your eyes, gasp, speed the heart, and prepare to flee or freeze. That is the view in the law, and it treats emotion as an obligatory, mostly unhelpful reflex. Barrett has reviewed thousands of papers and run large meta-analyses. Measure faces, posture, peripheral physiology, brains, or voices — pick any output — and the evidence does not line up with that caricature. It lines up with itself. It just refuses the inner-beast story.

She has also sent teams into small-scale cultures far from urban laboratories. One visit was to the Hadza, hunter-gatherers in Tanzania, a modern people living in conditions more like those of our foraging ancestors, which is when emotional expressions were supposed to have evolved. She says "we" euphemistically. She had two research permits and gave them to students, staying in daily contact by satellite phone so they could have the formative experience. For years she lived with a contradiction. She feels emotion. She perceives emotion in other people. Scientists cannot find a single biomarker, or even a pattern of measures, that predicts what emotional state someone is in.

The hypothesis she came to is that the brain constructs these episodes from more basic ingredients. Figuratively, it asks: the last time I was in this situation, in this physical state, what did I do, see, and hear next? It is asking what in the past is similar to the present. Things similar to one another form a category. A mental representation of a category is a concept. The brain is constructing concepts continuously, on the fly. Machine-learning classifiers miss the point, she says, because the brain is not classifying. If the past it reaches for is made of emotion, the concept it builds is an emotion — and the contents change with the situation. Anger is a population of variable instances. Sometimes you scowl, sometimes you smile, sometimes you cry. Heart rate goes up, down, or nowhere, depending on the action you are about to take.

Anatomy suggests a specific order. The first prediction is visceromotor: how to change the heart, the lungs, a flush of cortisol. Cortisol, Barrett insists, is not a "stress hormone." It is a hormone that gets glucose into the bloodstream quickly because the brain is predicting a metabolically expensive act, either moving or learning. From those motor and visceral forecasts the brain infers what you will sense. Your experience of the world and of your own body is a consequence of those predictions, not their cause. When the concept being built is an emotion, that is how an instance of emotion is made.

Ask someone for the concept of a bird and you will get a list of features. Cognitive science used to agree: a concept was a dictionary definition, the necessary and sufficient features of a classical category. Then came the obvious trouble — not all birds fly, not all birds have feathers — and the prototype: one typical instance, with other members similar by degrees. Experiments later suggested something more situational. In a pet store, "bird" tends to mean a good pet. In a restaurant it tends to mean chicken. In an American park, a sparrow or a robin; in parts of South America, perhaps a peacock. The brain conjures a concept on the fly to meet the function the category is being put to.

Then there are ad hoc, or abstract, concepts, whose members share no physical features at all. Think of everything that can protect you from the rain: an umbrella, the apartment you are sitting in, even the thought that the rain is about to stop. The function is the same. Darwin's less celebrated move, Barrett argues, was to see that biological categories work this way too. Before *On the Origin of Species*, a species was a classical category, and variation from the perfect instance was error. Darwin said the variation is the point. Natural selection has nothing to work on without it. A species is a conceptual category held together by function, which is why modern arguments over "what is a species" never quite close. There is no single human genome. There is a population of genomes.

Civilization is built out of the same trick. What physical feature do all currencies share? Salt, barley, shells, immovable rocks in the Pacific, bitcoin, plastic cards, mortgages that are only a promise about the future: we impose value and we agree they can be exchanged for goods. The number of variables is not infinite, because physical reality holds the trump card. You and Barrett can agree that you can fly, and you still cannot. You can agree a virus does not exist, and the virus is unimpressed. In the wide band between those walls, though, we take physical signals and give them meanings they do not have by nature. Emotions, she says, are like money, like countries, like kings and presidents. Because we agree, the function is real, and so are the consequences.

That agreement is curated, not discovered. As infant brains wire to their physical and social world, caregivers bootstrap emotion concepts into them the way they bootstrap *dog*, *cat*, and *truck* — sometimes with labels, often with mental words while a child throws Cheerios or will not sleep. Words are special for infants: even preverbal babies can use a label to group things that do not look alike and then be surprised when the next "same" thing fails to squeak. Move to a new culture and you have to acculturate emotionally, learning on the fly how to make sense of your own sensations and of an eyebrow or a tilted head. The categories converge inside a culture. Across cultures they can look very different.

In the 1990s Barrett's group asked men and women about their emotional lives. Women described themselves as much more emotional than men; men agreed. Then the team issued little Hewlett-Packard computers that fit in a palm, pinged people about ten times a day, and had them report how they felt — experience sampling. Over two weeks, men and women did not differ. Some people were treading water in a tumultuous sea of feeling; others floated on a lake; everyone in between. Sex was not the variable. At the end they asked again how emotional people thought they were. The answers had not moved.

The inferences we make about other people are more stubborn. Give a man and a woman the same facial movements and observers, of both sexes, tend to say he is reacting to the situation and she is having an emotion. Barrett wrote a *New York Times* op-ed before the second 2016 presidential debate applying that research. A woman in authority, especially, is in a catch-22: serious and she is a bitch; empathic and she is weak. Teaching evaluations show the same asymmetry. People expect a woman to be nurturant. A man who is nurturant gets points; a man who is not is not penalized. Fridman admits that Hillary Clinton annoyed him in ways the same traits would not have, in a man. Barrett's counsel is not to pretend the feeling is absent. It is to let consistent evidence organize your behavior anyway.

She applies the same medicine to herself. Ask her what regulates her eating and she will say carbohydrates: pasta, bread, a love she can feel. Research she admires, because it so violates her beliefs, finds that most studied animals — humans, crickets, baboons — eat to regulate protein, and will overeat carbohydrate to get enough. She still cannot feel the protein. She corrects her behavior toward it anyway. That, she says, is the idea behind bias training. If you have a deeply held belief and the evidence shows you are wrong, you are wrong. How confident you feel can itself be explained by science.

The general name for the failure is experiential blindness. If the brain cannot make a concept that anticipates the sense data, you do not see what is there. Barrett shows lecture audiences a blobby black-and-white image; they see nothing. She shows a photograph, then the blobs again, and now they see an object. The image did not change. People operated for congenital cataracts can have light reach the brain and still fail to see some things for days, weeks, or years. Empathy is the same mechanism pointed at a person. If you are not similar enough, you cannot predict what they feel. In the United States, children of color are under-prescribed medicine; male physicians miss things in female patients. Barrett has stories, she says, where people die of the wrong inference. A police officer in Texas wrote to her that *How Emotions Are Made* would help with the hardest encounters. Sensory sampling is yoked to the heartbeat. When heart rate goes through the roof, you are more likely to go with the prediction, to see poorly, or to see things that are not there.

Brains did not evolve so you could see, hear, feel, or hold office. They evolved to control a body. The scientific name for doing that predictively is allostasis: anticipate the body's needs and meet them before they arise. Barrett's metaphor is a body budget. The brain is not budgeting money. It is budgeting glucose, salt, and water across a gazillion accounts, deciding when to spend, when to save, and whether an outlay will pay a return. Talk of reward, reward-prediction error, and punishment is, in her translation, talk of deposits and withdrawals. When the budget runs a deficit, you experience discomfort and distress.

Jordan Theriault, a scientist in her lab, has a paper on the sense of *should*. Why social norms? Because if I make myself predictable to you, you become predictable to me, and predictability is metabolically cheap. Extreme novelty is expensive. In the short run you feel jittery, which we name anxiety even when it may only be arousal from the chemicals that help you learn. If the uncertainty lasts, you become depleted and then truly distressed. Depression and anxiety, on this view, are ways of being in the world when the predictions are not working. They are not a separate layer from sleep, food, and social connection. Treat them as purely mental and you ignore the variables that are actually moving.

She sees a perfect storm, not a single cause: not enough sleep, pseudo-food, social media stuffed with ambiguity and no context (hard work for a nervous system), and an economy that has left huge numbers of people one bill from the street. Listening to Robert Reich on oligarchy, she heard that CEO pay had moved from something like twenty times the typical employee to about three hundred times, the difference coming from workers' pockets. The workers do not know that story. They know they cannot feed children or pay for care. Then someone offers a narrative: somebody else butted in line. Empathy for people whose beliefs are unlike yours means foraging for unexpected information. That is expensive. When your metabolic account is already in deficit, you stop spending. For a brain, stopping spending means stopping the body and stopping learning, and going with the model you already have.

Barrett is not lecturing from a full tank. She gets up every morning and runs and lifts weights. She hates it. She does it religiously because it is a high-return expenditure, and she fuels before other costly work the same way — a friend, a night of sleep, not always a protein shake. When her daughter left for college she offered one piece of advice: try to be around people who let you be the kind of person you want to be. Kindness, the research she cites says, is a deposit in both body budgets. The best thing for a human nervous system is another human, and the worst thing is another human. You get to choose which one you are.

That is also her account of love. Babies cannot run their own body budgets; feeding and cuddling are metabolic caretaking that wires a brain to manage itself later. Humans evolved socially dependent. Manage the system alone and you pay a small tax that accrues into middle age as heart disease, diabetes, depression, a higher chance of Alzheimer's. Loneliness, she says, kills — on the order of seven years earlier; the exact figure lives in the web notes to the book. Lose someone you love and it feels as if you have lost a part of yourself because you have. "We are the caretakers of one another's nervous systems, like it or not." She did not, for a long time, fear dying; she feared pain. As a mother she fears dying before her daughter is ready. More than her own death she fears her husband's. The meaning of life, true to everything else she has said, is not one thing. It is a population of instances: to make meaning, to leave the world slightly better, to clear a path for a daughter or a student, or to be stopped by a weed in a sidewalk and feel, for a moment, the wonder of the physical world.