On a Perfectly Flat Floor, Heights Still Work

Lex Fridman opens with the setup Huberman has mentioned elsewhere: in the Stanford lab, stress is induced in virtual reality. Huberman will not let the word fear float. You cannot really have fear without stress, but you can have stress without fear; you cannot really have trauma without both, but you can have fear and stress without trauma. Those overlaps are one reason the lab exists. The field of emotions and states, he says, is messy, and it needs better physiological, neuroscientific, and operational definitions.

Their assignment rule is blunt. Fear, in this lab, is a big inflection in autonomic arousal: heart rate up, breathing up, perspiration, pupil dilation, the hallmark signature of the stress response. In some cases they have neurosurgery patients with chronically implanted electrodes in amygdala, insula, and orbitofrontal cortex, so they can see central features as well. What is almost trivial, and still striking, is the stimulus that works on almost everybody: heights, and falling from a high virtual place. Optic flow coupled to the vestibular apparatus and the semicircular canals pulls the physiology — stomach drop, sudden sweat — even when you know you will not fall off the platform.

Classic lab fear studies showed a picture of a bloody arm or a snake. Unless you already have a snake phobia, Huberman says, it does not create a real enough experience. He wants presence: a few moments of forgetting you are in a laboratory, without injury. They built 360 video, including a cage-exit dive with great white sharks shot at Guadalupe Island with Hollywood portrait photographer Michael Muller, because in 2016 a lot of VR was still lame CGI. Heights always do it. Generalized anxiety can be tipped by claustrophobia or public speaking. Huberman's sister will not even come to the lab because of the sharks.

In the actual protocol, subjects play a speeding cognitive game, then are suddenly placed on a narrow platform between two buildings and cued to keep going. Some people will not. They get down on the ground and hold on to a virtual beam that does not exist, on a flat floor. That, for Huberman, is the power of the brain to enter virtual states as if they were real, anchored by vision and balance. Closed-loop coupling — your own movements making things scarier — drives the nervous system further. They also study tools that let people calm themselves in those environments. The short answer to what works is still heights.

Asked what would torture him, Huberman does not lead with snakes. He never considered himself claustrophobic if the space was well ventilated. Then he started going to a Russian banya, underground, very hot and humid, in a city with earthquakes, and a few times imagined the place crumbling with them stuck inside. Heart rate went. He put himself into the lab's own elevator paradigm — stalled, crowded with virtual people, oxygen levels modulable — and found it really uncomfortable. He thinks he unhatched a bit of claustrophobia. Fridman went skydiving to beat heights and did not; the terror, he says, is the minutes before, when the thing is not supposed to happen and might. Once you accept the jump, the world goes briefly three-dimensional and the concept of ground disappears. He is still terrified of heights.

What it takes to overcome fear comes from two tracks, mice and humans. Graduate student Lindsay Salay, now at Caltech, published in 2018 that there are really three responses: pause (freeze), retreat, or go forward. A hub of neurons in the middle of the thalamus biases which of those a mouse — and, it turns out, a human — is likely to do. The physiology is the surprise. The lowest autonomic arousal is the pause. Retreat is a bit higher. The highest arousal is forward movement toward the threat. Hide-and-seek in a closet is easy when you are calm; as stress rises, stillness gets harder. A stalking cat's teeth-chatter is top-down restraint. Freeze is an active response, but it is fairly low stress.

Forward movement, if you survive it, is linked by a collateral — a side wire — into dopamine circuits for reward. Safely moving through or toward a threat gets tagged as a positive experience. That maps onto cognitive behavioral therapy and most trauma treatment: you have to confront the thing. Otherwise you sit in a low-level reverberatory loop, the arousal circuits humming louder. Stress and fear were designed to agitate so that you move. People often think the maximum fear response is to lock up. It is not. The maximum is to advance, and it has positive valence.

The usual inverted-U of arousal and performance, Huberman says, has another bump further out the distribution, where people perform very well under very high stress. Time perception changes; you evaluate the environment at a faster frame rate — the Matrix, as a lot of people think of it. VR is useful because you can push people hard without physically injuring them. That far bump is a small fraction of ordinary life. He thinks that is where optimal performance lies.

What, then, is performance, and what is optimal? Huberman grants that it is subjective and varies with task and environment. He makes it operational: there is a sweet spot where internal autonomic arousal — stress, alertness, whatever you want to call it — is ideally matched to the speed of the challenge outside. Perception of the outside world is exteroception; perception of internal real estate is interoception. When those are matched along a couple of dimensions, performance tends to sit in an optimal range.

Learning a new passage on guitar, which Huberman does not play, may not want a super-high activation. Internal arousal may need to sit where your analysis of space and time matches the incoming notes. Under high threat, when things come in fast and you must react fast, higher autonomic arousal is better, because you are slicing time more finely. Pupil dilation and lens movement change the optics, and with the optics comes a change in how you bend and slice time: more frames per second of readout. For guitar, you might want to be almost sleepy, relaxed enough that fingers can follow a purely external cue.

There is no perfect autonomic state for performance. That is why he does not favor terms like flow: they are not operationally defined enough. He does believe peak performance arises when internal state is matched to the space-time features of the external demands. You can slice time more finely or less finely to adjust to the dynamics of the stimulus. The same 'optimal' does not travel from a loom to a slow, precise skill.

Fridman steers from action to thinking: Cal Newport's deep work, holding a complicated object in mind, taking a step, discovering it is wrong, turning around without panic. Huberman maps some of that onto working memory — keeping things online and updating them. He quotes a coffee conversation with Jack Feldman at UCLA, who has a physics background and studies breathing as a ground state that modulates other states. Feldman's sketch: most great mathematics is done in the late teens and twenties because theorems put a huge demand on working memory, many plates spinning. Physics leans more on a library, material that has moved out of hippocampus into cortex. Biology has much longer arcs; people in their sixties and seventies still do terrific work from a deep knowledge base, even a remembered figure in a paper.

Everyone in the lab has a cliff, including SEAL Team operators. Cliffs show up as speed of processing or as how many things you can keep online. Fridman pushes back: the taxing step in engineering is not holding the known pattern, it is stepping back and asking whether a better algorithm exists — testing algorithms, not querying facts. Huberman agrees they can be correct about working memory without being exhaustive. Neuroscience, he says, is still in its infancy; we probably know one percent. When you are trying lots of different algorithms, you do not want very high autonomic arousal. Higher arousal makes the analysis of space and time more rigid. Creativity is a lava lamp: you do not know which direction the wax will go.

In drowsy and sleeping states, space and time become somewhat dislodged and fluid, which is why solutions arrive after sleep and naps. Psychedelics, he is careful not to recommend, were once thought to work by spontaneous bursting. Most of the 5-HT2A receptors that LSD and psilocybin hit sit in the thalamic reticular nucleus, an inhibitory shell around the thalamus that, in ordinary conversation, keeps you focused on what you are seeing and eliminates a lot of sensory traffic. Activate those receptors and the shell loses its ability to inhibit. The larger effect is increased lateral connectivity in layer five of cortex across areas. Vision's rigid space-time prediction, audition's slightly different one, smell's different one, start mixing. People talk about hearing sights. Dreams do a related job through ponto-geniculo-occipital waves: whole-brain activation, algorithms lumped together, a new possibility in mind.

Space and time need to be fluid *and* rigid for something meaningful to come out. Dreams are amazing, Huberman says, but if you cannot bring anything back, they are just amazing. A lot of people say they take psychedelics for plasticity and insight; he thinks they should be honest that they want a peak experience inside an amplified experience. Unlock space-time too far and you can come out with a fuzzy memory of transformation and no tools. His magic-wand tool, pitched to a group called the Liminal Collective, would be psychedelics with on-off switches: ten minutes deep, then snap into a linear state, extract, go back in. Hard to do pharmacologically in humans; more likely with VR, AR, and devices that bring in more of the somatic. David Spiegel's hypnosis work at Stanford is another candidate: narrow, tunnel-vision context, deeply relaxed, a strong state for inducing neuroplasticity.

Fridman treats himself as an n-of-one. After a traditionally defined full night of sleep — his days drift, he is single, he schedules no meetings, he has lived something like a 28-hour day — there is a window of effortless clarity, the deepest intellectual dives, maybe five or ten minutes, and in those minutes the temptation is to do something worthless like check social media. Huberman talks about titrating caffeine and background sound, then about a period that is not discussed enough: the transition out of sleep. A mentor he respects told him to be very careful about bringing in someone else's sensory experience early in the day.

He sleeps well but emerges slowly and needs a lot of caffeine. In sleep you expunge what you do not need from the previous day and run variations on the algorithms of your life, short timescale and long. Lateral connections in layer five of neocortex are very robustly active across sensory areas. That brain state would be useless while awake; you would be the person talking to yourself in the hallway about something no one else can see. The theory is that solutions reveal themselves in the early part of the day unless you interfere by immediately entering somebody else's space-time. Social media is a clean example: someone else becomes the conductor of your thoughts.

Allow the download, he says, and ask whether you are more in your head or in an interoceptive or exteroceptive mode, then plug into the work, and only then allow music, news, faces. A brilliant, odd professor at Berkeley would not look at faces in the early part of the day. He did not have the most rounded life; if the metric is cognitive performance, the practice could be very beneficial. Writers' habits often look like this. Whitney Cummings, a mutual friend, bought an island and lives on a cliff because she wants to control sound. Fridman does not want to spoil himself: he used to take that mind to coffee shops and train it to shut noise out. Completeness of life and what is best for work are not the same list.

Huberman reaches for a coined term, not a scientific one, because he thinks it captures the concept: limbic friction. The limbic system wants to pull you, typically toward reflexive behavior. Prefrontal cortex, through top-down mechanisms, has to suppress that and say no, we are not going to respond to the banging of coffee cups, we are going to keep focusing. Put yourself in a café and limbic friction is high: prefrontal cortex has to work hard. There is another side. When you are very sleepy, nothing is incoming, the room can be silent, and it is still hard to focus because you are drifting. Limbic friction is high for the opposite reason: autonomic arousal is too low. Netflix in the background or a looped song can boost alertness into the sweet spot.

He offers a spatial model: a glass barbell, one sphere of attention on what is happening with me, one on what is happening with you or the room. The barbell is not rigid. The globes can get bigger or smaller. Close your eyes and both orbs sit on internal state. Do the opposite and both sit outside; you are not thinking about heart rate or breathing. The more relaxed you are, the more freely those orbs move. The more alert you are, the more rigidly they are tethered. That was designed for threat: if something is coming to attack you, you are not going to notice that your breathing cadence is a little quick. Both orbs lock onto the threat, and behavior is driven by something external even when it feels internal.

Neurons, he says, only really do three things: they are sensory, they are motor, or they modulate. Models of attention and perception in 2020 say we have interoception and exteroception, strongly modulated by autonomic arousal, and that forming an optimal relationship to sleep, a threat, or coding means adjusting the internal space-time relationship to the external one. Pain pulls you interoceptive; salient things in the environment pull you exteroceptive. The practical drill is the one Fridman has already started: notice, on waking and while working, whether you are mostly inside or outside, how alert or calm you are, and watch that process through the day.

Interoception, defined cleanly, is awareness of anything within the confines or on the surface of the skin, including gut and liver sensors. Exteroception is anything beyond the reach of the skin: a bottle, a scent, a sound. Headphones make sound seem inside the head; a speaker in the room does not. Ventriloquism lives there. Greg Recanzone at UC Davis mismatched auditory and visual cues until a sound seemed to arrive from the corner of the room and hit you, and people recoiled. Everything in the brain is an abstraction. The space-time firing that encodes a face has no resemblance to the face. All neurons can do is fire in space and time.

Vision is tractable because you can control the stimulus. The retina is three layers of neurons at the back of the eye, about as thick as a credit card, and it is a piece of brain — a forebrain structure squeezed into the embryonic eye cups in the first trimester, bone forming around a hole for the optic nerve. For a thick-skulled mammal it is the only window. Birds and lizards can get light to the pineal; snakes have a hole. Some retinal neurons care only about luminance and tell the circadian clock the time of day, which then tells every cell in the body. Light in the eyes at the right times and many good things happen; light at random through the 24-hour cycle and many bad things happen, from cancer outcomes to diabetes. Those photon counters are intrinsically photosensitive ganglion cells, discovered by David Berson at Brown. They are not spatial. They do not think 'it is daytime.' They count.

The rest of the retinal neurons mainly signal things lighter or darker than background: pixels, little circles of dark or bright, red versus green or blue. That is basically the only information sent down the optic nerve — not bits of red traveling, but spikes in space and time. The outside world is converted into a language of a few syllables, shouted down the nerve, turned into Morse in the brain. Thalamus does similar work plus gating: faster or brighter than everything else goes up to cortex; more of the same red gets thrown out. In V1, Hubel and Wiesel's map, you get oriented lines built the unromantic way: dots in a vertical stack converge on one neuron. Show a black line and you can be sure of a vertical representation in V1. At that point it is concrete, not abstract.

Further up, it gets mysterious. By the fusiform face area, which Nancy Kanwisher at MIT has studied for much of her career, you find single neurons that respond only to your father's face or Joe Rogan's, regardless of orientation. You recognize a profile across a room. Challengers said there were no face neurons, only space-time shape cells. Kanwisher used greebles, faces morphed toward alien; the neurons mostly do not respond to greebles. Huberman's summary: the visual system stays fairly concrete until about V4, with its color pinwheels. Then you depart point-to-point wiring and abstract representations appear. Audition does the same walk from a tonotopic map to musical motifs. Memory space, creativity, love — your memories are not his, but somewhere early there are zeros and ones, and then you leave them.

The exception, Huberman wants to be clear, is that he has been talking mainly about neocortex, the six-layered sheet. Subcortical structures are a lot more like machines: plug and chug, heart rate, breathing, a neuron that responds to temperature on the top of the left hand. Drop an electrode into cortex, convert the recording to audio, and you hear hash, noise, until you find the stimulus that makes the cell crack like a bat hitting a ball. Drop into thalamus, hypothalamus, or brainstem, even with a grungy old tungsten electrode, and you never hear hash. Walk in front of the eye and the neuron goes; walk away and it stops; put your hand there again and it goes; you could do that for two days. Two plus two equals four. No abstraction.

That is why his question to people at Neuralink — he is a fan, no business relationship — is: when are you going to go subcortical? If you want to control an animal, you do not do it in cortex. Cortex is the abstract painting of a face; remove a piece and the picture may still be the picture. In subcortical areas a stimulating electrode can evoke an entire behavior or an entire state. Parkinson's improved by stimulating the subthalamic nucleus because it gates motor activation in predictable ways. Academic labs are still figuring the brain out; Neuralink picked a concrete endpoint, restore motion, restore speech, nothing abstract about that. A lot of layer 2/3 recordings exist because two-photon imaging still does not go deep without prisms and narrow endoscopes. Cortex is also scientifically comfortable: no one knows how six-layer neocortex works, so anything you see is novel. Thalamus and hypothalamus come with models you have to combat.

The eerie cortical demonstration is Bill Newsome at Stanford and Tony Movshon at NYU. Dots move up; neurons in area MT fire for up. Reduce coherence, firing drops. Lesion MT and the animal loses the discrimination. Then the eerie step: find a neuron that likes up, move the dots down, stimulate that neuron, and the animal reports — by lever — that the dots are moving up. Reality on the screen is down. Perception is up. Your perception of the outside world depends entirely on activation patterns of neurons. Donald Hoffman's evolutionary case that we may be almost wholly detached from physical reality is, Huberman thinks, a bit hardcore, but a useful thought experiment that cleanses confidence. Susumu Tonegawa and Mark Mayford's memory-replay work is the other shiver: capture the neurons that fired while an animal learned, then reactivate them all at once, slamming every piano key, and you still get the same behavior. In some structures the space-time sequence may not be what you think it is.

Could there be an experiment around consciousness? Huberman has three short answers, and he is not usually succinct. The cynical one: in the 1970s, 1980s, even early 2000s, very impressive speakers fell in love with the consciousness problem and overlooked that the technology was not there. He admires the love; they gleaned tremendous taxpayer resources for essentially nothing. He carves out Francis Crick, whose idea that the claustrum is involved he still thinks is great — a skewer of a structure almost nobody studied. The second answer is definition. Psychologists, he says, are smarter than a lot of hard scientists here: they put operational definitions on motivation so two labs know they are studying the same thing. No one agrees what consciousness is. Attention had the same fight in the early 2000s. As a young graduate student he decided he was definitely not going to work on attention or consciousness. He wanted to see the circuit, hear it on the audio monitor, do gain and loss of function.

From special operations he borrows language that is not abstract: three-meter targets, ten-meter targets, hundred-meter targets. You do not pick the hundred-meter target because it is more beautiful. If you do not take down the three-meter and ten-meter targets first, you are dead. Scientists, he thinks, could stand to adopt more of that thinking. Falling in love with an idea, even a molecule that later proved moot, is not mostly ego; there is not enough money in science for cartoon villainy. Neuralink's advantage, as he reads it, is a team that does not need first-author papers or a completed Ph.D., with a neurosurgeon whose failures are not tolerated, mixed with high-level theorists, aimed at a mission. Fridman wants to sit in the consciousness room anyway, the way Turing tried to make intelligence concrete. Huberman's plea is a tractable definition. People who like the high-level stuff should sit in it uncomfortably until they ground it; people who only trust what they can touch should sit in the high-level conversation too. What is missing is talk at multiple levels.

The last ten years in neuroscience, he says, were a huge influx of tools, circuits, connectomes — wonderful, and a bit like demonstration day at the science fair. Ten or twenty years ago people went to meetings and discussed ideas and models. He is grateful theorists, psychologists, even philosophers are converging, and that treatment of disease, reducing suffering, is a huge initiative. Consciousness remains an exotic bird: no one quite knows how to handle it, and it dies very easily. His own public goal includes paving the way so younger, more quantitative scientists can make the previous generation look stupid. That, he says, is what they were all trying to do.

From consciousness they drop to David Goggins. Huberman's read is that whatever lives in the abstraction part of the brain, Goggins has figured out how to grab limbic friction by the scruff and send it where he has decided it needs to go — to himself, every day, with no exceptions. The process is not pretty. He has created a rewarding element inside the horribleness. Fridman did a thirty-day push-up and pull-up challenge that Huberman noticed on social media. Huberman's comment at the time was that neuroplasticity loves a non-negotiable contract. The brain is doing analysis of duration, path, and outcome, and that is a lot of work. The more it can pass those off to reflex, the more energy it can allocate to other things. Fix the number of push-ups, the distance, the days, and you actually have more energy for the push-ups.

Once you know the core mechanics, you just go. People get caught in the tide pool of do I really have to, what if I get injured, can I sneak out. That is work. Fridman's hard-won lesson is that if you quit once, the door opens; it is convenient to trick the brain into thinking you would have to die first. The number was not the hard part. The entirety was, especially the early days, hours he was embarrassed to say in public. Huberman's friend Pat Dossett, nine years in the SEAL teams, assigned to more desk time than he wanted, asked himself what he hated most. Bear crawls. He decided to bear-crawl a mile a day. Why pick the thing you hate? Because it creates the most limbic friction.

If you can overcome that, there is carryover — not because a marathon magically improves your marriage, but because you are not exercising a circuit for bear crawls or pull-ups. You are exercising a circuit for top-down control, and that circuit was not designed for bear crawls, coding, or waking in the night to do something hard. It was designed to override limbic friction. Neural circuits generalize: the internal feel of a physical threat is the same class of response as an exam, a divorce, a marriage. Train top-down control when the desire to quit is utmost, from fatigue or from being too stressed, and the circuit is forever with you. Leave it idle and it atrophies, a plant that gets no water.

Which mind tells you to quit? Huberman puts something concrete on it: a paper in *Cell*, two years earlier, on effort. Animals swam toward a target and a reward in a visual environment. Physical effort was the same every time; the virtual world sometimes said you were making progress and sometimes, because nothing drifted by, said you were not. With each bout, epinephrine and norepinephrine are released in the brainstem. Glia, traditionally treated as support cells, measure that concentration. When it exceeds a threshold they send inhibitory signals that shut down top-down control — the quick stop. That is the physiology of I cannot endure this anymore.

Endurance can be rescued with dopamine. That is where subjectivity comes in. You quit because you have learned how to turn the brake off, or because you have learned to self-reward. Some people reward the pain process so much that friction becomes the reward. Goggins talking about eating souls is, in Huberman's hearing, a dramatic form of that — not picking up a trophy, but a reward that gives more neural energy, more dopamine, to suppress the noradrenaline circuits in the brainstem. Viktor Frankl, Nelson Mandela, people Huberman knows who have gone through cancer treatment three times: he thinks the same processes are involved. Again, generalizability, not a circuit for one action.

The pathways are ancient, built to bring resources through foraging, hunting, mates, water, and they sit in circuits where the zeros and ones are known. Reach a three-meter milestone and do not reward it, and it is just effort. Self-reward it and it is effort minus one on the adrenaline output. Fridman wants to keep the race in the mental because the body is limited and stupid physical damage can take you out of running for years. Huberman's key is dynamic control of output: sometimes reduce effort, and also restore through those subjective reward processes.

Why public education, from a Stanford lab? Pat Dossett, on a cold-water swim at the turn of 2018 into 2019, asked what Huberman was going to do to serve the world that year. Running the lab was not an acceptable answer. If I had my way, Huberman said, I would teach everyone about the brain. Then do it, Dossett said, and they shook on it. The governing philosophy: increase interest in brain and nervous system; increase scientific literacy slowly, graphs and p-values, not rammed down; put tools in the world for understanding and directing state — reduce stress, and also raise the threshold so that the goal is not always calm, sometimes learning to tolerate not being calm. The eight- and ten-year-old version of him spent weekends on weird animals and medieval catapults, then asked to teach the class on Monday. He does not like being the point of focus, he says; he compulsively has to tell people about the gems. He refuses the phrase 'dumb it down.' Information should be offered the way you offer food to a guest: cracker, then cheese, then maybe the stinky French kind — doors of interest, not of confusion — until you have people thinking about themselves neuroscientifically.

It is a precarious place to stand as a research academic. Some colleagues wonder if he is still serious about research. He is, and he also thinks the field's work needs talking about, and not every scientist is good at the translation. Fridman asks the ridiculous question about the meaning of it all, via Viktor Frankl. Huberman's answer is that our sense of meaning is very elastic in time and space. Someone locked in a cell or a camp can bring the horizon in close enough to micro-slice the environment and find reward and beauty in a square box. If the building alarm went off now, vision and hearing would lock to a small space-time bubble; the only meaning would be get out, figure out what is going on, contact loved ones. Sit back and do the pale-blue-dot zoom — this room, this country, this planet — and it can all seem meaningless. Shrink to a YouTube comment thread and everything inflates. The brain contracts and dilates its vision of space, time, and meaning. That is what lets us be dynamic.

Frankl and Mandela were not just grinding it out. They had to find dopamine rewards in the little boxes they were forced into. Huberman's undergraduate, graduate, and postdoctoral advisers all died young. What gives meaning, he decided, is the understanding that you can move between these dimensionalities. Two days later the thing that felt so important can look like nothing, because you had a different processing algorithm. What he wants in this lifetime is as many different levels of contraction and dilation as possible. Pull over; there is probably an anthill, a whole fascinating world. You cannot stay there, and you cannot stay staring at the clouds. The key is the journey back and forth up and down that staircase, as many trips as he can get before the reaper. He watched a postdoc adviser wither and still find beauty in closing moments — kids, a Giants game — because time was closing and the bins felt huge. Learning to do that more fluidly, without getting attached to one correct answer, is the goal. Fridman thanks him for contracting space and time onto a few hours of conversation.