An AI That Walks Among Us May Have to Know Depression

After the sponsor reads, Lex Fridman opens the January 2021 AMA with a question from a young man who has battled depression. When we try to build human-like AI, the listener asks, will we hit a wall where the systems themselves suffer from depression and other complex mental issues — and if so, is that a patch, or something you have to leave in? Lex does not take the safe off-ramp that machines cannot feel. Suffering, he says, is a deep, fundamental property of consciousness. Depression is one flavor of the suffering that comes with the human condition, "a kind of dark side street on the path to intelligence."

If we create systems that are truly intelligent in the sense that they can interact with people in deep, meaningful ways, they will inherit many of the characteristics of the full human experience. Depression is part of that package. Humans long to strip cruelty out of the world; it is the old theological question of why God allows suffering, translated into an engineering impulse to build a system void of the parts we cannot stand. Lex thinks those parts are intricately bound up with being human. Systems that move among us, he says, have to take on properties of consciousness, including the mystery and uncertainty of the human experience and all its flavors of pain.

He reaches for yin and yang. Love, beauty, and joy are lifelong partners of pain, confusion, and the oscillation of moods. The interesting part of a conversation, he has noticed on his own show, happens when there is disagreement, a bit of turmoil, a push and pull, a changing of minds, or even just the slow morphing of your own thoughts. That mess has to be engineered into AI that is trying to create meaningful human interaction. He draws a bright line for task-specific systems that will be smarter than people at particular jobs: those need none of this. For the ones that walk among us, "unfortunately, depression has to be part of the experience, or the possibility of depression has to be part of the experience." It is a hypothesis, he admits, offered from an engineer's chair, not a proof.

An Indian international student asks how Lex assimilated after coming from Russia, and whether he feels he belongs. The second half of the question hits him hard — maybe because it is late, maybe because he is over-caffeinated. In Russia he had been a popular kid. In America he became the opposite, a total outcast. The place he landed seemed to value material possessions over the things that were deeply meaningful to him: friendship, human connection, mathematics, scientific discovery. That mismatch, he says, was the catalyst for self-exile. He went into a brooding phase of learning English and reading more philosophical books. The first one he remembers finishing in English is *The Giver*, which helped him start thinking about the world.

The heaviest turn in the episode is here. He was so in love with the people around him as a child, with so many close friends in Russia, that he never noticed how deeply alone everyone is. The immigrant experience was the first time he faced that hard human truth: we are born alone, live alone, and die alone. Even in the arms of someone we love, we are still fundamentally alone, with thoughts, hopes, and fears trapped in "this conscious meat vessel between our ears." The realization terrified him and also liberated him. In the same moment he understood that the beauty he had felt in connection with other people was a gift that took him away from that dark fact. Love, he says, is almost an escape from the muck of life.

That is where the rest of his work begins. Love of the human mind, curiosity about it, and a love of programming grew up together — little programs as a kid, robots in college, the same interest with different outlets. The gift of being ripped out of home was that he finally saw how precious connection is, started looking for it, and became more grateful when he found it. A second curiosity started boiling: if you look at AI from that longing for connection, you stop asking whether a program can finish a task and start asking whether it can carry the richness of experience he had learned to appreciate as a human. When he talks about love in machines, that echo is underneath. He compresses the lesson into one line: unfortunately, the first step of deep gratitude is loss.

Someone asks what he would ask an alien. Lex treats it as a serious thought experiment, not bar chat. First he would ask for advice, assuming a civilization far older and wiser than ours: what should this species do to survive and prosper for a long time? He wants the alien's list of the things we are doing on Earth that are getting us into trouble. Along that line he would ask about great filters. When, in their history, did they almost destroy the entire species? For humans that is the nuclear age and the few moments that could have become all-out nuclear war. He would listen for whether they mention AGI, viruses, wars, or dangers we have not even thought of.

Then he would get selfish and follow his own curiosity. Do you have love where you come from? Do you have death, mortality? He suspects mortality might not even be a concept that makes sense to an immortal species, even if they still run some enforced selection like evolution. He would try to tease consciousness apart: is subjective experience a self-centered, over-dramatized quirk of evolution, or a useful feature that lets a brain think of itself as an individual? Are their intelligence and their consciousness separate, the way they are tightly wound together in us? "Intelligence, consciousness, love and death. All the greatest hits."

Physics comes next. He would look into the alien's eyes, if it has eyes, and see whether they even share a language of mathematics. He would ask what is outside the universe, why there is something rather than nothing, what dark matter and dark energy are, what the basic building blocks of reality are, and whether we are on the right track with quantum mechanics and general relativity. He might sneak in a request for hints on unification, or ask what physics you need at a black-hole singularity or at the start of the Big Bang. That is a natural moment to ask whether there is a God, or at least a centralized designer or team of designers launching and managing the universe, and whether treating the universe as a computation rings any bells. Then the practical round: what computers do you use, what games do you play, how do you build superintelligence, how do you travel near or past light speed, how do you harness a star or a galaxy, and can we build a time machine — and is that a good idea?

He appends a caveat to the dangerous questions: if you know the answer, will I be better off if you tell me? "Sometimes knowledge is not power. Sometimes knowledge is a burden that leads to self-destruction." From his personal stash he would still demand a yes or no on P versus NP and work out the rest himself. He would skip investment advice, on the theory that an alien would find money silly, though he might ask about Bitcoin. And he would definitely ask whether Elon Musk is one of them, and which galaxy he came from. As the alien politely starts walking away from the intergalactic bar, Lex admits the list has gotten ridiculous, and also that it is a useful way to notice which mysteries we actually want unlocked.

A 36-year-old in medical technology and research wants to pivot into computer science. Lex translates that as the broad field of software engineering, machine learning, and robotics, with less emphasis on theoretical CS. His first advice has nothing to do with job titles: find a simple project, get excited about it, have fun, fall in love with it, and be proud of the thing you create. The word that matters is *simple*. Do not start ambitious. Most people, if they allow themselves, can get a huge amount of joy from making a little thing, even by following a tutorial. That joy of creation is sitting there waiting. One of the magical properties of computer science, he says, is that it lets you build things that are almost entities on their own. That is what programs are.

He then offers the path he actually walked. First, basic software engineering: pick a popular, accessible language such as Python or JavaScript, write a hello-world, then something only a little more complicated. Second, use those tools to automate something you already do on the computer — scripts that find files, reorganize folders, rename things. He has one that finds filenames with spaces and, after confirmation, turns them into underscores. Helper scripts like that are joyful because you use them every day, you made them, and they make life a little easier. That, for him, is the fuel that feeds a love of programming. He started in C and C++; now it is mostly Python and JavaScript.

Third, branch into algorithms and data science, both full of beautiful things to fall in love with. On the algorithm side you can enjoy making things more efficient; on the data side you learn to clean, reorganize, visualize, and run statistics, still short of machine learning. That is already a career. A lot of jobs are just using computers to process, visualize, interpret, and analyze data, and the learning curve, if you push, is weeks or months, not years. From there, boilerplate supervised-learning projects, then more specific work, reinforcement learning, even little robots with basic computer vision. Each step can stay tiny. Only then do you see which career path actually excites you.

He names two skills: passion and Google. Passion, treated as a skill, means allowing yourself to get excited and treating that permission as a central part of progress. Google matters because computer science is drowning in good resources; the real ability is discovering the exact page that lets you take the next step. It sounds ridiculous, he says, but knowing what to search and which link to click is one of the essential skills of the modern lifelong student. Passion plus a search engine plus curiosity, one small next step at a time.

Asked whether he is most excited about bipeds, quadrupeds, arms, humanoids, or something more obscure, Lex admits the question is hard because he really likes robots. The love, though, was born in software; hardware just makes it more fun. So the progress that actually excites him is in simulated worlds, and he thinks that will stay true for quite a while. For human-robot interaction, the robots that will matter first are the ones that live in virtual reality or even just on a screen: human-like entities, or entities we cannot help anthropomorphizing a consciousness and a spirit onto, inhabiting the digital world. Slowly those entities will take physical form, probably humanoid.

Humanoid is extremely hard to engineer into a realistic, natural, fulfilling experience, and it is still, to him, the most exciting physical shape. He also likes Boston Dynamics' Spot, the robot dog, from the perspective of having a pet. The physical versions of these challenges are, as far as he can tell, several orders of magnitude harder than the same challenges in digital space. That is why simulated robotics will advance faster and hit the world at larger scale, especially as more virtual worlds get built. He does not even mean VR or AR specifically. He means any medium where you can interact with an AI in digital space, and he counts a humanoid or semi-humanoid digital entity you can connect with on a human level as a form factor of its own.

A listener quotes Einstein: ease and happiness as ends in themselves are the ideal of a pig sty. The ideals that lighted Einstein's way were kindness, beauty, and truth; without kinship with like minds and occupation with the unattainable in art and science, life would have seemed empty; possessions, outward success, and luxury struck him as contemptible. Lex usually agrees with Einstein on philosophy, and he agrees about material possessions. He thinks Einstein is unfair to the word happiness, and also unfair to pigs. He would like to defend both.

Happiness, he says, is a baggy word. Einstein is using it as a synonym for hedonism, a narrow definition. Lex wants to reclaim it as something that includes pleasure but also a deeper fulfillment. His real position is that happiness is a side effect of a life well lived, not a goal. The moment you make it the target, it is easy to lose your way — and that may be part of what Einstein meant. Still, it is an excellent signal. If you throw yourself into the battle of excellence, progress, and growth, a running average of day-to-day happiness over weeks and months is a good measure of how you are doing. The practical way to collect that signal is gratitude: sitting back and noticing where you started and what progress you have made. Momentary pleasure does not tell you much. If the average will not rise, you have probably lost your way.

Kindness, beauty, and truth are fine ideals, he says, but a bit ambiguous day to day. His own compression: struggle is the process and happiness is the measure. Daily life looks like a constant struggle to improve, and the other face of that struggle is gratitude for how amazing life is, for the progress made, and for the chance to struggle at all. You have to imagine Sisyphus happy. Looking back, most of his days are spent truly happy to be alive, so the pursuit of happiness is a good one if it is not a hedonistic local optimum. As for pigs, he thinks they are among the most intelligent animals, and he is still holding out for the possibility that pigs or dolphins have life figured out better than humans. On the pursuit of happiness and the brilliance of pigs, he and Einstein part ways for a brief moment.

How he chooses podcast guests is difficult and still evolving. First, the person has to be truly passionate about what they do, and passion can look like a thousand things — Lex notes that his own face is often blank even when he cares. Coupled with that, they have to be drawn to long-form conversation as a way to express that passion. Not everyone is. Some people would rather write, with several passes of editing; a long conversation has almost no editing. He also wants them to want *this* podcast. There are plenty of hosts who talk, converse, and explain better than he does, which he finds both humbling and motivating. If someone has not listened, or has listened and is not drawn to the particular weirdness of a guy in a suit who speaks slowly and asks strange questions, there is no reason to talk. If they are drawn to that weirdness, there is a chance his oddness and theirs collide into something that surprises both of them.

He is looking more and more for people unlike him: different backgrounds, worldviews, personalities. You can smell a clash of flavors, like chocolate and salt, that might turn into pineapple pizza — which some people love and he still does not understand. The risk has a cost, and he seeks it because that is where a magical conversation might be. He also answers the platforming objection: if a guest holds controversial views, why give them a larger audience? He understands the worry and does not like it. If he is successful, he says, that is precisely the point, because he is not very good at this, especially at hard conversations. But if it works, the tension between worldviews and personalities will create wisdom. He wants to talk to challenging people, including people at the outskirts of society.

He is not afraid of being canceled. He is concerned about doing a terrible job on an important, difficult conversation, and as a result adding division instead of love, knowledge, or inspiration — not because of the guest, but because he failed to catalyze the talk. He does not think he is a strong conversationalist. Some of those talks might take years of intellectual preparation; he reminds listeners that he is just an engineer, and there are topics he will not handle well. So guest selection is partly about whether his preparation matches the difficulty. He is trying to look outside the Overton window, build enough reputation-driven freedom to take risks and make mistakes, and invite the community to let one another change their minds without pinning a scarlet letter on a single sentence forever.

The combination he wants is extreme thorough preparation plus permission to be wrong: excellence and not giving a damn, together. He tries not to walk in with an agenda. He wants to be fragile, open-minded, almost boring and naive, placing trust in mutual respect even when he plays devil's advocate. Part of the filter is whether he trusts the guest not to take advantage of that childlike curiosity. He does not want curiosity to run so far ahead that preparation cannot catch it. What public discourse is missing, he says, is patience with nuanced conversation and with people who later change their minds.

A listener can find a scrap of hope, then lose it the moment the arguments start. How does Lex hold on? He tries to ignore the noise of the world and the bickering of the moment. If you give yourself a chance to see how amazing people are, he says, those people will reveal themselves as amazing. You will see it. Gratitude for how good things already are, and optimism about how much better they could be, he treats as a superpower. It makes life fun to live. For anyone who creates, especially things other people say cannot be built, optimism is a necessary precondition. It is the energy that lets you carry a fire of belief for months and years.

The second move is environmental. In his personal life, his reading, and his conversations, he tries to surround himself with people who are also full of optimism. He is unapologetically a fan of a lot of people. Off the top of his head from the podcast: Chris Lattner always brings a smile, one of the great engineers; Jim Keller, a different personality, equally deep and kind; Elon Musk as an embodiment of optimism; George Hotz, whose chaotic thinking is unlike his own and still inspiring; Joe Rogan as a long-time model of success without jealousy, a person who is a fan of others. On the darker-seeming side, Dan Carlin is often read as a pessimist, but Lex thinks he is optimistic — just soaked in the muck of human history, so the words can sound cynical while the spirit, in their conversation, was clearly hopeful.

The list keeps going: Stephen Wolfram; Sean Carroll and his love of everything about physics; Eric Weinstein and his love of geometrical shapes, symmetry, topology, the weird curvature of visible and invisible dimensions; Joscha Bach, whose flow of consciousness is so brilliant that Lex's "Commodore 64 brain" burns out trying to take in a Pentium. He is especially inspired by software engineers, because the creative process has optimism built in, and by a long trail of the dead — Dostoevsky, Freud, Jung, Camus, Kerouac. He feels as if he lives among people excited about the future. When the noise seeps in, that is what meditation is for. He does not fully ignore it; that would be running away. He takes it in and does not let it linger. From trolling and destructive criticism he picks the pieces that can help him grow, and lets the rest go. That is a muscle. Every once in a while he disconnects entirely and recharges in the simple silence of nature.

A kind message from Brussels asks what he has changed his mind about. He changes his mind constantly; he describes his thinking as a quantum system in flux. A few shifts have been stable over the past year or two, especially through the podcast. On psychedelics he had always been fascinated. What changed is hope: he now thinks they can enter the realm of science, with large-scale studies starting up. That excites him as another window into neuroscience, a way to understand how the mind works and maybe how to engineer some of what makes the human mind special into AI systems.

On social media, he used to think the complicated harms were balanced by community-building and giving ordinary people a voice. More and more he thinks the set of destructive trajectories social media can take civilization down is much wider and more severe than he accounted for. In the conversation about AI existential risk, his attention over the past two years has moved toward social media as the greatest threat of artificial intelligence — and also the greatest set of possibilities. The fan of possible futures is wider than he expected, and it is driven, managed, and directed by platforms. That is work he has been trying to help with.

The biggest shift is that extraterrestrial life, intelligence, and consciousness are worthy of serious scientific investigation. He already felt that way about human consciousness: we lack the tools, and we are very early. The same is true of SETI, of listening, of checking exoplanets for habitability, and of the hundreds of thousands of UFO reports that are only now getting some high-resolution sensor data. What he has come to understand is that when scientists close the door on consciousness and on extraterrestrial life, they prevent discovery. The topics get left to curious minds who are not well equipped — and those curious minds are magical, he says; he is one of them — but the rigors, tools, and funding of science are what can crack the fields open and force paradigm shifts in how we think about intelligence and life.

Conversations with people such as David Fravor, the pilot who saw the Tic Tac UFO, Avi Loeb's reaction when the 2017 material came out, and a more serious look at the Drake equation and the uncertainty in its parameters all opened his eyes to how little we know. There is also a subtler engineering point: staying open to mysteries we barely understand changes the productivity and creativity of an engineering mind. Ideas visit you from outside the little box of what we already know. He is careful on this line because he wants to remain first a scientist and engineer. He also wants a childlike curiosity, in himself and in other scientists. Incremental progress is essential, he says, but it has to be coupled to the kind of first-principles thinking that can throw the citations out the window and make the last few decades look, in retrospect, beside the point.

Why the mostly meat-based diet? The door was combat sports. Lex has wrestled and fought in weight classes his whole life, which means constantly figuring out how to perform physically and mentally while going to school and cutting weight. That made him fascinated with diets, but never as a prescription for other people. He thinks of himself as a nutritional scientist running an n-of-one study: what makes him happy, what makes him perform. About fifteen years ago he found intermittent fasting. He had been in the powerlifting world of six or seven small chicken-and-broccoli meals a day, and then discovered he could eat once a day, train two or three times, and have more energy, more focus, and better performance. It was mind-blowing. Fasting, learned from ultra-endurance athletes and from the idea of the fat-adapted athlete, was the paradigm shift that taught him to keep experimenting on himself.

Very-low-carb keto sounded crazy at first. He had grown up thinking low fat was good and high fat was bad, so eating fat as a real meal felt like cheating. Once he did it properly, with electrolytes and water, it felt immediately great, and he has stayed with it. The physical energy is more stable, useful both for explosive work — heavy lifts, jiu-jitsu, wrestling — and for long endurance. The bigger gift, for him, is mental focus: a cleaner, more stable attention that helps in grappling and also in deep work, programming, writing, or sitting with a blank page designing systems. After he acclimated, the number of hours in a day when his mood is good seemed to go up. Sleep got easier. Joint aches from training seemed lesser. Stable energy made skipping meals easy, which is a gateway into longer fasts.

The biggest liberation is that food stops being the obsession that structures the day. When he ate many meals, he noticed in retrospect how much he thought about food, talked about food, let it organize the hours. On keto and fasting he can spend long stretches lost in things he loves, and food does not enter the mind. The costs are real. He likes the satisfaction of high-volume eating; keto is lower volume, so the body is not hungry but there is less of that stuffed fullness — which is also why he feels lighter and less bloated. Socially, keto used to be unfriendly, because the fun foods at parties and restaurants are full of carbs, and not eating them can feel like sitting out. He thinks that is changing. At McDonald's he orders beef patties for a dollar fifty, and in his experience nobody looks at him weird.

He stops himself from preaching. The stack makes him feel good, focused, and physically capable, and he strongly suggests people try being scientists of their own bodies, but he will not conclude that the diet is great for everybody. He does not know enough, and it does not sound right to say. He has loved meat his whole life and feels best eating a lot of it, and he is also reading and thinking about the ethical side of veganism. That is an ongoing journey; his mind may change. For now, for many years, he has been enjoying keto mixed with carnivore. We will see what the future holds.

What was the darkest time, and what did recovery look like? He loves life, so the question is hard to sit with. Briefly: the darkest stretches have been when he put his faith in people, opened his heart, and they turned out not to be the best versions of themselves, or not the amazing people he had hoped. His heart has been broken in small ways, as he figures many people's have. The fire of hope still burns, perhaps brighter.

Recovery, in his telling, is almost stubbornly simple. He focuses on the positive moments with those people — there always are some — and has gratitude for them, without lingering on the negative. He just remembers the good times. That is how he recovers, how he keeps his optimism, and how he keeps his heart open so that future amazing people are still worth the risk. He is sure his heart will be broken again, perhaps many times. He still thinks the risk is always worth it. He reads a line he wrote down from Marcus Aurelius: love the people with whom fate brings you together, and do so with all of your heart. That, he says, is all we can do. It is 4:20 a.m., a good time to end, perhaps the best. Because they had been arguing with Einstein about pigs, he leaves Winston Churchill on the way out: he is fond of pigs; dogs look up to us, cats look down on us, pigs treat us as equals.