His Wife Calls Him the Radio

Lex Fridman opens episode 142 by noting the joke sitting in the episode number. Forty-two is, as Douglas Adams fans know, the answer to the ultimate question of life, the universe, and everything, so he and Manolis Kellis set out to talk about that unanswerable question in whatever way descendants of apes could muster, from biology and psychology to metaphysics and music. Kellis is a professor at MIT and head of its Computational Biology Group; this is his fourth appearance on the show. Fridman warns that the first forty minutes are all songs, a happy accident that revealed another layer of a mind already known for genomes, single cells, and the transcriptional control of disease.

Asked for three songs he draws the most meaning from, Kellis refuses the trap. Countless tracks have marked him in joy and in sadness. His son jokes that he has a song for every sentence, because he will break into melody the moment a phrase reminds him of a lyric. His wife calls him the radio: he can recite hundreds of songs that actually shaped him. The tour he offers instead is a life in music, beginning in the Greece of his childhood.

Growing up in Greece, the soundtrack was misery, poverty, and the stubborn hope of getting past both. One traditional song his parents used to play is, roughly, "Oh, if I was rich." It paints the noises of a poor neighborhood: the train going by, the priest walking to church, children crying next door. With all of that, the singer cannot sleep, and he dreams of being rich. Rich, Kellis says, does not mean wealth. It means out of misery, able to be captain of a ship and see the world. Like the poetry he has discussed with Fridman before, the Greek songs acknowledge how cruel and difficult life is, and they keep the longing for a better one.

Another song splits in half, a fast joyful dance flipping into a slow lament and back again. The poor state of being itself becomes the addressee of the lyrics; then the singer describes growing up in someone's arms, suffering, and standing with clear vision, young men compared to cypress trees. That image is among Kellis's earliest memories of metaphor, the kind of dense picture that trained how he would later think. A third, bleaker song describes how vigorously they took on life and pushed toward a direction they later realized was the wrong one. English songs, he says, rarely smack you in the face quite like that: all that passion and force, and then it turns out you followed the wrong life.

He had been reading a story to his children in the dark. Six-year-old Cleo asked to see the pictures. Eight-year-old Jonathan told her that if you look at the pictures it is just an image, and if you close your eyes and listen it is a video. Kellis calls it brilliant. The line is a child's version of what the songs had been doing all along: the human imagination is larger than the few illustrations a book, or a life, will hand you.

Before he knew English, Kellis learned American pop by ear, Madonna and Michael Jackson and George Michael, repeating the sounds until the meanings arrived later. Michael Jackson's "Man in the Mirror" taught a responsibility he still recites: if you want to make the world a better place, look at yourself and make the change. Songs, he thinks, can be heard shallowly or as a way of touching the psyche. The neuroscience is part of the argument. People who lose language after injury to one brain region can still sing, because the symmetric region on the other side is intact. That duality, musicality and rhythmic pattern on one side and language on the other, is a clue to how the two evolved.

Why did song develop at all? Kellis does not treat it as optional decoration. In his view, language comes after music. Early cultures wrapped every stage of life in ritual, and mate selection is an evolutionary drive: a bird that cannot string together a complex dance does not get a mate. Not every bird can sing, and not every bird can learn a complicated song; some tunes are genetically encoded, others are taught. The mating rituals of birds of paradise are extravagant, and he thinks the mating rituals of ancient human tribes were not far off. Sequential movement, selected rather than accidental, is a prelude to the cognitive machinery that later made language possible. It is a prerequisite.

Even after language arrived, artistic expression is badly needed by the brain, not a break the brain takes. He recalls the scene in *The Shining* where Jack Torrance tries to write and produces only gibberish, and he says he has that image when he works. The huge number of hobbies are not merely tolerated by his research; they are required by it. Stretching the brain in many directions connects the emotional self to the cognitive self, which he treats as a prerequisite for being able to think. His children play piano and violin and sports, run mazes, kick a ball, avoid obstacles. Moving in three dimensions, he argues, uses the same underlying machinery as navigating a theorem. He cannot talk with students without his hands, a whiteboard, or a Zoom canvas.

George Michael's "Careless Whisper" sat on a tape with only a fragment of the song, played over and over. Kellis calls it almost Greek, so heartbreaking, and notes that George Michael is Greek. At twelve the family moved to France and the soundtrack shifted to Serge Gainsbourg, the composer who stood on stage with a cigarette, barely pretending to please, mumbling songs whose lyrics made entire sentences rhyme. Then Georges Moustaki's "Avec ma gueule de métèque." *Métèque* is a French word from Greek: *meta* for change, *oikos* for home, a person who has changed homes, a migrant. The lyric offers a patchwork face, thief's hands that stole gardens, a mouth that drank and kissed and bit without ever pleasing its hunger, and a heart that knew how to make others suffer as much as it suffered, and also how to set people free.

That song still gets him every time. A cover by a singer from Amsterdam adds English lines, and Kellis loves how few songs show such unapologetic desperation for another person. High school took him to New York, where he started learning English in earnest, and Sting's "Englishman in New York" became a manual. It takes a man to suffer ignorance and smile; be yourself no matter what they say. It takes more than combat gear to make a man, more than a license for a gun. Confront your enemies, avoid them when you can; a gentleman will walk but never run. Modesty and propriety can lead to notoriety; you could end up as the only one. Songs, he says, teach you how to live.

Sting's "Fragile" reached him in Greece before he knew the words. Written after U.S.-backed Contras murdered Ben Linder in 1987, it is a political song that became a realization: you cannot win with violence. If blood will flow when flesh and steel are one, tomorrow's rain will wash the stains away, but something in our minds will always stay. Perhaps this final act was meant to clinch a lifetime's argument that nothing comes from violence and nothing ever could. Kellis also notices a craftsman's trick: the more obvious rhyme-word often sits in the second line, so the first can be the surprising one and the couplet still flows. Contrivance, he says, is what you hear when a writer shows off the unusual word too early.

Pink Floyd's *The Wall* marked him as a teenager. "When the Tigers Broke Free," a song that never made the album and lives in the film, describes a miserable morning in 1944 when a forward commander was told to sit tight and a few hundred ordinary lives held the tanks. Kellis traces the word *pioneer* to the pieces you send to the front to be slaughtered; we now treat those disposable men as heroes. The song, for him, explains the whole movie: a person stuck inside his own mind, the wall, the loneliness of "is there anybody out there." No other film, he thinks, has captured that prison so well. Sting's "Russians" he wants as a new American national anthem, not for Russians but for red versus blue. There is no monopoly of common sense on either side of the political fence. We share the same biology regardless of ideology. There is no such thing as a winnable war. To send people into battle you first have to dehumanize them, the way *The Lord of the Rings* turns enemies into monsters. At 99.9 percent genetic identity, he says, you cannot bear weapons against someone like that. The things most meaningful to us are the same on every side.

Joni Mitchell's "Both Sides Now" walks through clouds, love, and life. Clouds that were ice-cream castles later only block the sun; she has looked from up and down and still does not know clouds at all. Old friends act strange and say she has changed; something is lost and something gained in living every day. That, Kellis says, is growing up: the view you had at six is not the view you have as an adult, and if you have love and passion you can see the same rain as a thing to run in. Leonard Cohen's "Dance Me to the End of Love" opened his wedding. "I'm Your Man" is the song he and his wife keep returning to, a catalog of roles: partner, boxer, driver, passenger. Buried in it is the line that a man never got a woman back by begging on his knees, which makes the song, like Moustaki's and like "Careless Whisper," an apology. The last verses still undoes him: if you want a father for your child, or only want to walk a while across the sand, I'm your man. A day, a walk, a life, a child. He will be there.

He ends the music tour on a country song, Alison Krauss's "The Lucky One," the lyric he has identified with most. His friend Serafim, the person who got him into genomics, once pulled over after a conference so a passenger describing a family destroyed by war could get out of the car, then told Kellis he was only reassured because Kellis was there. From the smile, Serafim said, you can tell he is the luckiest man on the planet. Kellis does not claim better luck than anyone else. It is attitude: when something horrible happens he looks for the new chapter it might open. The song's last verse, though, is the warning. Freedom comes at the price of not committing. You cannot be the happy-go-lucky jack of all trades forever and also have a long relationship. Sometimes you have to embrace loss and suffering; if you keep a safety net, you are not all in. Fridman names the pairing: the feeling of being lucky has to sit beside the muck, the messiness, the ways life refuses to turn out as expected. Kellis's version is that the only way not to make mistakes is to never do anything. He would rather be the person whose bicycle is stolen than the thief, rather the one criticized than the critic.

After the songs, Fridman steers toward the episode's excuse for existing. Is the question of life's meaning a ridiculous pursuit, or does raising it have value even if it cannot be answered? Kellis first pays the joke its due. In *The Hitchhiker's Guide to the Galaxy*, a mega-computer thinks for millions of years and returns 42, at which point everyone has forgotten the question. Adams later admitted he pulled the number out of a hat. Turning forty-two himself, Kellis researched why it felt like such a cool number and read the findings to guests at his birthday party. After you have the number, he concedes, you go looking for reasons. That does not make the list less fun.

Light reflects off water at forty-two degrees to make a rainbow, rain with the sun already out, a picture of hope. Forty-two is the sum of the rows and columns of a magic cube of consecutive integers starting at one. It was the last number under a hundred whose sum of three cubes was not known, and it does have a solution. In binary it is 101010, yin and yang, the balance of the force. Giant pandas have forty-two chromosomes. Light takes 10^-42 seconds to cross the diameter of a proton, tying space to time. Fold a paper forty-two times and you reach beyond the Moon. Messier 42 is the Orion Nebula. In ASCII, 42 is the asterisk, a regex for life, the universe, and everything.

The mythologies pile up the same way. In Egyptian tradition the goddess of truth and justice asked forty-two questions of the dead; those who answered became stars. Judaic tradition ascribes a forty-two-letter name of God, trusted only to the middle-aged, pious, and meek. Christian counting gives forty-two generations from Abraham to Jesus. Kabbalah has God creating the universe with forty-two, twenty-five for "let there be" plus seventeen for "good." The Forty-Two Chapter Sutra was the first Indian Buddhist scripture translated into Chinese. The forty-two-line Bible was the first printed book, a mark of the age that led toward Enlightenment. A yeast cell, the single-celled eukaryote Kellis studies, has forty-two million proteins. Fridman is delighted. Kellis's punchline is that it is still, basically, a random number.

Fridman had asked whether the search might ultimately lead to our destruction. Kellis takes the opposite view. Asking about meaning is so inherent to human nature that it makes life beautiful and worth living. Searching is the point, not finding. When you have found it, he says, you are dead. Never be satisfied that you have got it. He likes the line that life is lived forward and only makes sense backward, though he cannot remember whose it is. The search itself is the meaning, and it runs on two scales at once: each person hunting through a single life, and humanity hunting for a collective one.

People look at animals and assume the meaning is obvious, a pile of instincts, a dog running around and loving everything. Humans are the species that spends a tiny minority of its time feeding and still has an enormous cognitive surplus. That surplus is where art comes from, and a healthy mind exploring things not tied to a function, the playing of life for no particular reason. The question is not whether a dog has a purpose. It is whether *Homo sapiens*, having paid so much for the apparatus between the ears, should spend it on something other than dominance and calories.

Fridman asks whether this mind is unique in the universe, perhaps the most beautiful thing the cosmos has built. Both the most beautiful and the most ugly, Kellis says, but certainly the most complex. Dinosaurs ruled Earth for 135 million years. Humans have been around for a million, a ratio of one to 135. The dinosaurs vanished about sixty million years ago; mammals had already been evolving as tiny creatures for thirty million years, then took over and radiated. Out of those mammals came the neocortex, the outer layer that sits on the so-called reptilian brain we still share in structure with dinosaurs. They did not have that cortex, and they ruled anyway. How many other planets, he wonders, still have mindless dinosaurs, with strength as the only dimension that matters.

Human cognition took a strange path. We spend some 15 to 20 percent of bodily energy on the brain, an enormous caloric commitment no other species matches, paid for with metabolic efficiencies that still leave the organ a glutton. Skulls could only grow so large before they would no longer pass the pelvic opening and would kill the mother at every birth, so folding continued instead, packing more capacity into the same volume. Other human-like forms existed and then vanished, leaving us as the only species with this hardware. That can make the story feel unrepeatable: maybe the asteroid missed, dinosaurs still rule, and any proto-human is breakfast. Once a neocortex is in play, though, the contest shifts to intelligence, social structure, larger groups, coordination, and empathy.

Here he offers a sequence that Fridman flags as striking. Empathy, the ability to project another person's intentions, may have come before self-awareness. Building a mental model of how *you* think may have preceded a model of how *I* think; introspection after projection. Empathy means feeling what the other is feeling, mapping an emotional state onto your own apparatus, and Kellis thinks that is what finally exploded human cognition. Life itself he treats as inevitable on every planet. The evolution of life into self-awareness, and into the things humans have done, may not be. Dinosaur longevity is his evidence that the cognitive revolution was not guaranteed. That feeds the Fermi paradox: if the universe is full of life, why the silence? One answer is a great filter, a short side road with a dead end. Kellis's reply is that the universe is enormous in time as well as space. We have been broadcasting for a blink. He loves that *Star Wars* begins a long time ago in a galaxy far away, not in some distant future. Alien civilizations would have to be watching every insignificant planet just as we started to listen.

He still thinks we will become a galactic species. Cognition won the dominance battle on this planet; other histories might have gone to strength. The flip side of being the lucky ones is that the same hardware can invent ways to destroy itself. Steven Pinker's *The Better Angels of Our Nature* is his counterweight: civilization, the state monopolizing violence and then handing governance to people, has dramatically reduced death, suffering, poverty, and war. The arc has setbacks, but it bends toward less destruction. Fridman imagines two dinosaurs having this same conversation while an asteroid slides into view. Kellis answers with another number. In 2068, forty-eight years from the taping, the asteroid Apophis has a very high chance, as he puts it, of wiping us out. We have forty-eight years to get our act together.

The deadline does not scare him. What let humanity move so fast is not only hardware upgrades, the neocortex and its folds, but software upgrades: language, culture, environment, education. The hardware has barely changed in seventy thousand years. Take a person from ancient Egypt and bring them here, he says, and they are equally fit. What changed is that we grow up in far more complex societies. Neoteny keeps the brain from fully stabilizing until after the teenage years, which gives sixteen or eighteen years to infuse it with the latest software. Ancient Greece exploded, in his telling, because of a shift from Egyptian hieroglyphic software to Greek language software, a toolkit for abstract notions, layers of meaning, ideals such as harmony, peace, justice, even doom. Without the word, the brain has trouble getting to the concept.

After the Enlightenment and the printed press, knowledge began to spread sideways, not only down genetic lines. Hardware improvements happen through vertical inheritance; software improvements happen through horizontal inheritance. Civilization exploded because of the second kind. Coronavirus is his exhibit. A century ago it could have killed us. In January the genome was published; less than a month later the first vaccine designs were done; ten months later a vaccine at 90 percent efficacy was in hand, ridiculous by any historical standard, and the reason is sharing. Forty-eight years until an asteroid, set against eight years of recent technological change, looks to him like infinity. Digitization, compute in a sliver of hardware, private rockets, the democratization of space: science has not slowed down. Social and political life can thrash; the lab, he says, is moving at breakneck pace. He is not worried at all.

Would he get on the boat to another planet? Probably not on the short list; he will be older by then. He would love to see Earth from above, that thin blue shield. Drive to New York, he jokes, and you are basically in outer space, the atmosphere is so slight. It is a privilege to be on this planet, and he thinks the species is in for good things, that we will overcome our little problems and come together, even if that feeling has not yet permeated the whole Earth.

How, exactly, are we special relative to dinosaurs? Cognitive improvements are only the start. Picture a lion hunting a gazelle on the Serengeti. The lion smells molecules on the wind, hormones and receptors priming it for impulse; it attacks against the wind so it is not smelled. The gazelle scans. Mitochondria in the lion's muscles wait for a jump's worth of energy that began as sunlight in grass, passed through chloroplasts, into the gazelle, into the cat. Humans, watching, can hold the whole scene. The lion experiences only one layer. The mitochondria experience one layer. The chloroplasts experience one layer. Photoreceptors, smell receptors, chemical receptors: one layer at a time.

He maps the same idea onto a computer. Physical layer, electrical layer, gates and logic, assembly, high-level code, a deep-learning routine, and eventually a cognitive system. Now imagine that system becoming not only self-aware but aware of the hardware and the atoms it is made of. The scene in *2001: A Space Odyssey* when HAL, being disconnected, sings and tells Dave he can feel himself losing his mind is, for Kellis, that breakthrough: knowing the hardware is going. Humans who have had accidents know they have had accidents. We share a reasoning layer with the machines we built; something else on the planet can integrate equations and play chess. That something, so far, still lives only in its layer. A bat does astonishing signal processing, echolocation at slightly different frequencies in a crowded cave, motion planning it is not even aware of. It just has a three-dimensional view, the way a baby sees without doing the math. You live in the layer you live in.

Humanity, he argues, has climbed through perception, organs, genes, molecules, atoms, quantum mechanics, even the fabric of spacetime, and united them cognitively. Anyone who finished high school can watch the Serengeti and know the layers are in play. Particle physics is the same move in another direction: the very small and the very large explain each other; every gold atom is the debris of a stellar fusion chain. Time opens the same way. Looking at Kilimanjaro you can see the volcano; Death Valley lays history on its side; we can see back to the Big Bang and forward to Apophis. In ancient times that would have been magician stuff. If there is a God watching, Kellis says, it is about time somebody figured the place out.

Fridman had asked last time whether we live in a simulation. Kellis's twist is that we already do. Inside the skull, sensory inputs become perceptual signals become a conceptual model, a matrix we did not need Morpheus to reveal. Hardware explosion plus scientific software explosion let us break through, realize we live in a matrix, and still have a consciousness that lives through the layers. We are, so far as he can think of, the only thing that has permeated space, time, and abstraction like that. The next frontier is the inside: the basis for the soul, the ego, the spark that makes you you. Completing the journey through that matrix, and maybe duplicating it in AI, is the unfinished work.

What he loves about current AI is that it is unpredictable. There is emergent behavior in cognitively capable artificial systems that we can model but do not encode directly. Saying the systems are not really intelligent because we wrote the code and tuned a few billion parameters is, he thinks, like saying we understand humans because they are made of neurons. Every person is encoded by a ridiculously small number of genes relative to the complexity of the mind: about 20,000, only a tiny fraction of them for cognition. The rest is emergence, cortical layers doing their thing. Deep learning now puts an architecture in place and lets the system go, a long way from early AI that failed if you had not hand-coded what to do with a grocery bag of cold and hot and hard and soft. The origins are humble. The result, deployed, is infinitely complex. It is still not cognizant of space, time, or the electricity it runs on.

Humans have the same cognitive architecture as monkeys and great apes, just a lot more of it. GPT-2 to GPT-3 is the same architecture, more of it, and already able to do much more. Do you need a fundamentally different design for whatever comes next, or just a ton more hardware? These systems are nowhere near what humans have between their ears. People shy away from the idea that a later GPT might be general intelligence, as if something totally new had to be added. They do the same with consciousness, reaching for quantum or non-physical extras. Or it could be more of the same, enough neurons for self-awareness to appear, an emergent behavior of running, seeing, reacting, predicting a fly in the air, a giant neural lookup table in a complicated key. The hardware, for a geneticist, sets an upper bound; the environment makes the bound shine. Nature is genes. Nurture is the richness that lets the apparatus reach its potential. We are far from that potential. Children a hundred years from now, he suspects, will find our educational systems primitive.

Fridman brings up Richard Dawkins and memes, ideas as organisms breeding in minds. Kellis likes the horizontal transfer, and then twists it. Humanity is already one giant idea-sharing network beyond what Dawkins imagined. But we do not only ship ideas. Fellowships and university admissions select for the cognitive hardware that produced the ideas, then move those brains across the planet. Physics minds go to CERN, genomics minds to the Broad Institute, computer science to Stanford or Carnegie Mellon or MIT. Students come to Boston because that is where certain cognitive systems thrive; once there, the interbreeding of memes is far denser than if ideas evolved in a vacuum. Some stay, become professors, and raise genetically related cognitive systems of their own, children growing up among parents who love this kind of thinking. Genes and memes, in his image, evolve in the same soup. Online groups do the same without geography: a person can belong to a computer science department, hang around biology, read poetry, and join a twelve-person community that exists only as a shared obsession. Teams stick because they meet on a common passion and stay because they are complementary, which is also how he describes his marriage: extremely different, common goals, dates that are papers and ideas.

How fundamental is language to sharing all this? Some people say language dictates thought and that thoughts cannot form outside it. Kellis disagrees. He does not dream in words but in shapes, forms, three-dimensional space with extreme detail; he records dreams in a Dropbox file or by dictation, and a recording can bring a whole night back even when he does not remember dictating it. No other person could recreate that geometry from the audio. Ideas he cannot put into words he will describe with a thousand, and the idea is still more precise, or more abstract, or simply other. Language is a projection from three-dimensional thought into a one-dimensional stream. It also gives you apparatus for concepts you did not know you had. His team coins terms such as the regulatory plexus of a gene so they can build on them. Ideas and language coevolve; they are not one-to-one. And human communication was never just words. Eyes, face, posture, intonation, pauses, volume: great apes already grunt, posture, shriek. We kept the ape channel and added the reasoning one. We are not merely a conversational system. We are a grunting, emotionally charged system that can also reason, standing inside a shared knowledge base of presidents and 9/11 and *2001*, so that naming one scene can raise a feeling in both chests.

Neuralink's early dream, transferring whatever is encoded in one brain into another, he finds beautiful and full of holes. We barely understand how knowledge is encoded. A paper with colleagues looked at mouse neurons that co-fire on a stimulus, then fire again on recall, the beginnings of a distributed engram. Reading a scene such as HAL losing his mind out of a human brain is already hard; writing the same connections into a second brain is another problem, and one he would rather not watch go wrong. Even then, encodings may not transfer. His version of that scene is twisted with childhood pebbles in Greece, Fridman's with Russia; dumping one into the other would mess things up and clash with unique experience. Ask a smaller model to give a larger one all its knowledge and the honest move is not to hand over ten billion parameters. It is to keep talking, generating every sentence it can, while the larger net re-encodes them. Language, on that view, is already the protocol: run a generative model forward, parse it on the other side. Interruptions and misunderstandings are not bugs. Kellis will stop a student, say what he thinks he heard, watch someone else rephrase it, and by the third pass they are somewhere none of them started, a diversity you would lose if you could copy weights directly.

What does he hope the world will remember him as? He does not think in legacy the way most people mean it. Students ask for career advice and he says he is the wrong person; he has never made a career plan. He is consciously living in the present, the lucky-one habit of being happy winning and happy losing. Make every day a local maximum of awesomeness for the town, the academic family, and the biological one. Be generous to friends, kind to enemies. You cannot stay upset at people if you truly love them; he looks at adults who insult him the way he looks at his screaming three-year-old, a brain stuck in a mode it may snap out of, or may not, and that is all right. A friend he counts among the smartest people he has met told him she loves being wrong because it makes her feel human. Through the podcast he has been discovering a legacy he was not tracking: generations of MIT students, lectures he has put online since 2001, strangers in a conference hall who know his voice. A question he asked at a meeting three years earlier had, someone told him, inspired a whole company project. Genes are already shared with cousins whether or not you have children. Ideas will touch everyone alive a hundred years from now, not just his, Joe Smith's too. America, in his telling, exports trained cognitive systems; political isolation cannot stop that. His academic family already has great-grandchildren, former trainees who are professors with their own labs, papers in which he can still see the lab's touch. The first paper with a student is always a joint editing session, screen shared, thinking out loud, creative destruction instead of a page of red ink two days later.

The research is how that teaching happens, mens et manus, mind and hand. A recent single-cell schizophrenia paper, which a stranger called a master class in scientific writing, found a transcriptional resilience: some people with schizophrenia seem to carry a protective cell state that keeps other cells' expression closer to normal, alongside master regulators that are themselves genetic targets of the disease and sit in synaptic connectivity and early brain development. The work has on the order of 100,000 to 120,000 citations, not readers only but people who built on it. He still feels like a kid in a candy shop, jack of all trades, master of none. They close with two of his poems. "Write Me a Poem" was composed on demand after a breakup, mocking the request for a pretty, romantic, rhyming product and then, by accident of typing without going back, rhyming anyway; love is no recipe. "Burn Me Tonight" refuses the offer to be friends: shout I love you or send me to hell, go all in rather than waste the rest of a youth. Fridman says there is no better way to end a meaning-of-life conversation. Whatever the meaning is, go all in.