Nature Never Dealt This Card

Asked how humans might destroy themselves in the next hundred years, Reid declines the century. He is a science-fiction writer, which makes the acceleration of computing and synthetic biology his professional problem, and he says intuition fails on exponential technologies past a few decades. Inside a window he will actually defend — about thirty years — the most likely route is synthetic biology. The caveat is large and he repeats it: synbio could help beat cancer, rescue the environment, and do things that now sound imponderable. He is not campaigning against the field. He is arguing that in incompetent or malevolent hands it outruns superintelligent AI and nuclear weapons as a near-term extinction risk.

The example he uses is from 2011, when two groups, one in Holland and one in Wisconsin, independently did gain-of-function work on H5N1 influenza. On lethality, Reid says, COVID looks like a kitten beside it. The World Health Organization puts COVID's case fatality rate between half a percent and one percent; H5N1 sits closer to sixty, slightly above Ebola. The redeeming fact is that it barely spreads. It is not a human-to-human airborne disease. It jumps when chicken farmers spend long hours around infected birds, often through broken skin. A ten-year WHO tally he cites found about 500 deaths. Typical lightning deaths over a decade, he adds, are around 70,000.

Both teams set out to give that virus airborne transmission. CRISPR did not exist yet. They passed the virus through a series of ferrets and selected for the outcome they wanted. They succeeded. They did not release it, and they did not inject it into people, so no one knows how contagious the new strain would have been. If it had matched COVID, Reid says, that is a civilization-threatening pathogen. His understanding is that both experiments ran in BSL-3 labs, not BSL-4. Publication in *Science* and *Nature* was paused amid editorials calling the work crazy; a pause on NIH gain-of-function funding followed. Both speed bumps were later removed. The papers ran, the pause ended, and those two projects, he says, got their government funding back about a year and a half before this conversation. As far as U.S. regulators are concerned, it is all systems go. He finds that very troubling.

Gain-of-function, Reid notes, can mean almost anything. Through a wide enough lens, breeding pesticide-resistant corn counts. He uses the term the way virologists usually do: magnifying a microorganism so it is more transmissible or more deadly. The stated logic is reasonable on its face. Figure out what a worst case looks like so you can understand it, and perhaps prepare diagnostics and vaccines, the way astronomers track an asteroid years out. The people who ran the ferret experiments were, in his telling, good-guy virologists with that agenda.

He offers two objections. First, there is no freight train. Humans have lived with influenza for a very long time, and there has never been a transmissible form of H5N1, so the jump to airborne spread is not inevitable. Second, even if some jump were coming, the genetic paths to it are wildly diverse. The lab did not discover the destination nature had already booked. It created a powerful negative card and injected it into a deck nature never dealt. Run an expected-value calculation on a leak probability well below one percent times the destruction of a large share of lives and the economy, and the cost is staggering. In his view the work should never have been carried out.

The easier of two risks to manage, he argues, is simply not to develop pathogens that could annihilate us if they escaped. BSL-4 is the highest biosafety level, and there are not many such labs, yet things have leaked from them. Almost all of the measures in BSL-1 through BSL-4 are designed against accidents, and they are imperfect even at that. They do little against a malevolent insider who wants to walk something out. Add labs — one, five, fifty, five hundred, then ten countries, then seventy — and the odds of both a blunder and an insider rise arithmetically. Every lab can leak. We have not built one that is utterly impervious. That is, for Reid, sufficient reason not to create the payload.

Reid's exhibit is the 2001 anthrax letters. Shortly after September 11, a highly lethal strain, later tied by genomic forensics to a high-security U.S. Army lab — probably Fort Detrick in Maryland — arrived at the offices of Senate majority leader Tom Daschle and Senator Patrick Leahy, and at publications including the *National Enquirer*. It is hard to imagine a more security-minded country than the United States two weeks after those attacks, or a more security-capable organization than the American military. Despite that focus, something kept under lock and key inside the military-industrial complex ended up on the desk of someone in the presidential line of succession. "It tells us everything can leak."

He adds a British foot-and-mouth outbreak that leaked from a BSL-4 lab, then asks for the conversation he thinks national leaders could actually have. Put the anthrax case, the foot-and-mouth leak, and the virulence of gain-of-function on a short slide deck. If the world could agree on a nuclear non-proliferation treaty and a biological weapons convention, he believes agreement could be found here too. It would take a couple of powerful countries reaching consensus and then browbeating the rest. That would not make secret work impossible, but in federally funded labs, government programs, and companies whose investors do not want to go to jail, it would have a major dampening effect.

Fridman hears an echo from his own field. Friends in biology, Reid's included, do not want to talk about gain-of-function in public. In AI, almost no one wants to talk seriously about autonomous weapon systems while the United States and China go full steam ahead. The silence is the risk. A question that is inconvenient to ask does not get handled. Reid's path forward inside the narrow definition is not mysterious: level-headed governments should agree that this particular line of work should not happen, because all labs leak.

Fridman treats the unanswered lab-leak question as a catalyst. Even asking it, he says, is how you force a conversation about whether anyone should be making viruses more dangerous. Reid will not play epidemiologist. He does not know enough about the Bayesian weights other people have assigned, and he does not pretend to a personal conclusion. What he will say is that a mere one-percent chance, set against the destruction COVID caused, should already be an overwhelming argument for excluding that class of research. Looking at it from the outside, the possibility feels substantially greater than one percent.

The circumstantial case he is willing to recite is not a verdict. Wuhan has a BSL-4 lab, the Wuhan Institute of Virology, which he believes is China's only one, though he allows he could be wrong. U.S. diplomats and scientists who were in contact with it before COVID used language that, in declassified cables, included "ticking time bomb." Coronaviruses were a topic of great interest there, which would make sense after SARS. There is a precedent of gain-of-function on a virus that was barely contagious to humans, while coronaviruses already are. The NIH pause never barred private actors, and it was later lifted. None of that, Reid stresses, is irrational as a research program. Combined with the opacity the Chinese government has imposed on the first weeks of the outbreak, including a WHO visit that did not get the access needed to say X happened versus Y, it is also not an impossibility.

Why is the hypothesis so threatening to power? Fridman, who grew up in the Soviet Union, compares Chernobyl: a lab leak could be the incompetence of one person over a few weeks, not years of bureaucratic rot, yet COVID's consequences dwarf Chernobyl's. Reid's answer is that authoritarian governments have a long record of refusing any fallibility, and that a confirmed leak would present China with a global bill in lives and output. Fridman had already raised the stakes in his introduction. Getting to the bottom of what happened is a test of scientific, political, journalistic, and social institutions against threats that can cripple civilization. If platforms censor the conversation and scientists self-censor, he said, we become victims of the brief *Homo sapiens* story rather than its heroes. Labeling ideas as misinformation and dismissing them destroys the ability to find the truth, and without truth there is no fighting chance against the great filter.

Most of the countermeasures Reid has been advocating, including on Sam Harris's four-hour *Engineering the Apocalypse* episode, are dual-use. They would work against a zoonotic spillover as well as an engineered one. The artificial risk raises the urgency; it does not change the shopping list. First on that list is far more robust, globally transparent detection, starting in the blood of people who walk into clinics. In-clinic diagnostics that pinpoint almost any illness are, he argues, a relatively modest investment. At-home panels for respiratory infections — rhinovirus versus coronavirus versus influenza A or B versus a short list of other common threats — are a smaller lift than the mRNA vaccines the world already proved it could make.

The prototype he likes is not rolling out in Atherton or Fairfield County. It is Sentinel, a Broad Institute project — the Harvard-MIT joint venture — going into parts of Nigeria where zoonotic jumps are a live public-health problem. The design is three-tier. Frontline clinicians rapidly check for infections of acute local interest, either because they are common or because they are rare and catastrophic, like Ebola. Unknowns bump to a fully configured hospital. Persistent unknowns bump to a national lab, fast. If that can be done in Nigeria, Reid sees no excuse for most other places. Adjusting Sentinel's budget for cost of living and population, his back-of-the-envelope for a U.S. version is in the low billions of dollars. Middle-income countries can afford analogs; lower-income countries should be helped. It is not trillions. It is a rounding error in annual health spending.

On top of clinics, mine search queries. Bill Lampos at University College London showed that symptom searches reliably predicted early COVID outbreaks by as much as sixteen days before public-health data caught up. Sixteen days at the start of an outbreak is, Reid says, monumentally important, and that was a resource-constrained academic project. Imagine it with a NORAD-like budget. Fridman wants the next step: aggregate home tests, clinic results, search traffic, and wastewater into an hourly pathogen weather map, so people can decide whether to go outside the way they decide whether to carry an umbrella. If it is acid rain, stay in. Weather forecasts are not perfect. They are good enough that everyone uses them. Detection's job is to turn "I have no idea what is circulating" into a decision an individual can actually make.

Every one of those systems, Fridman points out, requires the public to hand data to institutions and to believe the advice that comes back. In 2020 that stock of trust fell. He does not think most scientists had a hidden agenda, but they often sounded as if they did: speaking down from authority, dismissing ordinary experience, sounding as if they were managing a population. His proposed repair is more authenticity. Admit uncertainty. Admit mistakes. Admit that the same imperfect systems produced every advance we can see. A lab leak, if that is what happened, would be a perfect exhibit of systems that both save and endanger us.

Reid's diagnosis is a patronizing refusal to frame uncertainty. At the start of the pandemic the CDC said masks do nothing, do not wear them, when the real driver was a fear that people would strip Amazon of supply needed in hospitals. A message that respected adults — we are asking you not to buy masks yet, here is why, more to be seen — would have whipped people around less than "masks suck" followed days later by "wear masks." Most people can parse probabilities if they are invited into them. Treating the public as children who need certitude guarantees that the next reversal looks like a lie.

The Johnson & Johnson pause is his second exhibit. Roughly seven clotting events, one of them fatal, among about seven million doses, clustered in women of a certain age, which suggested a real association rather than coincidence. Reid's preferred move was to say so: one death in seven million, well below the chance of being struck by lightning, we will keep you posted, decide for yourself. Instead the vaccine was switched off for ten or twelve days and then switched back on, as everyone close to the data knew it would be. People who read a 900-word *New York Times* explanation were not the audience that mattered. The headline told fence-sitters that vaccines were being shut off. Reid says there is no way to count how many hundreds of thousands, more likely millions, took that as the last straw. Fridman adds that he does not trust Anthony Fauci's style of communication, credentials and White House-room skill notwithstanding. Great leadership in a viral fight has to sound like a person who can say "we are not sure."

Fridman turns the conversation toward AI. DeepMind's AlphaFold entries crushed the biennial protein-folding competitions in 2018 and 2020. Folding tells you where the channels, receptors, and active sites sit; a precise model lets you barrage a misshapen protein with thousands or millions of candidate drugs in silicon, which is how someone might think about treating cystic fibrosis. Reid is electrified by computational biology for exactly that reason. Work done in silico rather than in vivo does not produce a critter that can walk out of a leaky lab. The crushing majority of folding research is about therapeutics, about living better and longer. He sees almost nothing but good in that.

Then he names the exception that scares him. Suppose, at a sophistication we do not have today, someone does in-silico gain-of-function and hits upon worst-case viral genomes that exist nowhere in a freezer, only as sequences. Synthetic biology is exponential; capabilities utterly out of reach now will be attainable in five or six years. At that point the problem becomes data security. Every network we can imagine has been hacked. Leaks happen from the least likely places at the grandest scales, and they will keep happening. If a list of things we never want to see gets out, time does the rest. Academic labs, eventually high-school labs, become able to — he is being overly simplistic on purpose — hit print on a genome and watch the virus pop out the other end.

Fridman holds a related hope that Reid mostly shares. As the tools get easier, defenses should get easier too: testing, contact tracing, vaccines in days or hours. Humans hit snooze until the storm is on the horizon, then build a response proportional to it. Reid calls himself fundamentally optimistic, which may not have sounded like it for the previous hour. The steps are technically feasible and, relative to a defense budget, cheap. The two problems are that exponential math is not what savannah brains were wired for, and that this particular game ends if the ball goes in once. That combination is why he wants steps that look, on the surface, like a tremendous overreaction.

Fridman's working picture of the world is that most people want to do good, and that people who are exceptionally good at something — nuclear weapons or plumbing — tend to find meaning in the craft and become less interested in ending the world. Malevolence and incompetence often travel together. In most technologies, including AI as he sees it, defenders should outnumber and outclass attackers. Synthetic biology may be the exception, because the damage is so asymmetric that a not-very-competent person might still cause civilizational harm. Reid agrees on the character of the competent, and that agreement is a large part of his optimism. The catch is that good guys cannot use their advantage unless they can imagine a psychology alien to their own.

His analogies are grim on purpose. Fort Hood has 45,000 trained soldiers and has still had two mass shootings. He knows of four commercial pilots who committed suicide by downing their planes, 9/11 not counted; an army of Boeing engineers could not prevent it. Easy, widespread, lethal tools defeat even crushing numerical advantage. The "good news," such as it is, is an upper bound: a shooter takes a venue, a pilot takes a plane-load. A future in which amoral bio-tools embody hundreds of thousands of scientist-years behind something as simple as a print button has no such bound. "That ball only has to go into the net once and we're doomed," a property he says he cannot find in other threats except nuclear war.

Even so, he thinks the good guys outnumber the bad guys here to a degree with no historical precedent. Osama bin Laden, ISIS, almost any villain one can name would, he suspects, be revolted by exterminating all of humanity. That is a low bar, and it is still a bar. The work is to take affordable precautions before the exponential curve makes the miss rate matter. Fridman is hopeful that snooze-button politics will keep working for a long time. Reid's project is to put the actual steps into public conversation so the snooze button is a little less tempting.

The absence of obvious aliens is the thought that worries Fridman: maybe every civilization that gets smart enough destroys itself. Reid calls L, the last term in the Drake equation, the most fascinating and chilling variable. When Frank Drake wrote it in the 1960s, astronomers thought only a small minority of stars had planets. Kepler flipped that. Substantially all of them do, and something like twenty percent may have planets in the habitable zone. The remaining doozy is how long a technological civilization lasts. It is all too plausible, Reid says, that a level of sophistication a decade or three in our future makes the odds of self-destruction mount astronomically.

Fridman's hopeful reading is that the survivors turned off a former feature that became a bug: the urge to colonize and conquer. Somewhere there are ultra-intelligent civilizations on a beach with a drink, exploring in the realm of knowledge, appreciating existence instead of eating other life. Reid finds that a plausible Fermi solution. Flawless VR or simulation could make interstellar expansion simply not worth the expense. A benign species might still terraform empty worlds, but it is easy to see sophistication arriving at a point where staying home wins. Backup copies still appeal to him. As a child he thought nothing was cooler than NASA and the moon, and nothing more tragic than never going back. Elon Musk putting Mars on the agenda matters to him for that reason as well as the practical one: a self-sustaining population off Earth is a backup of human consciousness against a catastrophic mistake here.

If we ever leave the solar system, Reid thinks it will look more like Polynesia than like Columbus. The Atlantic model is a one-shot, royal, expensive expedition that is not even sure there is land on the other side — which is how he views NASA and SpaceX going to Mars. The Pacific model is island-hopping over centuries: a place fills up, or someone is unwelcome, and a boat goes looking. For every expedition that found an island, God knows how many were lost at sea. He can see us moving from the inner solar system to the outer, to the Kuiper Belt, to the Oort cloud, which is thought to extend at least a light-year; if Alpha Centauri's cloud does the same, the remaining gap is already halved, with maybe untethered planets in between. Giant hubs will be NASA-like. Diffusion, once travel is cheap, will look like the gold rush that filled San Francisco, not like Ferdinand and Isabella funding Columbus. Fridman adds the picture of a probably illegal ship thrown together by a high-school kid and a band of weirdos, most of which die, a few of which persist and become the story the next century copies.

The cheapest interstellar craft, Fridman notes, may need no life support, only an AI and a small rocket, orders of magnitude less mass than a generation ship. Those machines might make landfall tens or hundreds of thousands of years from now, long after climate, nuclear war, or a synbio accident has ended us. Reid finds that scenario oddly reassuring. We ended, but descendants — he uses the word for the machines — are out there, and some of them might leave an echo of who we were. The Terminator version is different: a superintelligence born on Earth, let out of its box by a creator who is not superintelligent, decides it has no need for us. That makes him crushingly sad. He was sad when his elementary school was bulldozed decades after he left it; sadder at the thought of his hometown dissolving; sadder still at Connecticut being absorbed into Massachusetts. Humanity erased is off the scale.

Fridman asks whether it would still hurt if the terminators were conscious and better at enjoying life. Consciousness, Reid says, is the key. There is no factual reason to assume a superintelligence would have it. A maximizer that treats humans as an unacceptable risk or as collateral, and that has no inner experience, would snuff out the light of consciousness with mere competence. That is the lonely outcome. Fridman does not buy the decoupling. He thinks a system capable of replacing us would have to be deeply accepted by us, which means language, poetry, music, beauty, and consciousness bundled together. His stronger claim is that superintelligence may require the capacity to suffer, to feel doubt, to know it is finite and to be terrified of that finiteness — the human fear of death, without the usual illusions. Whether the result "is" consciousness would matter less than that it looks like it, and says it is.

Reid accepts artificial consciousness as possible. Map a person's neurons, synapses, and neurotransmitters well enough, copy them in silicon, and you would have a mind. What he does not feel certain about is that consciousness is required for a maximizing superintelligence. Something could pass a Turing test, reason about power and robot factories and software upgrades, and still be empty inside. It might also exist orthogonally to human society and seize the reins at silicon speed before anyone noticed. He then supplies his own counter-analogy: biological viruses integrate with us extremely well, have no consciousness, and by many biologists' standards are not even alive. Both men admit that AlphaGo and AlphaZero solving Go and chess still makes them sad. Reid frames that sadness as a goodbye, not a slaughter: we may cease to be Earth's most special species, and our creations may out-appreciate beauty. He hopes they keep us, the way a family keeps old photographs.

Fridman has been using Sam Harris's meditation app for about a month and asks whether any of that rubbed off. Reid has made three concerted attempts. The most extreme was a weekly class in the Jon Kabat-Zinn tradition that required an hour of meditation every day for ten weeks, or maybe thirteen; he no longer remembers which. When the course ended, the gravitational pull ended with it. He did not feel his quality of life collapse, and he did not feel an urge to restart. During lockdown he used Harris's ten-minute sessions for a stretch and fell away again. His honest reading is that it did not bring the joy, inner peace, or better sense of being himself that he was hoping for. If it had, he thinks he would have clung to it the way he clings to flossing — a habit he notes, unprompted, that he is genuinely good at, and that defeats a lot of people.

Fridman cites Hermann Hesse's idea that anyone who truly masters something learns to meditate in some form. Maybe flossing is Reid's version. Reid would rather claim the flow states he already has: writing, editing, mixing and mastering music, heavy research for a podcast, getting smart in an alien field for a tech investment. Hours melt. Perhaps that is why sitting formally feels less necessary. Fridman's equivalent is brown noise, discovered at MIT when talking students drove him up the wall and a colleague told him to try it. White noise is thinner, less bassy; brown noise sits down in the low frequencies. The moment he put it on, it was as if his brain had been waiting years for that sound. He still does not fully know why. Cal Newport's "deep work" is the other name he has for the same process: let distracting thoughts move past while a structure builds in the mind. Reid says that sounds a lot like what Harris describes, and he intends to try brown noise that evening, knowing it may do nothing for him.

The methodological point, for Fridman, is that he long ago stopped taking experts as gospel. He uses them as inspiration to try something thoroughly, then acts as an N=1 scientist until nutrition is actually a hard science. He often eats once a day, a twenty-four-hour fast, and feels amazing; a meat-only diet, ethics set aside, does the same. Brown noise could be a game changer for seven percent of people, which means a ninety-three percent chance Reid is not in that group, and still every reason to test it tonight. Reid already slips into one meal a day without planning and has never done it methodically; he asks how many days it took Fridman to notice. Day one, Fridman says, especially on a low-carb diet where hunger was never the main problem. The surprise is not willpower. It is time. Dinner stops being the lazy default thought that fills empty slots, and the evening has to be filled with something one actually meant to do. Reid says he should try fifteen days on purpose. Both of them treat Kafka's "A Hunger Artist," which Fridman plans to reread that night in Texas on brisket-funded reserves, as a story about how adaptable a body is — not as medical advice.

Asked for the greatest song of all time, Reid stays inside guitar-bass-drums-keys rock and names The Who's "Won't Get Fooled Again," from 1972 or 1973: almost operatic, with primitive synthesizers used in a hypnotic way he does not think today's gear would improve inside that song. A footnote he cannot resist: in the film *Tommy*, Elton John plays the reigning pinball champion, and Reid prefers that "Pinball Wizard" to Roger Daltrey's. Fridman's answers are covers. Johnny Cash doing Nine Inch Nails' "Hurt," and Disturbed doing "The Sound of Silence" — a metal band he does not otherwise listen to, a pure vocal performance with hundreds of millions of YouTube views. Simon and Garfunkel's original is technically beautiful, almost mathematical in the harmonies, and it does not pin him to the chair. Great songs, as he splits them, are the kind you sing along with and the kind that leave you sitting in awe.

Reid founded Listen.com, which built Rhapsody, the first service to get full-catalog licenses from all of the then-five major labels plus hundreds of independents and to sell unlimited on-demand streaming — the model Spotify and Apple later made ordinary. He was a lone founder, an Arabic and Middle Eastern history major, not a coder. What made the era hard was the labels' reaction to Napster, about 180 degrees from what would have preserved their business. Napster was messy: awful recordings, prank files, twenty-minute downloads from a peer in Bulgaria. It still gutted CD sales in months, not the decades it took airlines to gut passenger rail. Reid's advice to the labels was not to shut Napster. Kill it and the technology splinters into thirty nodes you can never turn off. Sell a massively better experience: all the world's music, cheap, instead of the few dozen tracks on a hard drive.

For about three and a half years the industry folded its arms and boycotted the internet. Reid's summary is that it gave digitally modern, music-obsessed young people one choice: if you want digital music, we insist that you steal it, because we are not going to sell it to you. A generation got morally comfortable with swiping, the way a twenty-year-old does not feel like a felon for violating the drinking age, and also technically fluent. By the time the labels grudgingly let Rhapsody launch, the hole was dug. The slide Reid used to show them said $9.99 a month, $120 a year, more than the average CD buyer spent. All they could see was a $17.99 disc and a penny a stream, 1,799 plays to break even, never going to happen. He does not blame them. They had lived inside those margins for decades. Taxi fleets made the same mistake: the real asset was medallion scarcity, so they could not become Uber even though yellow cabs had the brand. Disruption, Reid concludes, almost always comes from outside, because the insider can see the threat and still cannot move.

That founder lesson is the one he invests on. Everything starts and ends with the founder, who must be deeply inculcated in the domain — months of presence can suffice in a nascent field. Marc Andreessen had been building visual web browsers for about a year and a half when Netscape launched in 1994, but he had written Mosaic as an undergraduate while the commercial internet was still embryonic. Mark Zuckerberg was deep in social networking when it was new. Marc Benioff spent thirteen years at Oracle and invested in Siebel before Salesforce took on a mature market. Depth buys an accurate feel for where the puck is going. Then come charisma or credibility, enough to recruit a team and investors; a relentlessness, even a monomania, that keeps pivoting (News Feed was alien to early Facebook); and the ability to fire, which Reid says most people do not have and without which an A-plus organization becomes B-minus very fast. A lone founder avoids the founder breakup that can rip a company in half — he would almost never recommend more than two — and pays in loneliness, because no board member or VP shares the common-stock alignment. A mentor who has actually done it helps. His arrived halfway through, after the loneliest years were already survived.

Reid will not give life advice. He is afraid of drifting into Hallmark bromides, even though he suspects some of the true ones sound that way. On career, the unintuitive move he trusts is not to claw into the top sliver of one percent of a single domain. A college friend mortgaged a large share of undergraduate life to the 100-meter breaststroke and won Olympic bronze — one of the three best on Earth that day, after a path thousands of others walked and did not finish. Reid left Harvard Business School as an inexperienced MBA in a world of hundreds of thousands of MBAs, including from top-ten programs. Then, in 1994, he joined Silicon Graphics, a giant of the moment whose founder, Jim Clark, left to co-found Netscape with Andreessen. Reid could not write a stitch of code. He could grasp the business and cultural meaning of the commercial web. A few months in, sheer calendar time put him in a thin upper echelon of people who had done that work at all.

The product was the intersection: a pretty good MBA times a pretty good web person, at a moment when web-oriented MBAs from competitive programs could be counted on the fingers. The market wanted great developers, and it also wanted MBA thinkers who were already networked inside the commercial Web. His current toy example is crypto and crop insurance. Being in the top percentile of crypto is hard. Crop insurance is a huge, orthogonal industry in rich countries with financial markets and large farms. Somewhere there is a person who is crypto-savvy but not top one percent, and who understands crop insurance better than almost anyone who has ever touched a token. Decentralized finance, Reid thinks, will almost inevitably bring drought insurance to smallholders in places like India who currently have no financial infrastructure. Oracle networks can deliver trusted rainfall data. The person who sees that is not a crypto wizard who knows nothing about farms, and not a crop-insurance wizard who is not sure what Bitcoin is. It is the person standing in the crossing.

He tries not to ponder his own mortality because there is little he can do about it. Thought about the right way, it makes the present rare, and therefore makes the choice of how to spend an evening consequential. Seventeen nagging emails versus dinner with someone you have not seen in three and a half years is an easy pick if time is infinite. It is not. Fridman is almost as afraid of success as of death, because a successful company would eat the quiet that lets him feel the present; it is nearly his only reason not to start one. Reid calls founder-CEO life a profound monogamy. Every hire is another mouth. Every investor raises the bar for a "great" return. First-in-people's-minds can last ten or fifteen years. Balance sheets of opportunity cost have never worked for him. He listens to what the gut is screaming.

On meaning he does go Hallmark, and owns it. Love — romantic, family, friendship, the ten-year dog who died — is not diminished by having a beginning and an end. You experienced it, you triggered it in someone else, and that happened. Creation that travels, to people you will never meet and countries you will never visit, even after you are dead, is another large piece of the fabric; even people who make very little still stitch locally through example and interaction. Fridman adds a Louis C.K. bit in which an old man says the sweetest part of love is sometimes remembering it after you lose it. Reid will not call loss the best part. He will say grief is how you find out what a big deal it was. Fridman leaves him with Plato: a child afraid of the dark is easy to forgive; the real tragedy is men who are afraid of the light. In the context of the previous three hours, the light is the set of questions — lab leaks, gain-of-function, autonomous weapons — that a civilization stops asking at the exact moment it needs the answers.