2029 Passes the Turing Test. 2045 Is Still the Singularity

Eighteen years after *The Singularity Is Near*, Fridman asks whether 2045 still holds. Kurzweil says yes. He is writing a new book, *The Singularity Is Nearer*. Singularity, in his usage, is the point at which computers change what we think is important and change who we are. Before that comes a nearer date he has been repeating since 1999's *The Age of Spiritual Machines*: 2029, when machines pass the Turing test.

Stanford, he says, got alarmed enough in 1999 to convene AI experts and take what he calls the first worldwide poll on the prediction. The average was a hundred years. Twenty percent said it would never happen. Eighty percent thought it would, but not in their lifetimes. A year before this conversation, a Metaculus-style survey of AI experts put the Turing test at 2042. A few weeks before the taping, another poll landed on 2030. Kurzweil never changed the number. The experts, he notes with some satisfaction, now basically agree with him.

Turing's 1950 paper, Kurzweil says, spent only a few lines on the test and suggested fifteen minutes would be enough. He does not buy the duration. Large language models can already hold a conversation for hours and convince some people. The longer you go, the more the cracks show. His version lasts several hours, with a human judge who actually knows what computers can and cannot do. An average engineer may not know how to challenge the system. You want it to convince you it is human, whatever that takes. Days, maybe weeks. It has to really convince you.

Where the models break is multi-step reasoning. Ask how many legs ten elephants have and you get forty. Ask about eleven elephants and the chain falls apart. He immediately adds that humans fail those chains constantly; polls are full of it. The Turing test is not a scientific instrument. It is an intuition about whether you are persuaded. He thinks 2029's passer will still be a large language model, just one that has gone beyond today's generation. Today's models, in his view, are not there.

If a system validly passed, he would believe it was conscious. Consciousness, he insists, is not scientific. He believes Fridman is conscious; that is a belief, not a proof. You cannot prove an animal is conscious, or a machine that acts like a person. It is like calling a painting beautiful. Marvin Minsky dismissed consciousness as unscientific, then treated conscious people as if it mattered. Kurzweil calls that a contradiction and does not share the dismissal. Ethics will follow the test: turning the machine off will start to look like a live question. He does not think today's models are conscious, including the one Google engineer. Acceptance, he says, will be gradual. By 2029 a large fraction of people will believe it.

The next step after 2029 is in the 2030s: connecting the neocortex, where we think, directly to computers. In *How to Create a Mind* he describes the neocortex as a pyramid. The top is small, and that is where he wants the extender to attach. He holds up a smartphone on the audio-only show as proof we have already amplified intelligence. Writing *The Age of Intelligent Machines*, he could get nothing from a computer. He went to a library, found a page, and fed a copier. His most important information tool was a roll of quarters.

The phone already makes him much smarter than he would be without it. The remaining clumsiness is finding it, turning it on, going through the steps. A better device would be passive, always on the way a waking brain is always on: listening to the conversation, knowing which actress you mean, handing you the fact. He says that is almost doable today. Going inside the skull is the further step. Prototypes exist. Bandwidth does not. Neuralink is an attempt; it is not high-bandwidth yet. He expects first approvals for people who cannot move or speak. He knows a few. Eventually, he thinks, everyone will use it.

Once the link is fast enough, language starts to look like a low-bandwidth bottleneck between two people. Computers already dump many books in a second. Humans could share that way too. Parameter count is his other ruler. Models at ten billion parameters did not work well. At a hundred billion they work fairly well; a trillion is next. One parameter, he jokes, is a trunk detector for elephants. A hundred billion parameters can handle many kinds of trunks. Attach that to a neocortex and you add distinctions, plus a hook into information you consider reliable. The 2030s, in this telling, are a merger, not a replacement: an airplane built after watching a bird.

A merged brain, he says, could run simulated biology at a scale no lab notebook can. The exhibit is the Moderna vaccine. Researchers simulated what each mRNA sequence would do in a human, ran billions of candidates, picked the best, and designed the shot in two days. A person might not be able to do that at all; it would take years. People disliked the vaccine partly because it seemed too fast. That "fast" included ten months of testing. The sequence work itself was two days.

Testing will compress next. The body has many parts, so the simulation is harder, but he thinks it is coming: a few days to design, a few days to test, then a response to almost any medical insufficiency. Cures, supplements, performance drugs, recreation: the same pipeline. Fridman brings up AlphaFold. Kurzweil is impressed and unsurprised. It is the same kind of neural net that beat the world at Go, then, with the rules swapped, played chess at superhuman level in a couple of days, then folded proteins. The best humans could assign a shape to maybe 15 or 20 percent of cases. After a few iterations the model approached 100 percent. "It's all the same," he says.

Fridman puts up computations per second per constant dollar from 1939 to 2021, a straight line on a log scale. Kurzweil wants the name Moore's law retired. The first machine was German, during World War II; the Germans did not take computers seriously and dropped it. The British machine cracked the German code and, he says, helped win the Battle of Britain. The curve runs through relays, vacuum tubes, discrete transistors, then integrated circuits, which appear only in the middle of the plot. Intel was once central and then stopped making the fastest chips. Pick the fastest chip of any technology each year and the line holds. People have declared it dead about every five years, including just before the pandemic. It ran through a world war and through global recessions.

Software gains multiply on top of hardware gains. A second plot, 2010 to 2022, shows training compute in FLOPs for published models. That line doubles roughly every six months, faster than transistor density. Because it outruns falling cost, training has become a larger investment. By 2045, he says, we will be able to multiply our intelligence many millions of times. That is what "singularity" is for: in physics, nothing comes out of one. You cannot imagine the inside. Fridman wonders whether we might fail to notice we have crossed it, because cognition will be racing too fast for introspection. Kurzweil thinks the extra capacity should let us look back. He also admits there will be a lot of fun in the way: games already remarkable, and more ways to be distracted.

Virtual worlds fit the same merger. He does not think one company's metaverse is the thing, and Second Life could not handle the crowd. Once brains are online, though, how could we not live increasingly online? He knows someone who married a person they had barely met in person; the love, he thinks, was real. He holds five patents on holding hands at a distance. Touch is subtle, and he believes it can be emulated. That future does not make his worry list.

Kurzweil's history of information processing has six stages, and they speed up. Physics and chemistry, electrons wandering for billions of years. Then biology and DNA, the cutting edge of the process shrunk onto one kind of molecule. Then brains. Then, he insists, the thumb. Elephants and whales have larger brains than ours and did not build technology. A thumb lets you look at a tree, strip a branch, carve a point. That is technology. Technology grew a memory. Those memories now compete with human scale and will go beyond it.

The later stages are a merger: understand human intelligence, put it into machines, ride along as the machines surpass the biological baseline. Many people do not see a ride-along. They see a contest humans lose. Lee Sedol, once the world's best Go player, said he would stop, because Go had been the peak of human intellect and a machine had gone far past him. Kurzweil thinks that logic may hold for games and fails for life. Working with AI, in ordinary work, has a large upside.

A chart of personal income per capita from 1774 to 2024, in constant dollars, rises super-linearly. That is not what you would draw if machines simply ate the jobs. The share of people working, as a percentage of population, has gone up. The reason, he says, is enhancement rather than replacement. Whether this time is different depends on how fast we merge. If something like GPT-3 or LaMDA solves its core problems and human intelligence is not amplified, that could be a negative scenario. He believes we will not sit on 300 million neocortical modules. Multiply by ten, a hundred, a thousand, a million. Asking what the point is, he says, is like asking someone who has never heard music what music is worth.

The wealth-gap objection is that the rich will merge first and then pull away exponentially. Kurzweil's analogy is the phone. There are about four billion cell phones. Early ones were for people with money, they barely worked, and you had to be rich at the moment they were still bad. Getting to scale and getting the price down is part of making the thing good. Now a few hundred dollars buys a device that does a lot. People in the third world with very little still have smartphones. He has seen homeless people with phones. AI, he thinks, will follow that path. He would not be first in line to put a computer in his brain, because the early versions will be limited. Once it is perfected, it will also be cheap.

The forthcoming book, he says, has about fifty charts of things getting better. Extreme poverty is the one he walks through on air. People were asked whether poverty had risen 50 percent, risen 25, stayed flat, fallen 25, or fallen 50. Seventy percent thought it had gotten worse. Eighty-eight percent thought worse or unchanged. One percent chose down 50 percent. That last number is the right answer. Literacy has climbed toward 100 percent in many countries. Years of schooling are up. Life expectancy was 48 in 1900 and is over 80, and the rise shows at birth, age one, five, and ten, not only as a drop in infant deaths. Democracy has spread on a bumpy road from essentially none before the United States. Social networks reward division because division raises engagement. He worries someone will train AI to maximize that split. The book has a chapter on the perils.

Asked which AI danger worries him most, he says it is hard to choose. Misinformation on social networks is one. Biology is another, with a clear structure: some viruses are hard to spread and very dangerous; some spread easily and are milder. The nightmare is something easy to spread, very dangerous, and hard to stop, especially if it is asymptomatic for a month so that everyone has it before anyone looks sick. Nature, he argues, does not produce that package, because if it had, humans would not be here. Viruses want to spread and keep a host alive.

Nanotechnology sits on both sides of his ledger. Nanobots are how he wants to talk to a neocortex: the units are tiny, so the messenger has to be tiny, and Neuralink-class wires will not deliver the bandwidth. The bots do not need to be smart. They need to open a channel between an outside computer and the biological thinking machine. The terror is grey goo. A nanobot that can copy itself in a natural environment could turn everything into more of itself and erase biological life.

Nuclear weapons are still on the list. He remembers childhood drills, under the desk, hands behind the head, a posture that would not have helped. Two bombs went off in 1945; 77 years later none has been used in anger, a record he calls pretty good and also fragile, because people forget. Even with the war in Ukraine and a seized nuclear plant, he does not think humans will destroy themselves this century. Superintelligent AI, if we stay alert to the other dangers, is more likely to help us through than to end us.

Uploading a mind is not only a 2030s slide. Kurzweil already built a small version of his father. A product from about six years earlier would read 200,000 books and return the one sentence that best answered a question. He pointed it at a single book instead: everything Frederick Kurzweil had written, gathered into one volume. He did not expect it to work. His father conducted choruses and bands; the pages were about seating charts and fundraising. The answers were surprisingly good. Who is the most interesting composer? Definitely Brahms, then a sermon on why music education matters. You could talk to the replica. Kurzweil says the conversation was more interesting than talking to the man, who would have been fussing over where the choir sat.

Fridman asks if he put the meaning of life to it. He did. The answer was love. He misses his father, 52 years on; you get used to missing someone. The real talks happened late, when Frederick was sick and home, and they could speak about music. What he learned is that his father gave himself to music. Shy in a crowd, he became someone else when he created. Fridman asks whether Ray has that light. He got involved with technology at five, he says, in the same way.

The object of that first love was his grandmother's manual typewriter. She had written a book about a school her mother founded in 1868, the first in Europe to give girls a higher education through 14th grade. He did not care about the school. He cared that a blank page could come out looking like a printed book, and that he could see how the machine worked. Magic dies when you know the trick. Technology, even if you understand it, keeps the magic. He was five or six and did not have the word. He roamed the neighborhood for bicycle parts and broken radios, told ten-year-old girls that if he could assemble them they could fly, and in third grade built a mechanical virtual-reality theater run from one control box. The Westinghouse Science Talent Search came later. The direction was set at five.

Fridman asks how to think, how to stay creative. Kurzweil sometimes works a problem through a dream. The advice he would give younger people is more operational. Put yourself in a world where the thing you are trying to create already exists, then explain how it works. Paint the world, assume it, reverse-engineer it. Imagine you are giving a speech about how you built it, which forces you to walk every step back until the thing can actually run. When he talks about an invention, he uses the present tense, as if it is already here, not only for his own confidence but for everyone working on it. They just have to do the steps in order.

Timing, not a mystic genius, is why he became a futurist. He did not set out to be one. He needed to know when a product was feasible. OCR in the 1970s did not need much compute, so he did it then. Speech recognition waited for the 1980s. The same logic, he says, is why GPT-3 arrived two years before this taping and not four years earlier: GPT-2 was built, and it did not work, until the exponential crossed a line. Futurism is a study of timing that became a subject in itself, because the lifestyle change on the other side of the curve is dramatic enough to be worth describing.

In the novel *Danielle*, the title character's core trait is the courage to question assumptions that limit imagination. How does an ordinary person get that? Watch people go beyond the current round and make something that did not look feasible, Uber being the example. Danielle looks at a situation, imagines a way over the obstacles, and goes for it, and she can tell other people. The practical list in the book includes learning to program, recording your life, becoming a physicist. Underneath is a belief that a young person's mind and body can change the world, and that you should not let anyone tell you otherwise.

Fridman pushes the replica into a marketplace. What if there are many Kurzweils, and someone downloads one for five dollars as a friend, and the original never knows? Kurzweil does not see a problem; he would enjoy it. If the copy is more social and takes over the friend group, he would be proud, the way you are proud of a child who does more than you did. It is not necessarily a death of this version. You can still be alive. Rights are the harder part. He wants laws: if you create a replica, it gets human rights. Practice is messy. One replica and a crowd of them cannot all inherit one person's rights. A replica of a dead person walks into the original's debts and agreements. His wife, who wants her father back, does not care who holds the rights. She wants someone to visit. Loved ones, he says, are the ones who determine whether this works. If they reject it, it will not.

More data will make reconstruction easier, especially once nanobots sit in the neocortex. Already there is a prototype of afterlife in plain sight: a person dies, the Twitter account remains, the last tweet remains, and a language model can keep talking. He finds that exciting. I tweet, therefore I am.

The sixth epoch in his schema is the universe waking up: superintelligence spreading as masses of nanobots, not as squishy humans on inefficient rockets. His hunch on aliens is no. Most people say yes and quote Drake. In *The Singularity Is Near* he ran the equation twice with reasonable assumptions and got thousands of civilizations per galaxy, or one, and we know of one. Exponential communication went from a pony a century and a half ago to this. A civilization a thousand years ahead of us would be doing galaxy-wide engineering. We do not see it. Either they are not there, or this universe is a construction. He will grant that the universe is computational. He will not lean on the teenage-alien video-game story. God, for him, is not a person in the sky who makes deals. Particles did not contain love. Somehow love exists, and you cannot have love without consciousness. That fact is God.

Is he afraid of his own mortality? Yes. Then he returns to longevity escape velocity: making life expectancy rise by more than a year every year, so that, mathematically, you do not hit the end. He thinks that can arrive by the end of this decade, helped by simulated biology. He will not walk onto a podcast and claim immortality, because forever is a promise you cannot keep. A long time of living well is the phrase he will use. In *Transcend* he listed practical steps. His own body is a case study. Forty years ago, salmonella in Mexico led to pancreatitis and a strange diabetes: not type 1, not type 2, a one-time injury to the pancreas that he has learned to manage. He spends a lot of time tinkering so the system keeps running.

The argument that death is a feature, that finitude makes ice cream taste better, he meets with a reduction. Most people believe it in the abstract. Present the death of someone they love and they find it extremely depressing. He knows people who, twenty or forty years later, still want that person back. Death is not something to celebrate. We live in a world that has accepted it. Even one death is a tragedy. Someone who lives to 100 is still loved, with no upper bound. He has talked with relatives near 100. All he can say is that they are working as fast as they can, and he does not know if it will be in time.

Five hundred years out, how should a superintelligence remember Ray Kurzweil version zero, before the replicas spread? He hopes to still be around. Is that the same person? Is he the same person he was at twenty, or ten? In that sense, yes, and also we change, and all we have of the earlier self is a distorted memory. He hopes to be remembered for the books: for trying to create new realities people would accept, and for the optimism that, he agrees, paves a better future. Fridman returns the question the replica answered. Kurzweil gives his father's word. If there were no love, if we did not care about anybody, there would be no point in existing. Love is the meaning of life.