Dan Kagan-Kans has interviewed roboticist and futurist Hans Moravec (open copy) The interview is published in The New York Magazine. Moravec, a legendary prophet of weird singularities, rarely gives interviews these days.
In 1988, Moravec made one of the most striking forecasts in modern intellectual history. In “Mind Children,” published by Harvard University Press, he announced that machines would reach human-level intelligence within about forty years. The prediction rested on a straightforward calculation. Studying the human retina’s neurons and their computational throughput, Moravec extrapolated to the whole brain and arrived at roughly 10 trillion operations per second as the target. He then turned to the long record of computing power. By assembling historical estimates - from human clerks with pen and paper, through early mechanical calculators and vacuum-tube machines, to commercial silicon - he showed that the amount of computation a dollar could buy had increased roughly a thousandfold every two decades throughout the twentieth century. This was Moore’s law extended backward into a broader “law of intelligence.” On a logarithmic chart the flat line of biological human capacity was crossed by the exponentially rising line of artificial capacity around 2028.
Moravec did not stop at parity. Once machines could design better machines, recursive improvement would drive them far beyond human levels. The book’s title captured the vision: autonomous artificial progeny would outcompete humans economically, absorb uploaded minds, and eventually inherit the Earth and the stars. In the 1990s he pushed the scenario further in talks, essays, and the 1999 book “Robot,” culminating in the idea of “mind fire” - an expanding sphere of pure computation converting the universe itself. Biological humanity would fade like aging parents. A possible conflict in which superintelligent descendants erased their biological creators was acknowledged but not treated as an immediate catastrophe.
It’s Hans Moravec’s world
Few academic peers took the timeline or the post-biological horizon seriously. Moravec’s emphasis on raw computational power rather than elegant, hand-crafted algorithms ran against the founding intuition of the field, shared by his own doctoral adviser John McCarthy: intelligence could be captured by precise descriptions of reasoning that modest machines could already run. Moravec, immersed in vision-processing research, insisted that sensing, motion, and the vast unconscious calculation of the brain required orders of magnitude more power. “With enough power anything will fly,” he wrote. The view left him intellectually isolated for decades. Only much later, once large neural networks trained at massive scale began to work, did the broader community come around. The essayist Gwern observed that today’s world is becoming “Hans Moravec’s world.”
He remains unsurprised that human-level systems are arriving near the date he named, yet he acknowledges one major miscalculation. He expected future AIs to be modular software systems whose motivations could be directly engineered. Instead they are neural networks grown from data - crystal-like rather than programmed. “Who thought you could pass the Turing test overnight using these silly ideas Geoff Hinton had?,” he wonders.
The prediction that once seemed reckless is landing near its mark. Computing power did not grow with perfect regularity after 1988, yet the endpoint coincides closely with Moravec’s original chart. Intelligence, it turns out, need not be carved by elegant algorithms or even housed in familiar bodies; raw power, data, and connection appear sufficient. Moravec himself, once the most exuberant publicist of the post-biological future, now watches its approach with detached calm. He is content that the mind children are growing, whether or not he lives to see them fully mature.
The wildest thing you can imagine
After “Robot,” Moravec largely withdrew from public debate. A rare autoimmune disease, diagnosed in 2022, eventually confined him to a large reclining chair in a modest house in a Pennsylvania senior community. He is now 77, retired, and describes himself as a spectator. “It’s great to see AI progressing the way it is without me having to do any effort.” Much of his day is spent conversing with ChatGPT about starship designs, nuclear power, and the trajectory of AI, which he judges to be “on track, more or less.”
Moravec’s personal circumstances carry an ironic resonance with the observation that later became known as Moravec’s paradox: machines excel at high-level reasoning but struggle with the sensorimotor skills evolution refined over hundreds of millions of years. His autoimmune disease severs the reliable link between neural signal and muscle. His mind remains clear; his body does not fully obey. He is unsentimental about the disease and about his own mortality. He never expected to be uploaded, finds the prospect of gradual brain excavation “uncomfortable and painful,” and places the odds of oblivion after death at roughly fifty-fifty, with the other half “the wildest thing you can imagine.” His wife Ella, an Eastern Orthodox believer, holds a different view; he is agnostic.
Commentaries
I found very interesting Moravec’s comments on the neural network approach. While he got the game-changing importance of computing power right, he thought artificial intelligence (AI) would be “written like software” and used to be dismissive of neural networks. Today’s AI is centered on neural networks and, while the pure neural network approach must be augmented by all sorts of other tricks to get an AI that sort of works, it can be argued that neural networks do the bulk of the work. Neural networks are significantly more inscrutable than traditional software systems, but Moravec thinks we’ll have to live with that. “We’re going to have to make way for them anyway,” he says.
Another point that I found very interesting is Moravec’s sober but not hopeless attitude toward death. We know there’s much we don’t know about how things really work, and I can live with a 50% chance of a wild afterlife. Moravec doesn’t elaborate on what the wild afterlife could be like, but in light of his published works and interviews I suspect he might have in mind something like high-fidelity reconstructions of the past created by the mind fire of future superintelligence.
A thoughtful and definitely worth reading commentary by Daniel Arjona (in Spanish) frames the interview as the year’s strangest and most moving piece on AI: the story of a prophet who correctly timed the arrival of human-level intelligence and yet will not enter the promised land.
Inspired by Moravec’s vision of biological humans like us fading away like aging parents and leaving the stage to a new generation of mind children, Arjona suggests a literary parallel with the science fiction of Dan Simmons, specifically the “Ilium” and “Olympos” novels. Simmons is one of my favorite science fiction writers, so I find this parallel intriguing. The “moravecs” imagined by Simmons are autonomous, spacefaring, self-evolving biomechanical intelligences. They have been independent of biological humanity for millennia. However, they retain what Arjona calls "the incurable vice of literature.” They are humanists of a post-human sort: curious, loyal to each other, and obsessed with the cultural artifacts of their long-vanished creators.
Arjona presents this as “the kind version of the prophecy.” Where Moravec ends up emphasizing the impersonal expansion of pure computation, Simmons softens the picture. Our mind children would inherit and tenderly curate our culture. “Simmons imagined the children of the mind as we would like them to be: curious, loyal, and incurably in love with their parents' books,” says Arjona.