Cade Metz and Arthur Koestler
Already, generative AI seems historically inevitable. However, as Cade Metz shows in his powerful history Genius Makers, the current wave of large language models and their progeny (ChatGPT and Dall-E among them) was not a sure thing. It emerged through a torturous path of false leads, fierce opposition, and seemingly miraculous coincidences. The coders and creators whose names appear in headlines today – people like Yann LeCun, Geoffrey Hinton, and Joshua Bengio — may be reshaping our world, but it was hardly certain at the outset that their version of AI, with unsupervised code learning from random experience, would prevail.
Genius Makers reminds me of another great book about breakthrough pathfinders with an uncertain future. Arthur Koestler’s The Sleepwalkers, originally published in 1959, focuses on three 15th-century luminaries: the timid Nicholas Copernicus, the hapless “cosmic architect” Johannes Kepler, and the scientific pioneer and heretic Galileo Galilei. They stumbled their way toward articulating the true nature of the solar system and critical aspects of physical reality. They had an idea of where they were going, and they felt compelled to go there, but they didn’t quite know the way. Hence Koestler’s name for them: sleepwalkers.
The generative AI sleepwalker story starts in 1958, when a Cornell University professor named Frank Rosenblatt created the first computer that learned on its own. In this case it learned to recognize markings on 3”x5” notecards. The concept of unsupervised machine learning – and the associated concepts of deep learning, neural networks and language learning models (LLMs) – did not take hold easily. There were major debates, with opponents like Marvin Minsky dominating the field. Nonetheless, the “neural network underground,” as Metz calls it, persevered. Then-obscure figures like Hinton and Yaan LeCun continued to challenge the established conventional wisdom about artificial intelligence: that it should be created through top-down rules-driven programming. Instead, these researchers set loose systems that ramble around data, compiling patterns and setting their own directions.
Technological advance always reflects the personal idiosyncrasies of the discoverers. Sam Altman, writes Metz, was motivated by money. Hinton was propelled by his own drive to persist, despite extreme episodes of self-doubt and some major impediments, like his famous chronic back pain, which has prevented him from sitting down for years. Note that I’m writing about them in the past tense, though their story continues. But generative AI already has a past.
It didn’t have to happen quite the way it did. It didn’t have to happen at all. It happened, and here we are.
In telling this story, Metz has tapped into an archetype. Outsiders develop a new innovation. It represents a transgression against conventional wisdom. The innovators face resistance from insiders. They keep trying, around the margins. They reshape everything. Then they become the insiders themselves.
The quality of the future depends on knowing the history of the past. The Sleepwalkers told us, we need are not bound by the idiosyncracies of the early physicists. They were giants, but they are not everyone. Physics moved beyond them. The Genius Makers tells us, the giants who created generative AI are still at work (and play), but they are no longer the only ones in the field. It needs the rest of us. We were asleep before, unaware of the implications of this powerful new tool placed before us. Now we’re waking up.
