After he was diagnosed with cancer, his head began to explode with ideas of a kind he had never entertained before, and in such copious quantities that I came to fear that, if the disease did not kill him, mental overexertion surely would. This sudden, formidable fecundity was, of course, nothing new in him, it was as if he had recovered, seemingly overnight, the same zeal that he lost after Godel thwarted his attempts at entangling the entire world in a web of logic. Much more curious still was the fact that he also developed (though perhaps it would be more precise to say that he was beset by, as it was a violent and abrupt transformation) feelings that he had no prior experience with: spells of almost overwhelming empathy and a deep concern about the general destiny of humanity. These anxieties, which he could neither contain nor deny, would at first send him into flights of blind panic, though later, when he became more accustomed to the invasion of his psyche by all that he had previously chosen to ignore, he learned to channel these thoughts into himself, where they became the source of a fantastic thirst, an unquenchable curiosity regarding all matters of the spirit.
Perhaps it's because Hungarian mathematician and physicist John von Neumann dabbled in so much that he is rarely remembered today: he was an instrumental part of the Manhattan Project, and later basically invented game theory, before turning his attention to the self-reproducing "automata" that are the basis of today's artificial intelligence programs. He would fit in nicely among the mad geniuses of Benjamin Labatut's previous novel, When We Cease to Understand the World: a brilliant man who outpaces all his rivals, caught up in the whirl of wartime, and then suffers a breakdown of the psyche and spirit. Labatut's von Neumann is a dangerous man, whose intelligence allows him to be cruel to his wives and collaborators, and who expresses little interest in the deadly consequences of his work. In his last days, riddled with cancer, he is struck by a sudden attack of empathy, but at the same time his brilliance has been dulled: his daughter describes leaving his sick room in tears, as he struggles with adding together single-digit numbers. Von Neumann dies relatively young, but his legacy is profound. How much of his cruelty is contained within the ideas and systems he bequeathed the world? How much empathy? How much madness?
The Maniac is bookended by two sections only tangentially related to von Neumann's life. The first is about physicist Paul Ehrenfest, who killed his disabled son in 1933 before turning the gun on himself. Ehrenfest knew what the ascendant Nazis would do to his son if they were able; he so no future either for his son's disability or his own genius. Among other things, this section roots von Neumann's story in the grand political struggle of World War II, which evolves into the Cold War conflict of mutually assured destruction--a concept born not just from von Neumann's work with the Manhattan Project but his invention of game theory--and the destabilizing affects of modern AI. But it also introduces a question about the mental processes that animate men like Ehrenfest and von Neumann--is madness the handmaiden of genius? Or are they a single creature with two faces? The later bookend is a short history of AlphaGo, the AI program that defeated the world Go grandmaster.
In between, Labatut chooses to tell von Neumann's story through the first-person voices of his friends, family, and colleagues. Some of these are wildly entertaining--the whole book, I think, would have been better if it had been narrated by Labatut's version of Richard Feynman--but this strategy ultimately struck me as a mistake. The voices feel false as often as they are convincing, and are inherently less interesting and engaging than Labatut's own erudite voice that begins and closes the novel.
It's the last section, actually, about AlphaGo, that struck me as the best part of the novel. Labatut explains that Go is not like chess; whereas a chess board begins with a large but limited number of possible future configurations, a Go board is a blank space on which the pieces unfold. The number of possible Go boards dwarfs the number of chess boards by many degrees of magnitude; many believed even after Deep Blue's victory over Garry Kasparov that such a program for Go was nearly impossible. But the AlphaGo program takes down the braggadocious Go champion, Lee Sedol, down four games to one, a moment Labatut literally describes as epoch-making. It's funny, actually, to read this section at the moment, where intelligent people (as far as I can tell) are rather down on the power of AI. The language learning models that can't draw hands, or count to ten, are a far cry from the artificial intelligences that Labatut finds frightening, and which men like von Neumann and Turing thought had the power to outstrip their human creators. But perhaps those LLMs are not the same thing as programs like AlphaGo, which have no root or interest in language or art, only the cold purity of mathematics.
But it's not the four AlphaGo wins that make this section fascinating--it's AlphaGo's single loss. Beaten and discouraged, Sedol manages to find a move that baffles AlphaGo so badly it essentially begins making random, desperate moves with no strategy whatsoever. Spectators, having seen the program bait Sedol into mistakes before, assume that the machine is thinking a thousand steps ahead, but the programmers know the truth, because they have seen this before, the program descend into a kind of madness. In this, the program comes to resembles its godfather von Neumann in the most frightening way: it, too, has exploded genius to madness. In this way Labatut suggests that artificial intelligence may resemble our own in ways more frightening than we conceive; they, too, may be prone to psychic unraveling.
No comments:
Post a Comment