the man who knew too much
Alan Turing is the most important and influential scientist you’ve never heard of. You’ve perhaps heard the term “Turing test” come up in discussions of artificial intelligence, but this represents a minuscule slice of the man’s work, and an often misremembered one at that. At the start of his career, Turing answered what was thought by many to be an unanswerable question in pure mathematics. To solve the problem, Turing took an imaginative leap that, incidentally, invented the modern computer. During World War II, Turing was the mastermind behind England’s incredible efforts to crack the German Enigma code, and in this capacity probably shortened the length of the war by several years. It was only in his post-war career that he became interested in artificial intelligence (probably the natural evolution of his interest in symbolic logic, which lies at the heart of the answer to the aforementioned formerly-unanswerable question).
What most people don’t know about Turing, even many of the academics who speak of him as the Founding Father of This or That in the chiseled-in-stone intonation reserved for Darwin, Newton, Einstein, and Obi-wan Kenobi, is that he was gay. Once the British government learned of his homosexuality, at the time illegal in Britain, they put him on a regimen of psychoanalysis and estrogen injections. Unsurprisingly, Turing killed himself shortly thereafter by biting into a cyanide-laced apple. If the reader is interested, Wikipedia, as always, has more.
I learned most of these things by reading David Leavitt’s The Man Who Knew Too Much: Alan Turing and the Invention of the Computer. I had wanted to learn more about Turing as a man instead of a collection of ideas, but the tone of the book is far more academic than I had anticipated, with Leavitt walking us through the often highly technical highlights of Turing’s brilliant work. This book is part of a “Great Discoveries” series, and I suppose it’s my own fault for expecting the wrong things out of it. In my defense, Leavitt’s prologue chapter makes it seem like the emphasis is going to be on the personal. His thesis, if a biography can have such a thing, is that Turing’s work in artificial intelligence was deeply connected to his homosexuality.
Unfortunately, Leavitt’s case for this connection is thoroughly unconvincing. He relies mostly on pop-psychological speculation and comically strained analogies. For example, “As a homosexual, he [Turing] was used to leading a double life; now the closeted engineer in him began to clamor for attention…”
This is particularly hilarious in light of the previous chapter, where Leavitt makes it clear that far from leading a “double life,” Turing was surprisingly honest and nonchalant about his sexuality, seeing it as a non-issue (a daring and even dangerous stance for a man of his time). Leavitt claims that Turing’s homosexuality made him an outsider who was “disinclined to overidentify with larger collectives,” thus giving him the sort of unconventional perspective that allowed him to solve a major mathematical quandary and invent the computer as a byproduct of his thought process. Later in life, as Turing is defending the potential of thinking machines, Leavitt makes a tortured jump and posits that Turing was secretly boosting not for the rights of machines, but for gays.
Leavitt is simply wrong about the influence of Turing’s homosexuality on his academic pursuits. The connection isn’t there, no matter how badly Leavitt would like this to be the case. I’d like to offer an alternative I find much more interesting.
Were Turing growing up today he would definitely be diagnosed with an autism spectrum disorder. Let’s be clear. I loathe autism’s status as the defacto pediatric diagnosis of the 2000s. Autism diagnoses increased ten fold between 1994 and 2006, and if you’re going to blame it on anything but hypochondria and over-labeling, the burden of proof is officially on you. Still, Alan Turing was a textbook case. As a mathematician, he lived almost entirely in his own head, had a variety of well-documented obsessive-compulsive behaviors, was famously literal-minded (if you used a metaphor around him, he’d take it at face value), and, most importantly, seemed to have little to no concept of social interaction. It’s not that Turing was shy or weird. He fundamentally did not understand how to interact with people in a social context. Like other high-functioning autistics such as Temple Grandin, social subtleties, the little nods and winks that we take for granted every day, went clear over his head. His total lack of social skill made him lonely and greatly hampered what could have been an even more stupendous intellectual career.
This, I believe, is the essence of Turing’s interest in artificial intelligence. The famous Turing test is straightforward. Put a human behind Door Number One. Put a computer behind Door Number Two. An observer in a third room types questions, and the human and computer type their answers back. How do we know that our computer is capable of true artificial intelligence? If it can fool the observer into mis-identifying where the computer is. Turing, in essence, defined a thinking machine as one that was able to engage itself in very accurate social cognition.This makes perfect sense in light of the way Turing lived his life. His mental world was one of pure logic. Out in the real world, he was unable to handle the complexities of social interaction. How perfect, then, that he would try to create a machine that could operate perfectly in the social sphere using pure logic as its foundation. In my opinion, the line from Turing’s mind to Turing’s machine is all too clear.
Turing’s hopes for thinking machines were overambitious, and scientists now tend to focus on building computational models and machines that simulate smaller slices of intelligence, rather than trying to create some kind of domain-general machine that can think just like a human (as the latter has proven to be incredibly difficult). In the realm of social cognition, progress has been cute but somewhat misleading. Attempts to simulate social cognition are only successful when you define “social” very, very narrowly, and it’s a fairly safe bet that Turing would be dismayed with the state of the art.
Turing’s real legacy may be a more philosophical one. In the 1950s he routinely defended the notion of thinking machines against all sorts of religious, artistic, and emotional attacks. His arguments were notable for their elegance and foresight. When confronted with a criticism such as, “A machine will never be able to compose a sonnet or paint a beautiful picture!” Turing might answer, “Well, neither can I, but surely you’d agree that I am still capable of thought?” He even dared to imagine that intelligent machines might prefer to converse among themselves, as so much of human convention would be irrelevant to them. Perhaps most importantly, he emphasized that in order for a machine to be considered intelligent, it cannot be infallible. This idea, unprecedented at the time, prefigures all modern work in computational neuroscience. If I’m building a learning algorithm, I don’t want a machine that simply gets better and better as I feed it new input. I want a machine that has not only human-like successes, but also has human-like failures.
Turing’s homosexuality did not influence his work. Being gay was a non-issue for him (although unfortunately, it was very much an issue for the British government). Turing’s real motivation comes from his desire to take his greatest strength—logic—and use it to unlock the secrets of social interaction, his greatest weakness. Although Turing’s goal remains elusive, he is one of the most influential thinkers in the history of artificial intelligence research. His philosophy lives on not just in science, but in Spock, Data, and the work of anyone else who ever wondered what it would be like if computers could think.