HOUSE_OVERSIGHT_018420.jpg

think than on devising formulas to organize, store and probe data. What had begun as a problem of "artificial intelligence" became, in the end, a puzzle of mathematics. The mystery of human thought, that great unknowable sea of chemicals and instinct and experience that would have let you place your finger on just the song to open the heart of your date, had been unlocked by data. Here was the "Disappearing AI Problem." A puzzle that looked like it needed computer intelligence demanded, in the end, merely math. The AI had disappeared.

For several decades this accidental digital magic trick – "Hey, where’d the AI go?!" – bedeviled machine intelligence. It gave the entire thinking machines enterprise a bit of an occultish flavor. Many problems that once seemed to demand the miracle of thought really only needed data. The human was still doing the thinking; the computer was simply computing. It was extremely easy to draw a line between where the human ended and the machine began. This was a puzzle that had been, in a sense, anticipated at the very dawn of the digital revolution by the mathematician Alan Turing in a 1950 paper called "Computing Machinery and Intelligence" that he published in the British journal Mind. "Can machines think?" Turing began.257 His idea was to test this question in the following way: Have a research subject – a secretary, a graduate student, anyone – chat with an invisible interlocutor by way of a keyboard. Then ask: What are you connected to? Another human? A machine?

Turing figured you could call a machine "artificially intelligent" if it could fool a user into thinking it was human. "Please write me a sonnet on the subject of the Forth Bridge," Turing suggested a tricky user might ask. What computer could possibly know about this famous Scottish landmark; to say nothing of being able to rhyme "Forth"? When the response came back, "Count me out. I could never write poetry," you’d think that sounded awfully human. "Add 34957 to 70764," Turing suggested you might fire back. Say the computer pauses. Then, thirty seconds later, "105621." Are you dealing with a clever machine here? A dim, honest, slow-multiplying human? Impossible to tell. The distinction between machine and man blurred a bit.

Buried in the very premise of Turing’s test was an assumption about what machines could do, and how they might do it. "The idea behind digital computers may be explained by saying that these machines are intended to carry out any operations which could be done by a human computer," Turing said. His smart machines would be tuned by humans to do human tasks, in a human sort of way. Write poetry. Do math. His test of artificial intelligence was designed to figure out if a computer could think like a human. You and I might be able to spot patterns in movie habits given enough time, but as more complex problems emerge, as a world of a trillion connected points becomes a sea of data to examine, there is no chance we’ll match the machines. This opened a then nearly unimaginable possibility: What if, some day, a computer could think better than a human. Could, in a nanosecond, come back with that elusive rhyme for "Forth": North. For such a machine, passing Turing’s Test – thinking like a human – was an interesting challenge. But more interesting

257 "Can machines think": A. M. Turing, "Computing Machinery and Intelligence" Mind 49 (1950), 433

188

HOUSE_OVERSIGHT_018420