The Google at Delphi

Steve, still at Amazon in 2005, puts his thesis in the first paragraph so the reader has the option to leave: Google is quietly building the world's first practical large-scale AI. Not Turing-test AI, not human-level intelligence — a system that can solve interesting problems and make money doing it. The recipe: hire Peter Norvig (author of the AI textbook) as Director of Search Quality; hire across cognitive science, ML, compiler design, NLP, and kernel work; have access to data measured in billions or trillions of pieces; have access to compute. Steve walks the long winter of AI in the eighties and the statistical-learning renaissance that needed exactly the data and compute Google suddenly had. The middle of the piece breaks the fourth wall: Steve pauses to say that his brother Dave died six years and six days earlier of T-cell lymphoma at twenty-four, misdiagnosed twice as bronchitis, and that typing Dave's symptoms into Google today returns "cancer symptoms" above the fold. The working definition that gives the essay its title: anything smart enough to save your brother's life.

Twenty years on, almost every prediction in it landed. Steve closes by explaining why he is not going to Google — because Bezos is the smartest person he's ever worked for and if anyone can match Google it's Amazon — and then appends a short bracketed note dated February 2006: well, it's clear the writing was on the wall. Amazon was very cool, but I couldn't resist the pull, and I left six months later.