The Google at Delphi

Written from inside Amazon on the next-to-last day of 2004, and at least half of it is secretly about Amazon: the Google prophecy shares the page with an asbestos-suit rant against Amazon's C++ codebase and the story of a cognitive-science Ph.D. — a Los Alamos researcher whose phone screen was one of the strongest in Steve's ten years of interviewing — whom the hiring loop forced him to reject because the man's C++ was rusty. The prophecy itself grew out of homework. Steve had spent the year cataloguing the fifty things every engineer ought to know, got asked by the Dean of Princeton's engineering school why machine learning wasn't on the list, went off to repair his own long-rusted AI skills, and came back convinced that the old aphorism "AI is Search" had quietly turned into a business plan someone was already executing.

As a prediction it lands with unnerving accuracy, and not just about Google: statistical learning fed by oceans of training data and massive number-crunching is close to the recipe that actually produced the AI boom, written down while "AI Winter" was still the polite consensus. If anything, Steve undershot. He is careful to talk the ambition down to practical moneymaking systems, no HAL 9000, no Turing Test, and reality overran the modest version he thought he had to defend. The one call the essay gets flatly wrong is its ending, where he explains why he, personally, is staying put at Amazon; a bracketed note he added fourteen months later handles the correction.