Laplace’s Demon

In 1814, Pierre-Simon Laplace wrote a sentence that an artificial intelligence laboratory could adopt as its founding charter without changing a word: imagine an intelligence vast enough to know at one instant every force in nature and the exact situation of every body in the universe—and powerful enough to analyze all of it. To such a mind, nothing would be uncertain; the future and the past would stand equally present before it. He meant it as a thought experiment about the reach of Newtonian mechanics. It reads now as a mission statement. Strip away two centuries, and the dream of the perfect predictor—the system that ingests enough data to forecast what comes next—is the dream that animates the entire enterprise of machine learning. The language model predicts the next word; the recommender predicts what will hold your attention; the risk system predicts whether you