The Bitter Lesson

Rich Sutton argues that the biggest lesson from 70 years of AI research is that general methods leveraging computation, such as search and learning, are ultimately the most effective, by a large margin. He ties this to Moore's law, or more generally to the continued exponential fall in cost per unit of computation, which makes massively more computation available over time. He cites computer chess, where search-based methods defeated world champion Kasparov in 1997, and computer Go, where similar gains came about 20 years later. He also describes speech recognition, where statistical methods based on hidden Markov models beat human-knowledge-based methods in a DARPA competition in the 1970s. He concludes that building human knowledge into AI systems plateaus in the long run.