#332 — Can We Contain Artificial Intelligence?
Making Sense with Sam Harris
The Evolution of AlphaGo
The breakthroughs that got everyone's attention in the AI field were with AlphaGo, AlphaZero, and AlphaFold. AlphaGo's success in playing the game of Go demonstrated that traditional methods used for games before could not handle the complexity of Go. AlphaGo used a new suite of methods that scaled with more compute, and it progressively improved its performance. AlphaGo's success led to the development of AlphaZero, which achieved superhuman performance without any prior game knowledge. AlphaZero learned entirely through self-play, spawning instances of itself to play against in simulated environments. This uncoupling from human knowledge allowed AlphaZero to excel in various games, including chess and shogi. The ability to parallelize and scale up these methods, utilizing computing infrastructure and improving models' power and performance, is a crucial advantage.
Remember Everything You Learn from Podcasts
