Milestone · March 9 – 15, 2016
DeepMind’s AlphaGo won a Go match against Lee Sedol by four games to one.
Go was considered a problem where search alone fails. Combining learning with search showed that position evaluation can be learned from data and self-play.
The source cited is the Nature paper published before the match: it describes the system rather than the match itself.
Nature · Published January 27, 2016
Both concern public matches in which a system played against the strongest human players.
The same laboratory and the same line led to AlphaGo three years later.
The last human component is removed: no human games in training.
Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning AlgorithmThe closing post itself refers back to the 2016 match with Lee Sedol, whose moves, DeepMind says, brought new knowledge to the Go world.
AlphaGo's next moveThe paper sets Pluribus's two processors in play against AlphaGo's 1,920 CPUs and 280 GPUs in its 2016 matches.
Superhuman AI for multiplayer poker (Brown and Sandholm, Science 365:885-890), author copyThe preprint compares itself with AlphaGo: a batch 50 to 150 times larger, a model 20 times larger and training 25 times longer.
Dota 2 with Large Scale Deep Reinforcement Learning (Berner et al., arXiv:1912.06680v1)