A single algorithm can learn to play three hard board games
Science
7 Dec 2018
Vol 362, Issue 6419
p. 1118
Abstract
From the earliest days of the computer era, games have been considered important vehicles for research in artificial intelligence (AI) (1). Game environments simplify many aspects of real-world problems yet retain sufficient complexity to challenge humans and machines alike. Most programs for playing classic board games have been largely human-engineered (2, 3). Sophisticated search methods, complex evaluation functions, and a variety of game-specific tricks have allowed programs to surpass the best human players. More recently, a learning approach achieved superhuman performance in the hardest of the classic games, Go (4), but was specific for this game and took advantage of human-derived game–specific knowledge. Subsequent work (5) removed the need for human knowledge, and additional algorithmic enhancements delivered further performance improvements. On page 1140 of this issue, Silver et al. (6) show that a generalization of this approach is effective across a variety of games. Their Alpha-Zero system learned to play three challenging games (chess, shogi, and Go) at the highest levels of play seen.
Access the full article
View all access options to continue reading this article.
References and Notes
1
C. Shannon, Philos. Mag. 41, 256 (1950).
2
J. Schaeffer et al., AI Mag. 17, 21 (1996).
3
M. Campbell et al., Artif. Intell. 134, 57 (2002).
4
D. Silver et al., Nature 529, 484 (2016).
5
D. Silver et al., Nature 550, 354 (2017).
6
D. Silver et al., Science 362, 1140 (2018).
7
R. Sutton, A. Barto, Reinforcement Learning: An Introduction (MIT Press, 1998).
8
G. Tesauro, Commun. ACM 38, 58 (1995).
9
V. Mnih et al., Nature 518, 529 (2015).
10
R. Coulom, Int. Conf. Comput. Games 2006, 72 (2006).
11
M. Moravčík et al., Science 356, 508 (2017).
12
N. Brown, T. Sandholm, Science 359, 418 (2018).
13
O. Vinyals et al., arXiv:1708.04782 [cs.LG] (16 August 2017).