Incoming links: Chess
Generalization of AlphaGo Zero that plays Go, Chess and shogi.
- www.science.org/doi/10.1126/science.aar6404 A general reinforcement learning algorithm that masters Chess, Shogi, and Go through self-play by Silver et al. (2018), published without source code
- www.quora.com/Is-there-an-Open-Source-version-of-AlphaZero-specifically-the-generic-game-learning-tool-distinct-from-AlphaGo
www.quora.com/Which-chess-engine-would-be-stronger-Alpha-Zero-or-Stockfish-12/answer/Felix-Zaslavskiy explains that it beat Stockfish 8. But then Stockfish was developed further and would start to beat it. We know this because although AlphaZero was closed source, they released the trained artificial neural network, so it was possible to replay AlphaZero at its particular stage of training.
A discrete 2D game on a rectangular grid: towardsdatascience.com/reinforcement-learning-implement-grid-world-from-scratch-c5963765ebff
This is analogous to many traditional board games such as Chess, the concept is very natural and maps well into computer.
The downsides of gridworld games are:
- it is hard to model speed in discrete worlds. When you 10x faster, when do you collide with something else that is also crossing your path?
- they tend to not use vector representations of objects. So to have an object be 10x longer than another one, the naive implementation has to add 10 smaller objects. This becomes untenable as the number of objects increases.
This section is about games initially designed for humans, but which ended up being used in AI development as well, e.g.:
- board games such as Chess and Go
- video games such as Minecraft or old Video game console games
Related to Leela Zero, a Go engine