3 ms·
Even more importantly, the policy network that chooses which move to explore must choose human like moves in order to function correctly because it must choose
by moultano 11y ago
Even more importantly, the policy network that chooses which move to explore must choose human like moves in order to function correctly because it must choose to explore the correct moves of Alphago's opponent.
- asdfologist 11y agoBy that logic, AlphaGo would falter against an opponent who plays moves at random.
- moultano 11y agoI don't think that follows. To beat the machine the move must be both unpredicted and profitable. Random moves are not profitable. Training purely by reinforcement learning rather than on humans could create a policy network that ignores more subtrees that are profitable than the current one does. In short, it isn't good enough for the AI to be good at playing itself, it has to be good at playing every possible player, and while it is playing humans it is sufficient for it to be good at playing every human player.
- hyperpape 11y agoThat's not right. It just needs to choose equally good or better moves in playouts. It doesn't need to anticipate when its opponent plays bad moves, that's just a bonus. Basically: if you're good enough you don't need psychology, you just play the winning move.