In light of the recent news that AlphaGo's creators, Demis Hassabis et al., have expanded their self play learning AI to include chess and Shogi in addition to Go, I thought it might be interesting to issue the following challenge to them: Come up with a self learning AI which can play Pente so well that it can surpass the current knowledge base of turn based Pente as it currently stands as player 2. On our end, we get unlimited use of the database here and can call up any living players out of retirement for consultation purposes as needed. We can also consult any existing computer programs. We have 1 week per turn to come up with our moves as player 1, as does the AlphaZero program. Humans will bow down to our new Pente overlord if it can defeat us once in 20 games playing as player 2. Ready? Alpha Go!
Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
Re: Is a superhuman Pente AI possible?
Posted:
Dec 10, 2017, 12:29 AM
This would be interesting. Considering that Pente is a far simpler game compared to Chess and GO (And probably Shogi, though I've never played it), they would probably expect their computer to win easily.
I'm not sure that's the case though, Even if we give it more than 20 moves.
