Hi,
I have programmed a bot that thinks like a human, often blundering.
@hummus-bot
Here is an example from my game
where I played Ke1, hoping to exchange rooks. I totally missed Rxb2! because that pawn seemed protected.
Well, Hummus missed it too for the same reason, and the game continued. This could have happened in a game between two humans.
Some technical details: the bot works by assigning static evaluations to moves (search of depth 0) and then searching the moves with probabilities derived from this evaluation. Basically this means that if a move seems bad at first, the bot will not search it thoroughly - and tactics may be missed.
Another interesting thing is the opening book. Here I sampled millions of lichess games and prepared a distribution of the played moves. Hummus plays the opening using this distribution. For example, if 59% of lichess games start with 1.e4 then 59% of Hummus games will start with e4. The same goes for subsequent moves. This means that sometimes Hummus will try to catch you in silly traps like scholar's mate... because humans try it too.
At the moment the bot is rated ~1900 which is a good sparring partner for me, and I enjoy our games.
Feel free to try it out!
(time increment should be 5-20 seconds)
I have programmed a bot that thinks like a human, often blundering.
@hummus-bot
Here is an example from my game
where I played Ke1, hoping to exchange rooks. I totally missed Rxb2! because that pawn seemed protected.
Well, Hummus missed it too for the same reason, and the game continued. This could have happened in a game between two humans.
Some technical details: the bot works by assigning static evaluations to moves (search of depth 0) and then searching the moves with probabilities derived from this evaluation. Basically this means that if a move seems bad at first, the bot will not search it thoroughly - and tactics may be missed.
Another interesting thing is the opening book. Here I sampled millions of lichess games and prepared a distribution of the played moves. Hummus plays the opening using this distribution. For example, if 59% of lichess games start with 1.e4 then 59% of Hummus games will start with e4. The same goes for subsequent moves. This means that sometimes Hummus will try to catch you in silly traps like scholar's mate... because humans try it too.
At the moment the bot is rated ~1900 which is a good sparring partner for me, and I enjoy our games.
Feel free to try it out!
(time increment should be 5-20 seconds)