Evochess is a game where the pieces evolve. You start with pawns and a king,
nothing else. To be short, pawns can become bishops or knights. Bishops and
knights can later become rooks.
First time when I wrote here, I have checked statistics.
There were 34 games. Small sample. The only outcome was either "loss" (half of
the games) or "abandoned". No wins. No draws. Even in Easy mode.
I read the game logs from telemetry and those "abandoned" positions were already
lost. This gave me a hint: first-time players are easily beaten by AI and get
frustrated.
I have done my best to re-design the Easy opponent.
My first attempt with biased random moves didn't work quite
well. The Easy player felt "suicidal": sometimes it gave pieces for free.
Which felt broken.
Then I made it less consistent: the majority of time it does weaker moves by
performing shallower search. While sometimes it makes strong moves played
by a neural network. I have played thousands of chess games on lichess
and from my experience casual human game feels exactly like this.
When you play with a clock, you don't pick the best move. You pick
what looks like a good move. Or a fun move.
Here is an example game against Easy that you can scroll back and forth
(Easy was playing Black):
(Apologies for the long URL, it is the whole game encoded since there is no
game backend)
I left it on the most interesting moment as it felt. Such a game would be
impossible against Zen. Not because Zen is stronger, but because it never wavers.
Easy gives you a chance (like a blitz opponent whose clock is ticking) and then
takes it back.
Evochess is a game where the pieces evolve. You start with pawns and a king, nothing else. To be short, pawns can become bishops or knights. Bishops and knights can later become rooks.
First time when I wrote here, I have checked statistics.
There were 34 games. Small sample. The only outcome was either "loss" (half of the games) or "abandoned". No wins. No draws. Even in Easy mode. I read the game logs from telemetry and those "abandoned" positions were already lost. This gave me a hint: first-time players are easily beaten by AI and get frustrated.
I have done my best to re-design the Easy opponent. My first attempt with biased random moves didn't work quite well. The Easy player felt "suicidal": sometimes it gave pieces for free. Which felt broken.
Then I made it less consistent: the majority of time it does weaker moves by performing shallower search. While sometimes it makes strong moves played by a neural network. I have played thousands of chess games on lichess and from my experience casual human game feels exactly like this. When you play with a clock, you don't pick the best move. You pick what looks like a good move. Or a fun move.
Here is an example game against Easy that you can scroll back and forth (Easy was playing Black):
https://penkovsky.github.io/evochess/?p=AQEI_wAAAAD_CFAAAAAG...
(Apologies for the long URL, it is the whole game encoded since there is no game backend)
I left it on the most interesting moment as it felt. Such a game would be impossible against Zen. Not because Zen is stronger, but because it never wavers. Easy gives you a chance (like a blitz opponent whose clock is ticking) and then takes it back.
Tell me if Easy is now too easy.
Previous post: https://news.ycombinator.com/item?id=49146818