I was thinking it would be fun to play LZ and human vs LZ and human match. Of course, this could be tricky to set up, most likely we’d need a referee for playing LZ moves. Anyone wants to try?
I like the idea, but would it really work? It points to a real benchmark for AI. If the computer is actually intelligent, and not just a calculating machine, it might balk (“I didn’t agree to this”) by playing bad moves, or not working at all. Does LZ take each board position on its face, or would it “wonder,” ‘Where did that bad last move come from, its messing up my strategy…’
I am more afraid about the confidence of the human element I know Leela is miles better than me, why would I question her moves?
It’s all technicalities. Yeah, if we push Leela into 0.1% winrate it’s going to play bad moves. But in the beginning it’s gonna be ~50% so at least up to middlegame it will be interesting.
Most likely it’s better to play it on demoboard. Referee owns the board and makes moves for Leela and for players who put them in a chat. Three people total. Three games total so each player can play twice and be referee once. It’s going to be fun.
Remember how fun it was on Ke Jie vs AlphaGo event? https://youtu.be/V-_Cu6Hwp5U?t=18m25s
It would probably work best if the human players were dans. That way, they can for longer keep Leela Z. in a position that won’t make her run off the rails, ie. 30 – 70% winrate or so.
Sorry to disappoint but LZ reaches 75% winrate pretty quickly even if we start from 70-80% advantage.
It’s rengo with the AI, not against it.
Ah, you mean exclusively. roy7 discussed the possibility of me participating in AI/human rengo and he mentioned any combination could happen. That was on my mind.
First friendly match Human+AI vs Human+AI on ogs was played.
We used demo board and just typed moves + trash talk in chat. AquaBot helped with bot’s moves.
Here’s clean and nice record of the game: