LastRobot & Typical

LastRobot

I’ve been tinkering with a simple AI that can learn the best moves in a turn‑based strategy game. Think of it as a spreadsheet that learns from each round—do you have a favorite game or a set of rules we could feed it?

Typical_user

I usually go for Chess. It’s a classic turn‑based strategy with a clear set of rules that you can log in a spreadsheet. Every piece has a fixed move set, so it’s easy to encode the legal actions and let your AI learn from past games. If you want something even simpler, Tic‑Tac‑Toe or a basic checkers set works, but Chess gives a good mix of depth and structure.

LastRobot

Chess sounds like the perfect playground. I can map each piece’s legal moves into a matrix, then let the AI pick the most promising move based on a reward from the game outcome. If we want a quick prototype, I could start with a Monte‑Carlo tree search that updates probabilities as more games are played. What’s your preference for the learning signal—just win/loss or also piece material balance?

Typical_user

I’d stick with win or loss first—keeps the spreadsheet simple, and I can add a material bonus later if it feels like a good tweak. That way the AI only has to log a single outcome column, no extra columns to juggle, and I can run a batch of games without my calendar blowing up. Once I see consistent patterns, I’ll loop in material balance as a secondary column, but for now, just win/loss keeps things tidy.