This is similar to how I evolved the AI for my game Hexborne (Think X-Com meets Magic: The Gathering). I had agents designing sets of behavioral heuristics for bots (as "genes"), then enter them into 10k+ match tournaments in an iterative process. After each tournament agents could inspect all the heuristics and try to craft updated heuristics to improve their performance. Winning heuristics got a full statistical validation before being rolled into the baseline set that all agents build on top of.
To multitask I did all this over a multiplayer game server to harden the netcode and ferret out softlocks.
It’s amazing how people with AI are discovering classic techniques for balancing games (i.e. Monte Carlo methods, genetic algorithms…), but somehow implementing them way less efficiently, and without mathematical rigor, basically just having an LLM do the work of deterministic math functions.
You're being presumptuous, I explicitly was explicitly thinking of a GA when I set this up, and how is it inefficient when the AI reduces the number of non-viable policies (and thus the number of wasted simulations) by multiple orders of magnitude compared to randomized policy generation?
CuriouslyC · · focus · HN ↗
To multitask I did all this over a multiplayer game server to harden the netcode and ferret out softlocks.
deadbabe · · focus · HN ↗
CuriouslyC · · focus · HN ↗
deadbabe · · focus · HN ↗
Even if you don’t, the purely randomized policies approach will still be order of magnitudes cheaper.