3 ms·
I wonder when we are going to start seeing game AIs trained by reinforcement learning with self-play. Building a simple 0AD AI would be a fun capstone project.
by nerdponx 9y ago
I wonder when we are going to start seeing game AIs trained by reinforcement learning with self-play.
Building a simple 0AD AI would be a fun capstone project.
- eertami 9y agoThe Open AI guys made a very impressive Dota 2 AI that trained itself through self play (and against pros), there is a pretty interesting write-up about it: https://blog.openai.com/dota-2/ https://blog.openai.com/dota-2/
- Thorncorona 9y agoTo shed some light on why they won, they chose a character (Shadow Fiend) that heavily relies on tight timings to execute moves. Naturally, AIs beat humans in events where timing is the key differentiator. Even so it’s quite impressive