You are reading a single comment by @yoshy and its replies.
Click here to read the full conversation.
-
Yeah it should be interesting. To get the algorithm to scale to the number of potential actions at any given time will be a pretty big challenge I reckon. And then like you say to actually decide on which action to take.
They seem to have skipped forward to a hugely complex game, I'd have thought they would have gone for a racer first (clear goals and limited options for actions) but then basic racing games have already been done by machine learning.
https://youtu.be/S9Y_I9vY8Qw
yoshy
Oh, that should be fun.
The "problem" with Starcraft would be the speed of the decisionmaking.
If they can have the machine act fast enough to overcome a human player in clicks/minute, then I see little chance on the human players side. If the decisions done by the machine take longer than the decisions made by the human, the human wins.