'FTW' - Deep reinforcment learning bots play Quake 3

Perhaps something interesting for the News?

Human-level performance in first-person multiplayer games with population-based deep reinforcement learning

Just sharing this because it’s really cool open research.

Watch the Video, read the papers, perhaps give me your input on what you make of this. :slight_smile:

2 Likes

things like these are great for press releases and all, but they don’t really solve any open research problems in ML, and they’re not a good showcase of what this technology can do

we’ve been able to construct game AIs that beat humans easily for decades, without spending a ridiculous amount of computational resources. Dumb techniques like A* pathing in racing games and platformers, input read buffers in fighting games, and the like have foiled humans since we came up with the concept of the quarter operated arcade machine.

Solving the same problem with high technology and a massive amount of Compute hours doesn’t really prove much.