GOOGLE'S UK artificial intelligence (AI) division DeepMind has shown off that it can play classic frenetic multiplayer game Quake III Arena like a human.
Using reinforcement training, whereby AIs learn how to play the game from pursuing certain gaols based on receiving a reward signal, DeepMind trained a team of AI agents to work together and learn general strategies to play a Capture the Flag game in a customised version Quake III Arena in collaborative fashion while adapting to map changes and enemy tactics.
The idea was to have AI agents play the game like humans, which they managed to do adopting strategies such as camping the entrances to enemy bases and following teammates rather than going off alone - we know plenty of human players who can't even manage that in games.
"From a multi-agent perspective, CTF requires players to both successfully cooperate with their teammates as well as compete with the opposing team, while remaining robust to any playing style they might encounter," DeepMind's boffins explained.
"To make things even more interesting, we consider a variant of CTF in which the map layout changes from match to match. As a consequence, our agents are forced to acquire general strategies rather than memorising the map layout. Additionally, to level the playing field, our learning agents experience the world of CTF in a similar way to humans: they observe a stream of pixel images and issue actions through an emulated game controller."
The result of this effort was a team of AI agents that learnt from scratch how to see, act, cooperate and compete in unseen environments, all from pursuing a single reinforcer signal: whether their team won or got defeated.
Once trained, the agents were put into mixed human and AI teams fighting together and against each other. The agents, dubbed FTW (for the win) had a better win-rate compared to the humans and even worked a lot better together.
Does this eventually mean that smart robots will eventually work together to enslave us all? We can't be sure yet, but they're likely to get better at playing computer games.
DeepMind's smart folks didn't explain how this AI research could be applied to the real-world, but it does showcase what can be done with reinforcement training and AI systems. µ
It's an onomatopoeic week for Google
Hope that free lunch was delicious
It's like Bixby being terrible never happened
Notch to be outdone