What's this about?
DeepMind published a retrospective on fifteen years of games research — from the Deep Q-Network learning 49 Atari games from raw pixels, through Go and StarCraft — and announced it is partnering with studios to prototype new gameplay, including a research partnership around the EVE Online universe with Fenris Creations, plus work with Hello Games and Coffee Stain.
What exactly makes games useful for AI research?
A game is a world that is rich enough to be hard and constrained enough to measure. You get unlimited trials, a clear score, and no safety consequences for failure — which is why Atari and Go produced foundational results. The shift signalled here is from using games as a benchmark to beat, to using them as a place to build things people play.
What's the community saying?
There is no single large thread on the post itself. The connection worth drawing is to the Skyrim companion item above: an independent developer built a real-time AI companion this week while DeepMind announced studio partnerships, from opposite ends of the same idea.