Google DeepMind masters lip-reading, is now learning to play StarCraft 2 to improve AI

There are a lot of interesting developments going on in the artificial intelligence world, almost every week you can see several stories about exciting technology that's under development. For example, Tesla gave a sneak peek of its new self-driving car setup, which is powered by the NVIDIA Drive PX 2 system.

Earlier this week Google revealed its DeepMind deep learning system has achieved superhuman capabilities in lip-reading. In cooperation with the University of Oxford, the search company trained the AI system using 5,000 hours of television, with programs like Newsnight, BBC Breakfast and Question Time. The result is DeepMind is now vastly outperforming professional lip-readers in terms of word recognition:

The AI vastly outperformed a professional lip-reader who attempted to decipher 200 randomly selected clips from the data set.

The professional annotated just 12.4 per cent of words without any error. But the AI annotated 46.8 per cent of all words in the March to September data set without any error. And many of its mistakes were small slips, like missing an ‘s’ at the end of a word. With these results, the system also outperforms all other automatic lip-reading systems.

Now ARS Technica reports Google's DeepMind unit teamed up with Blizzard to further deep learning in AI. The research could help to move future games away from predictable, pre-scripted AI to more challenging, human-like AI. Furthermore, it could also help game developers to create better training modules to teach gamers the nuances of a new game, by being able to create in-game advisers that can recognize good or bad moves.

If successful, both DeepMind and Blizzard have their eyes on a raft of potential benefits. On the gaming side, Blizzard sees the project as a way to enhance and improve the Starcraft 2 experience, moving away from predictable, pre-scripted AI towards something altogether more rewarding and useful both for the publisher and its player base.

"The idea that we could get to something that isn't [scripted], that is something which has learned through watching human players—even it doesn’t reach the same skill level as players, it will provide some variety and new challenges that differ from the current level," says Sigaty.