Blizzard Entertainment announced that they have a partnership with DeepMind, the artificial intelligence that made waves when it became good at Go. Well, now it's going to attempt to get good at StarCraft II, and Blizzard is helping make that process a little easier for it.
According to The Guardian, DeepMind scientist Oriol Vinyals is working with Blizzard to give their AI, DeepMind, an opportunity to tangle with the best in StarCraft II. The AI will self-learn how to play the real-time strategy game, and Blizzard developed an API that features a topographical colored chart to help the AI identify sectors within StarCraft II and understand how to develop strategies based on the rules of the game.
The AI has a long ways to go before it's on a level of competing at a tournament, but right now Vinyals and the crew working on DeepMind have been setting up the parameters and giving the AI the necessary tools to play the game effectively. Since it doesn't use hands or have a physical body, it does everything through simulation, which can create a bit of a conundrum when facing off against humans, given that the AI could technically cheat and access keystrokes and mouse actions in ways that only a computer cold, thus cheating at the game by making millions of calculated moves per minute.
The Guardian has reported that Vinyals has limited DeepMind's calculated executions per minute to fit in line closer to a human's input-per-second ratio. They also wanted to limit the amount of processing power that DeepMind consumed when trying to make too many micro-calculations within StarCraft II.
So how well does it perform? We don't know yet. The announcement came during this year's BlizzCon in Anaheim, California. We'll likely see and hear more about DeepMind competing in the StarCraft scene further down the road.
One interesting thing about it is that it's not just a one-way street for the scientists to help improve their AI's ability to act and react in a real-time simulation where precise calculations and tactical strategies are key. It's actually a two-way street, with Blizzard gaining quite a bit out of the process as well.
The company plans on using the data to help improve their own AI for video games, monitoring DeepMind to see how they can custom-tailor an AI to a player's skill and play-style. That would actually be pretty cool... procedural difficulty and challenge is something that's kind of hard to pull off in a lot of games. Usually, it resorts in a game simply going extremely hard with enemies having massive amounts of health and doing absurd amounts of damage to scale to the high-tier players, or being super easy to accommodate the newbs.
If this little exercise results in Blizzard formulating AI routines that focus more on tactics and appropriate player-reads, it could do wonders for the way gamers engage against AI opponents in real-time strategy games like StarCraft or MOBA titles like Heroes of the Storm. Heck, it might even open up some really cool possibilities for MMOs such as World of Warcraft, where, instead of relying solely on patterned routines from bosses, the game might utilize dynamic boss battles that react to the player's tactics.
Eventually, DeepMind is expected to get smart enough to face off against pro StarCraft players, and that's when things are really going to get exciting.