Next-Gen Artificial Intelligence Learns to Play StarCraft II

AlphaGo, the AI from Google, was able to win four out of five games of Go against Lee Se-dol, a world-class champion, at the start of 2016. This was an important event because of the incredibly massive number of positions and plays possible in the game and how difficult it is to build an AI that can be efficient in evaluating the options before the extinction of the human race.

Blizzard has now partnered with Google to build a next-gen AI than can play StarCraft II, a computer game. It might seem redundant since most computer games feature a mode where the gamer plays against the computer. However, the difference is that an AI will be more intelligent, whereas the current simulators aren’t good at what they do – they’re just bigger thugs because of the various constraints or benefits offered by the programming.

At the moment, these games utilize scripts that determine how the games’ AI should respond to what the player does, which includes, at its foundation, the order buildings and units are created as well as a number of rules of how the AI should react to a number of scenarios. In other words, artificial intelligence in gaming is more an illusion powered by previously programmed scenarios and scripts.

The new AI that Blizzard and Google want to build would actually learn instead of simply reacting to conditions. It will remember how you play and learn from your play style, changing its reactions to suit the person it is playing against. It would also be able to teach new players the game more effectively. There will still be constraints to allow human players to win because a game like StarCraft is nothing more than a very large math problem, which means that an AI would wipe the floor with even the best human players.

Of course, we can’t expect to see games shipping out with an inbuilt AI as that would require a room full of servers. However, an AI could be used to create scripts for games modelled on its own behavior, resulting in a more effective computer opponent. Essentially, the AI would be teaching the game how to play more effectively against a human.