Google DeepMind, Blizzard Use StarCraft 2 For AI Research; What About Other Games?
Tech giant Google is teaming up with gaming publisher Blizzard to improve artificial intelligence (AI) through one of its games: StarCraft. However, it seems other games may fit the "teaching" bill as well.
According to Google, Blizzard announced in the annual BlizzCon that it will release tools that will allow third parties to teach AI how to play their hit game StarCraft 2. This is in collaboration with Google's DeepMind project, and the tools will be using the DeepMind platform.
The DeepMind team said StarCraft is an "interesting platform" to develop current research on AI. The gameplay itself may be a useful environment to simulate the "messiness" of the real world.
However, aside from StarCraft, there appears to be other games that AI systems may use to "teach" itself. Although an RTS may be a considerably good platform, other genres can help gauge certain aspects of an AI's program.
The nature of StarCraft as a real-time strategy (RTS) game is what makes it a hit among players. The game involves interconnected layers of decisions as players have to decide the effective use of resources, which buildings to make, and which units to build. There is also a separate piece of thinking when combat is being considered.
While there's a significant amount of research to AI's applications in games such as chess or go, StarCraft and its compatriots make a better simulation in the real world. This even tackles problems computers encounter real time such as logistics.
According tot Fortune, insights while playing StarCraft can help generate solutions for computers that are facing immense challenges.
However, other games offer nearly the same amount of intensity and require the same amount of problem-solving. Examples of these are massive online battle arenas (MOBAs) such as League of Legends and DoTA. Unlike StarCraft, the games require the usage of a single character through a massive campaign against other "heroes."
While the latter requires immense resource and unit management, MOBAs demand anticipation of team members and assumption of roles as well. This time it's not just you versus an opponent, it's a team versus a team. It will completely change the way AIs interact with each other.
StarCraft's "messiness" is even emphasized with the "fog of war" that is only unveiled when units cross this. Players then have to predict the decisions of their enemies while planning their own. This is not faced when AI challenge players to Go as the board is visible at once.
StarCraft's endgame is also a factor in consideration, as everything happens in real time instead of turns. This is extremely helpful for computer agents as they have more processing power than the human brain.
However, it appears these AI will have to work within the same limitations as human players - somewhat. AI players will still have to control the camera to learn about the conditions of the game and even execute mouse "clicks" within the "limits of human dexterity."
If an AI may also want to test its limits within human movement, it can also try playing first-person shooters (FPSs). It has the same amount of "human limitation" that playing StarCraft may offer, but only in a limited environment. The AI will have to assess its environment quickly and avoid dangers in its way with covers and hiding. The rather immense selection of "weapons" will also help AI determine the best "tools" depending on the situation and the environment. This means the randomness with opponents and game styles will opt the AI to try to adjust to its environment as best it could.
Regardless, there is a multitude of games that AI can use aside from StarCraft in order to meet its full potential. While gaming did have a negative image for "negative influence" on children, teenagers, and its players, it appears its contribution to the field of AI can change our perception toward the popular industry.