AI’s aren’t necessarily new to StarCraft, but Google and Blizzard thought that the concept needed a little more ‘umph’. If you’ve had your own fair share of StarCraft, you’ll know that no AI is match for human player, much a Gosu, which is a Korean term for “gaming god”. Well, seeing how Google’s impressive AI has crushed a number of games already, the question is if it has what it takes to conquer this famous RTS (real-time strategy) game. Let us give you a brief rundown for those of you who are unfamiliar with the game.
Source: Blizzard Entertainment
StarCraft is a RTS game created by Blizzard Entertainment in 1998, which had an equally famous sequel, StarCraft II, released in 2010, who received multiple expansion packs released across the years. The game typically starts with a simple choice between 3 different races, all of which have unique play styles and unit trees.
Once a game starts, you’ll have to mine for resources before you start expanding your base or building an army, which you’ll have to upgrade, send into battle, and hopefully, destroy enemy forces. It might sound simple, but all of the choices you’ll be making would be based-off of certain aspects of the game. Choosing the wrong time to build a specific building, or creating an army late, can very much make-or-break your chances of winning. If you think this is all too much for an AI to handle, there already is an excellent AI like that has existed for over the past few years: the Automaton 2000 Micro bot.
Source: Automaton 2000 via YouTube
Micromanagement
Micro, or micromanagement, is a type of strategy that StarCraft players use where they manage their units into full detail. Spreading out your units, going for a hit-and-run, and the notorious zerg rush unit management, these are all basic tools for a StarCraft Gosu. This strategy focuses on scouting, positioning, harassment, and even combat. As we mentioned earlier, Automaton 2000 is an AI famous for mimicking these types of play. But no, this is not the kind of AI that Google wants to create. They’re striving to create an AI who can learn how to do things like these, and not just doing it because that’s how they’re programmed.
Source: DeepMind via YouTube
Google’s aim is to create an AI environment where the AI can dissect all the information and core principles of the game. Differentiating between units and seeing how Terrans are not the same as Zergs, how to tell if the unit is an SCV or a Drone, and even how to understand and interpret unit health bars. It’s not about creating an AI to play against, it’s about understanding how an AI can compare against a game that’s basically a visual mess for non-gamers. It’s interesting to think how an AI can cope and react to different types of scenarios, and seeing how DeepMind is already the best at the field of Chess, this should be something to look forward to.
https://www.youtube.com/watch?v=eThrmFcZ_ro
Source: YouTube
Blizzcon 2016
Google announced at Blizzcon 2016 in Anaheim, California the collaboration with Blizzard Entertainment in developing AI that can deal with the highly-complex workings of StarCraft II, which could directly translate into the messiness of the real-life world. The use of games in AI development is very efficient, seeing as they provide an environment that allows for smart testing and also allows quick feedback with scores. Seeing how daunting RTS games like StarCraft are, one can treat AI’s at an astounding rate of efficiency with all the complex calculations needed to win a simple game. DeepMind started from 2D Atari games, Chess, Go, and are now venturing into more advanced 3D games like StarCraft. Seeing how DeepMind had dominated their previous opponents, it’s exciting to see how StarCraft II is going to fare against it.
Article Sources:
