
The Google DeepMind project will learn its AI how to play Starcraft II after learning other games.
Beacon Transcript – Google’s DeepMind, the Artificial Intelligence development project, will be teaming up with Blizzard Entertainment so as to teach the AI how to play Starcraft II.
With the Artificial Intelligence (AI) being one of the most up-and-coming domains in technology, the system is being introduced in a multitude of varied fields so as to enlarge its databases.
Google, the giant Internet company has its own separate AI development division and its DeepMind project will start acquiring knowledge through the use of an unexpected medium.
The project’s research team will be collaborating with the Blizzard Entertainment group as they develop an API or Application Program Interface.
This will allow AI research environments the possibility to gather data and learn based on the Starcraft II game. The API will also be made available sometimes next year to other interested researchers and developers in the AI domain.
The API will allow programmers to use and train AI players to play the game, which could contribute to the acquiring of new skills in both human players and the AI.
As the players chose to use the aforementioned artificial intelligence players, the system itself would benefit from this as it would come to gather new, more varied data than in a laboratory which features no real-world scenario scripts.
AI agents have already mingled in a number of games, mostly Atari classics and the Asian Go, both of which they have learned to play as good as any human player, even managing to defeat their competition.
According to statements, Starcraft was chosen as the next game challenge for the DeepMind project thanks to its more complicated set of rules which cannot be pre-programmed, as compared to classics such as Chess and Go whose rules can be introduced.
The game will not feature readily available information so the AI player will have to explore the constantly changing game environment so as to be able to plan its future move.
The AI agent will also have to develop a quick strategy so as to combat its opponents and counterattack their moves.
The artificial intelligence player will as such have to learn and train itself to develop an effective memory system which will also help it generate spontaneous plans as opposed to the long-term planned moves specific to chess or other such games.
The Starcraft II real-world scenarios will also contribute to the program’s knowledge as it could account for the possibility of transposing Google’s DeepMind to more real-world applications sometimes in the future.
The DeepMind research team will continue their partnership with Blizzard so as to develop a series of training missions or “curriculum scenarios”.
This will help researchers determine the acquisition and effectiveness of certain skills based on different advances and algorithms.
The future API release will allow the interested developers the option of developing their own Starcraft II scenarios.
Image Source: Flickr