In their article, Vinyals, Gaffney, and Ewalds (2017) discuss the collaboration between Blizzard Entertainment, a video game developer, and DeepMind, an AI (artificial intelligence) researcher, to use video games as a platform for AI design and testing. A base for the research environment is StarCraft II, one of the most popular real-time strategy (RTS) video games (“StarCraft AI competition,” 2018). The researchers of DeepMind worked with Blizzard to design a machine learning APA that allows the AI developers to use the game and test their agents. Agents’ education happens in the form of missions where the AI acts as players, collecting materials, completing campaigns, and playing mini-games (DeepMind, 2017). The use of mini-games, however, is only a base for further exploration, and players, as well as researchers, are encouraged to create new games and missions for the agents to complete.
This opportunity to utilize games with a heavy focus on strategy and long-term planning allows AI developers to test their current knowledge and abilities in an area that requires multiple agents to complete various interrelated tasks. According to Peng et al. (2017), such environments as StarCraft are better suited for complex testing than previously used games such as chess. Here, the concept of collaboration leads to the increasing quality of communication and scalability of the agents (Peng et al., 2017). This joint effort between the companies is interesting because it introduces gamification of research in an environment with a large number of human players (Blizzard Entertainment, 2017). Therefore, in contrast to settings that are designed only for agents, StarCraft and Blizzard can offer DeepMind an enormous amount of data gathered from playtime. Here, the agents can be compared to human players, taught to perform a set of complex tasks that require an extensive understanding of the campaigns’ progression.
References
Blizzard Entertainment. (2017). The StarCraft II API has arrived.
DeepMind (2017). StarCraft II ‘mini games’ for AI research. Web.
Peng, P., Wen, Y., Yang, Y., Yuan, Q., Tang, Z., Long, H., & Wang, J. (2017). Multiagent bidirectionally-coordinated nets: Emergence of human-level coordination in learning to play StarCraft combat games.
StarCraft AI competition. (2018). Web.
Vinyals, O., Gaffney, S., & Ewalds, T. (2017). DeepMind and Blizzard open StarCraft II as an AI research environment. DeepMind.