To learn how humans and AI systems can best live together, we may need to kill a whole lot of Zerg.
Starcraft II has been a target for Alphabet’s DeepMind AI research for a while now – the UK AI company took on Blizzard’s sci-fi strategy game starting last year, and announced plans to create an open AI research environment based on the game to make it possible for others to contribute to the effort of creating a virtual agent who can best the top human StarCraft players in the world. Now, DeepMind and Blizzard are opening the doors to that environment, with new tools including a machine learning API, a large game replay dataset, an open source DeepMind toolset and more.
The game is more challenging than most of those tackled by AI programs to date. Not only is StarCraft extremely complex, it also requires planning far ahead and trying to second-guess what your opponent is up to. This means developing AI programs capable of matching humans ought to help researchers explore new facets of humanlike intelligence with machines. One other potential benefit, according to those involved, will be exploring ways for humans and artificial agents to play together.
“StarCraft is interesting for many reasons,” says Oriol Vinyals, the DeepMind researcher who is leading the project. The fact that players often get only a glimpse of their opponents’ activities, for instance, mean that algorithms will need to develop better ways of storing information in memory. “Memory is critical,” Vinyals says. “What you see now is not what you saw a while ago, and something specific that might have happened a minute ago might make you want to act differently.”
DeepMind has built an impressive reputation on building AI programs capable of playing various types of games with superhuman skill. The company began by conquering various Atari games and more recently it took on the extremely complex and abstract board game Go (see “DeepMind’s AI Masters the Game of Go a Decade Earlier than Expected”).
Within StarCraft, players compete as one of three races: the humanlike Terrans, the cyborg Protoss, or the insectoid Zerg. Battles involve complex strategic actions like mining resources and constructing bases, as well as protracted battle sequences. StarCraft is also the most popular spectator e-sport, and in South Korea especially, tournaments are often played in massive stadiums and shown live on television. Prominent players have welcomed the prospect of matching up against AI programs, but DeepMind hasn’t yet said when this might happen (see “StarCraft Pros Are Ready to Battle AI”).
StarCraft II is such a useful environment for AI research basically because of how complex and varied the games can be, with multiple open routes to victory for each individual match. Players also have to do many different things simultaneously, including managing and generating resources, as well as commanding military units and deploying defensive structures. Plus, not all information about the game board is available at once, meaning players have to make assumptions and predictions about what the opposition is up to.
It’s such a big task, in fact, that DeepMind and Blizzard are including “mini-games” in the release, which break down different subtasks into “manageable chunks,” including teaching agents to master tasks like building specific units, gathering resources, or moving around the map. The hope is that compartmentalizing these areas of play will allow testing and comparison of techniques from different researchers on each, along with refinement, before their eventual combination in complex agents that attempt to master the whole game.
The whole goal here is to come up with AI that can play StarCraft II better than any human can, in much the same way that DeepMind did with its AlphaGo software for playing the ancient physical board game of Go. DeepMind wants this to propel the existing research forward, hence its appeal to larger research community and this open release of tools.
The new release of the StarCraft II API on the Blizzard side includes a Linux package made to be able to run in the cloud, as well as support for Windows and Mac. It also has support for offline AI vs. AI matches, and those anonymized game replays from actual human players for training up agents, which is starting out at 65,000 complete matches, and will grow to over 500,000 over the course of the next few weeks.
Jacob Repp, a principle engineer at Blizzard, says his company is interested in seeing if sophisticated AI agents could make the game more interesting, either by playing against people or collaborating with them. It is already possible to create agents in the game that follow scripted commands. Repp says it would be interesting to have those agents use machine learning to some degree, too. And he says the company is exploring these sorts of ideas. “We’re finding that these tools are very useful for the process of making games and designing features in games,” he says.
Source: DeepMind
