Learning by playing: AI can dominate most human competitors in Freeciv

Beating the AI opponents in a strategy game is the first step any gamer takes before heading online where the real challenge is. Whether they cheat or not, most game AIs are beatable, often with simple, repeatable strategies once you find their weak points. Not so with Arago’s flagship AI. Named HIRO (Human Intelligence Robotically Optimized), the algorithm can beat almost all players — and that’s not even its main job.

Arago is an IT automation firm, which develops smart AIs that can streamline businesses and automate many of their functions. HIRO is one such AI, and while it does an excellent job of improving workflows at a number of corporations, it’s the way it’s trained that is most fascinating. It plays — very well at that — the freely available civilization-building game called Freeciv.

Much like Sid Meier’s series, Freeciv is a game about deep strategical choice over many turns and many hours. Typical AI in games like this are easily outstripped by human players, but not so with Arago’s HIRO, which has beaten 80 percent of the humans it’s gone up against and is still getting smarter (thanks TechCrunch).

Getting better at anything requires training, whether you’re a human or AI like HIRO. To that end, Arago made it possible for HIRO to understand words like “city,” and “tile” in order to teach it in a more humanlike manner. It’s the restructuring and recombining of the lessons it’s been taught over time that make HIRO so versatile and ultimately capable of beating almost any opponent that’s thrown at it.

Arago is rather proud of this achievement, pointing out that Freeciv is a highly complicated game, with many more permutations of moves than the Go game that Google’s AI recently bested a world champion in.

Those same abilities are transferred over to the more business-centric tasks HIRO handles in its day job.

Developments like these are why OpenAI recently announced an outsourcing of its AI development platforms through its new Universe initiative. Training AI to play games like we do may be one of the best ways to teach them to be versatile.