Computer System Learns Game Play by Reading Manual

Researchers over at MIT have developed machine learning systems that are capable of learning human language by simply reading a game’s manual.

MIT’s Computer Science and Artificial Intelligence Lab developed a computer system that taught itself to play popular online strategy game Civilization by reading the game’s manual, without human supervision.

The system has been developed on a script developed by researcher Regina Barzilay that allowed a computer to install software on a Windows system using the instructions provided by a Microsoft website.

Researcher S. R. K. Branavan, who is a part of the team that developed the system, says that the manual provided the system with open text in the form of instructions, which informed it about the actions that can be taken but did not tell the system how to win.

The computer system begins with no prior knowledge of what it has to do and learns to play the game gradually, when it reads the manual and starts experimenting with the controls.

“Games are used as a test bed for artificial-intelligence techniques simply because of their complexity,” Branavan said.

“Every action that you take in the game doesn’t have a predetermined outcome, because the game or the opponent can randomly react to what you do. So you need a technique that can handle very complex scenarios that react in potentially random ways”, he added.