The trek towards a Terminator reality goes forward as scientists at MIT’s Computer Science and Artificial Intelligence Lab have successfully ‘taught’ a computer how to play a game. The results were encouraging, as the computer was able to win 79% of the games that it played.
Working under a project supported by the National Science Foundation, the team at MIT worked to improve machine learning by utilizing games. In this instance, the computer was given the task to ‘learn’ how to play Civilization by running over the manual. The results came back rather positively, showing that computers can be made to learn the meaning of words and work to translate their meanings into proactive use.
MIT rationalized that by giving the computer a set of instructions, the game environment could be used to put the system through its paces. "Game manuals have very open text. They don’t tell you how to win. They just give you very general advice and suggestions, and you have to figure out a lot of other things on your own,” said associate professor of computer science and electrical engineering Regina Barzilay.
She noted that after applying additional techniques to supplemental abilities of the computer, there was only a 62% success rate.
All in all, MIT remains hopeful on the project, saying that this is yet “another step closer to the real world.”
