Human brains are remarkably inefficient in some key ways: our memories are lousy; our grasp of logic is shallow, and our capacity to do arithmetic is dismal. Our collective cognitive shortcomings are so numerous I’ve written a book about them. And yet, in some ways, we continue to far outstrip the very silicon-based computers that so thoroughly kick our carbon-based behinds in arithmetic, logic, and memory.
Take, for example, our capacity to flexibly learn new things. Sure, I.B.M. has built monstrously fast computers to play chess and “Jeopardy!,” but Deep Blue and Watson were purpose-built, dedicated machines that excel only at the particular games for which they were built; Watson, built to conquer the game show “Jeopardy!,” would be hard-pressed to play chess. Your average ten-year-old, in contrast, can learn to play any number of games and well, if not quite at a world-class level.
How do human beings manage to be so flexible, and what would it take to make a machine equally supple in learning new things?
Earlier this month, Dr. Thomas Walter Murphy VII (the seventh in long line of Thomas Walter Murphys, originating with a Civil War soldier), put out a machine that could learn to play not one game but many, without any specialized prior knowledge about any particular game. For the past couple of weeks, Murphy’s invention has been making the rounds on the Internet, and for anyone who was a fan of the original Nintendo, the video of its performance is a must-watch. The program conquers a number of original Nintendo Entertainment System games, like Super Mario Brothers, where it manages to discover trick moves, attempt strategies, and push its way to the end of the game. (You can download the source code for yourself.) Sometimes, even when it doesn’t quite succeed, it still fumbles its way into doing something clever. It can’t win at Tetris, but it figures out that, if it pauses the game and doesn’t restart, it won’t lose.


