These Computers Are Learning By Teaching One Another Pac-Man
Researchers at Washingon State University have touched on a very fascinating way to improve machine learning and artificial intelligence: by having said machines teach each other to play video games. Their plan is to havew one computer show another how to play Pac-Man. Instead of simply moving the programming from one system to another, they've developed a number of algorithms to simulate the learning process.
"We designed algorithms for advice giving, and we are trying to figure out when our advice makes the biggest difference," said AI professor Matthew E. Taylor. According to Taylor, the 'teaching' computer shows the 'student' computer how to play "well;" showing it the best techniques not get eaten by ghosts and how to grab the most points. Eventually, the researchers decided to make things a bit more complex, and set the two systems to play Starcraft. At certain points, they noticed that the student system actually ended up being better at it t han the teacher.
The trick, they explained, was teaching the system when it should focus on giving advice, and when it should simply sit back and let things happen. If a teacher gives advice too often, explains a release, then the student becomes annoyed and doesn't learn to outperform the teacher. If, however, the teacher gives too little advice, the student won't ever manage to learn what it's doing. There needs to be a balance.
Not surprisingly, for this reason, the research could have impressive applications beyond simply
This research could have applications beyond simply improving artifical intelligence and machine learning. Computer-assisted learning, for example, could greatly benefit from what the study's results. After all, if a robot could teach another robot to play Pac-Man, it stands to reason that it could also teach a human being mathematics, science, or any of a wide array of other knowledges or skills. Of course, that's probably not a terribly comforting thought for flesh-and-blood teachers, who I'd imagine aren't keen on the idea of being replaced by a robot.
Check it out! I have an RSS Feed now!