http://www.wired.com/2016/01/in-a-huge-breakthrough-googles-ai-beats-a-top-player-at-the-game-of-go/
This article published a few days ago talks about the game of Go and how google developers in Britain have developed code and then ran the code against itself to self learn. We talked about in class how we have figured out checkers, but not chess because it's too complicated and the game of Go is more complicated than chess. They believe the system used for learning can apply to almost anything and in 2014 thought it would take almost 10 years to get it to work. I think this is a large breakthrough for AI and it makes me wonder if we are finally evolving code quicker than expected.
In CS, there is a long history of finding that we have to re-predict our capabilities in the future. In the search for new protocols and capabilities for code to learn against itself, we may find entirely new ways of solving problems. So, for example, by *today's* means, creating an AI that can win any game of Go could require a decade of work. But given a new method of solving the problem (such as advancements in self-learning code), we could potentially solve the problem much sooner.
ReplyDeleteI think it is really interesting that ran the code against itself in order to get it to learn how to play. I know they have not figured out how to beat the world champion yet, but what will be the next impossible game to beat after go? Is there a more difficult strategy game like this that anyone knows of?
ReplyDeleteThe last two seminars in the Math department was focussed on Go and they claimed that the reason Go was so hard to code was because of the pattern recognition required in the game more than it does require logic. I think having a program solve Go will be such a break through in visual pattern recognition in computers given that it is one of the things humans are currently better at than computers.
ReplyDelete