There’s a lot of buzz in the media today about the news that a computer has beaten the two all-time top players of Jeopardy, the US game show that uses word play and fast reactions to test competitors’ general knowledge and lateral reasoning skills.
The computer is Watson, created by IBM, and it uses a sophisticated parallel program to try to come up with answers to questions. The New York Times published an in-depth report last year, which includes an overview of how the program works. Basically, it explores thousands of different ways to tackle the question in parallel and then ranks the answers according to confidence. The confidence check includes a basic fact check, such as making sure that somebody was alive at the time something happened if the two are implied to be related. If the program has enough confidence in its answer, it will buzz in. It’s a similar approach to the one I mentioned previously which could help increase software reliability by running multiple algorithms and then comparing the results.
The algorithms work by matching words in databases of information, and presumably performing some basic natural language processing to understand the context of some key words. Applications touted for this technology include advising doctors rapidly on the best treatment based on vast databases of medical papers, or providing automated customer service.
The good news is that you don’t have to just read about it: you can try out the technology yourself. The New York Times has a Flash game where you can compete against the software. I’m guessing this is a simulation using Watson’s stored answers, rather than being a Flash interface to the real computer. But it still gives some insight into the software’s strengths and weaknesses. Good luck!