Check this out.
That’s a robot winning at Jeopardy. Sounds easy at first. But think about the wordplay, the riddles, the esoteric details, which make up most Jeopardy questions… And in spite of that, Watson gets most of the questions right.
According to this Popular Science article, a modern computer would take about 2 hours to come up with the accurate answers Watson can get in about 3 seconds. Watson has 2,800 processors to do that.
Watson isn’t connected to the internet. I guess that’s to be technically fair. But it is pre-loaded with tons of information: encyclopedias, dictionaries, thesauruses, plays, books, movie scripts, and so forth.
The cool thing is that it even learns while it’s playing. If Watson doesn’t understand what a category is, it won’t select it until its competition has, and a couple questions have been answered. With that data, it can learn what patterns there are, and assess whether it’s a good category to select.
One of the biggest risks for Watson’s question answering process is whether it interpreted the question correctly. It’s far more likely for this computer to be completely wrong than ‘close,’ as many humans are when they answer. And it’s got a risk assesment algorithm as well, which decides whether it’s confident enough to answer at all. And if Watson is winning by a lot—or losing—it might take bigger risks.
There’s going to be a two-day tournament between Watson and Jeopardy champions Ken Jennings and Brad Rutter. The winner gets $1,000,000, while second and third get $300k and $200k respectively. That tournament is going to air on February 14th (mark your calendars). And look forward to Alex Trebek’s interview with a computer.