Watson on Jeopardy: Round 1
I’ve been waiting for this one!
Watson, an IBM computer system, started three days of competing on Jeopardy last night. It’s up against two of the games greatest players: Ken Jennings, with the longest win streak, and Brad Rutter, who has won the most money.
What’s interesting about a computer playing Jeopardy?
Jeopardy is full of puns and subtleties. Computers are not usually good at that.
That is Watson’s job…to understand the complexities of human language.
I thought Watson was going to the internet for the information, but according to the show, it’s not.
So, how did it do?
It was fascinating! Brad Rutter beat Watson to the answer, but Watson had a 70% probability assigned to the right answer. We get to see what Watson is considering. I’m guessing that’s why Watson didn’t beat Brad to the answer: it wasn’t sure enough to buzz in.
The next one went to Watson.
Interestingly, Watson next went looking for a Daily Double and got one. I’d anticipated Watson running straight through a column, the way many people do. I’m not sure Watson knows he is doing well on a category, by the way…he seemed to skip.
Watson got way ahead quickly. The human players started looking nervous, shifting their weight, practically rolling their eyes.
At the first commercial break, the score was Ken at $200, Brad at $1000…and Watson at $5,200.
At one point, Watson knew the answer was from Harry Potter…but apparently either didn’t know what a “victim” was or just knew the names and didn’t know what had happened. However, importantly, Watson knew it didn’t know…it had three possibilities shown to us but didn’t have high confidence in any of them.
Watson simply missed at least one question.
However, there were some disadvantages that had nothing to do with the programming.
Watson can’t hear. So, he apparently didn’t know Ken had already given a wrong answer, and gave the same wrong answer. That does happen with human players, but rarely.
This one bothered me a bit. The question had to with an “anatomical oddity” of a famous person. Watson correctly answered with the body part in question, but not the condition. It was scored wrong, and Brad got the money by getting it right. I believe a human player would have been asked to be more specific. If that was the case, Watson might certainly have gotten it.
I suppose that kind of instruction either couldn’t be given to Watson in whatever way it was getting the questions, or it couldn’t have understood it.
At the end of this first show, Watson and Brad were tied at $5,000. Ken was considerably behind at $2,000.
That’s only half of the first game, though. That game concludes today (Tuesday, February 15). There will be a second, full game on Wednesday, February 16).
The technology is honestly astonishing. I’m not sure how obvious it is when you are watching it, but Watson is understanding human language…deliberately tricky human language at that. Watson is answering questions on a wide-range of topics, with information it knows.
Where will this lead in the future? Help Desk, medical advice lines, reference librarians, personal shoppers…and you won’t have to press zero for an operator. 😉
For more information on Watson and Jeopardy, see this previous post.
This post by Bufo Calvin originally appeared in the The Measured Circle blog.