I forgot to post a heads-up yesterday about Watson, the IBM supercomputer, being on Jeopardy last night. There's lots to say here.
First, this is a great way to popularize science. I only wish they'd done more and better promos for the show. From my informal survey, not too many people were aware of it. It was a big event -- Machines v. Humans, round 1. (I know, I know: Big Blue beat Kasparov in 89 but that was merely chess; this is a game of knowledge and language.)
PBS had a show last week about how IBM created the computer. It was not a simple task. Jeopardy's questions are often double entendres, and computers have a very difficult time understanding human language even without that complication. They used machine learning to get over this hump.
They gave Watson all the past Jeopardy questions and all the correct answers, and let it find the patterns within the information. In other words, it learned what sorts of answers work for what sorts of questions. No one taught Watson this trick. It taught itself. After this sojourn into machine learning, the computer's ability to understand the questions jumped from really bad to damn good.
As for last night's show, it was fun. Watson ended up tied for first place with one of the contestants, and the match continues tonight and Wednesday. There's still time to tune in, even if you missed the first show.
It's interesting to note that Watson's answers were wrong a lot of the time -- though not that often when he actually answered a question. (They showed you onscreen what he would have said, even when another contestant took the question. So you could see that he got a lot of the answers wrong -- but it didn't count since it wasn't his turn to answer.) This means the end result is truly a matter of luck. Watson could get two-thirds of the questions wrong and still win if his actual responses to questions happen to fall into the third of the answers where he guessed correctly. I like that. It really is a game, despite everything.
The thing that was most fascinating to me was when Watson got an answer sort of right, but in a way that showed he didn't really understand what he was talking about -- and didn't even grasp the exact nature of the question. And that's because there's nobody home inside Watson. He's a Chinese Room. Nonsense going in; nonsense going out -- and in the middle, only a set of rules, no intelligent entity of any sort. There is no comprehension in Watson. There is only pattern recognition, sorting and ranking. Kinda fun.
Anyone else watch the show? What did you think?
First, this is a great way to popularize science. I only wish they'd done more and better promos for the show. From my informal survey, not too many people were aware of it. It was a big event -- Machines v. Humans, round 1. (I know, I know: Big Blue beat Kasparov in 89 but that was merely chess; this is a game of knowledge and language.)
PBS had a show last week about how IBM created the computer. It was not a simple task. Jeopardy's questions are often double entendres, and computers have a very difficult time understanding human language even without that complication. They used machine learning to get over this hump.
They gave Watson all the past Jeopardy questions and all the correct answers, and let it find the patterns within the information. In other words, it learned what sorts of answers work for what sorts of questions. No one taught Watson this trick. It taught itself. After this sojourn into machine learning, the computer's ability to understand the questions jumped from really bad to damn good.
As for last night's show, it was fun. Watson ended up tied for first place with one of the contestants, and the match continues tonight and Wednesday. There's still time to tune in, even if you missed the first show.
It's interesting to note that Watson's answers were wrong a lot of the time -- though not that often when he actually answered a question. (They showed you onscreen what he would have said, even when another contestant took the question. So you could see that he got a lot of the answers wrong -- but it didn't count since it wasn't his turn to answer.) This means the end result is truly a matter of luck. Watson could get two-thirds of the questions wrong and still win if his actual responses to questions happen to fall into the third of the answers where he guessed correctly. I like that. It really is a game, despite everything.
The thing that was most fascinating to me was when Watson got an answer sort of right, but in a way that showed he didn't really understand what he was talking about -- and didn't even grasp the exact nature of the question. And that's because there's nobody home inside Watson. He's a Chinese Room. Nonsense going in; nonsense going out -- and in the middle, only a set of rules, no intelligent entity of any sort. There is no comprehension in Watson. There is only pattern recognition, sorting and ranking. Kinda fun.
Anyone else watch the show? What did you think?