Talking robots are nothing new. My toy robot, 2XL, could do that. He could even laugh, and make me laugh. But of course 2XL, like every talking machine since, wasn’t really exhibiting intelligence. Behind those words, were just some tricks. In 2XL’s case, it was a clever use of an 8-track tape.
And, as computer science has advanced, so too have the tricks. In fact, the tricks may be getting so advanced, that it soon may be impossible to tell where tricks stop and intelligence begins. Indeed, we may discover that intelligence is really just a very advanced bag of biological gimmicks.
But the language trick for computers is a serious challenge. I remember when I worked in a Board of Education office in Japan, one of my Japanese colleagues approached me with a print-out with some text on it. Looking it over I could see that it was English…or was supposed to be. The words were English alright, but their meaning was lost in a scramble of nonsensical grammatical errors. I was then told that this was the work of a computer translation program and I was asked to judge its quality. Well, I had no idea what it was talking about, even after several minutes of decoding work.
This kind of linguistic screw-up by a computer won’t surprise anyone. What would be surprising is a computer that could talk…for real. No matter how great our machines, they just can’t crack the mysteries of language (and frankly I think this is perhaps the most intriguing mystery yet to be solved).
Despite the difficulty, we are getting closer to intelligent machines, at remarkable speed. Google Translate is now catching people’s attention for its “uncanny” ability to get translation right much more of the time than we’ve been used to. And now, comes next week’s Jeopardy match-up between the two best Jeopardy players and a computer from IBM called Watson.
[A small aside here: 2XL once told me how Stanley Kubrick came up with the name for HAL, the killer AI computer in 2001: A Space Odyssey. Kubrick just took the letters IBM and used the letters that preceded them in the alphabet. After explaining this, 2XL then gave one his endearingly ridiculous laughs.]
Anyway, it was interesting when Deep Blue (that other wonderbot from IBM) beat the world’s greatest chess player, Garry Kasparov…but that wasn’t really that amazing since we’ve known that games are often just plays on probabilities and statistics, things that computers can do quite well. Jeopardy, on the other hand, that’s a game based on language…at least to a large extent…so getting Watson to even be able to compete is a big deal.
As a librarian, this is starting to cross the interesting-line. Watching Watson tackle complex questions (or answers in the case of Jeopardy), sorting through the immensity of online databases and digital materials and then coming up with the right answer (in spoken language no less!), well, this is worth a standing ovation, if only Watson would appreciate that. In a sense, he’s doing reference work…a very challenging job indeed.
But watching this spectacle, you can quickly see the potential for online versions of Watson, which hopefully IBM will be working toward shortly after their PR coup on America’s smartest quiz show whets our appetite for interfacing with this proto-entity. We’ve seen similar tools through Google and, especially, through Wolfram-Alpha, which can use natural language questions and return quite accurate results. But what is really neat about Watson, is that he can actually learn from mistakes and get better over time.
Remember my posting about the kinds of skills incoming librarians need? Add statistics and computer learning theory to the mix. And if you’re Watson, take a tip from 2XL, and get yourself a cute laugh. People won’t be afraid of you so much.