Not exactly current news, Watson’s appearance on Jeopardy! I’ve had
exactly a month as of today to think about it, and here’s what I think.

I used to watch Jeopardy! all the time back in the 1990s or something like that.  I haven’t really watched it much in the past 15 years or so, although I did tune in for special events like Ken Jennings winning a zillion games in a row a while back.  I only watched some parts of the Ken Jennings/Watson/Brad Ritter match back in February.  I tuned in for the last part of Monday’s episode and the first part of Tuesday’s, but the show seemed more like an infomercial for IBM than a game show, and I’m not a big fan of IBM.  I didn’t watch on Wednesday.  Of course, I didn’t need to tune in to find out the result, namely that Watson won, since it was so well-publicized.  Incidentally, it seems funny to me to name a computer after Thomas Watson, who is the same person that said in 1943 that “There is a world market for maybe five computers,” but I won’t pursue that idea further.  What I am interested in looking at was the following points.

It’s pretty obvious that the reason why Watson won is not because it is “smarter,” however you define that, than the competition, but rather because it’s a lot faster on the signalling button.  I’m not really sure what the configuration was for Watson’s signalling button, but regardless of what it was, the signalling path had to be a lot faster than it was for the two humans. Their brains aren’t hooked up directly to the signalling button; rather, their brains have to register that the question has finished being read, and then the brains need to send slow electrical impulses to the fingers, which then need to move to activate the signalling button.  So, this achievement just demonstrates how electronics are faster than the human brain (which everyone already knew), not that computers are better at answering questions than humans.

Next point.  To start, I believe that it was Marvin Minsky who said something along like “AI is anything that we haven’t done yet”.  It’s a relevant quote, since it illustrates that, once we understand how to do something, it isn’t anything special.  So, on the one hand, we probably shouldn’t discount Watson’s “achievement” solely based on the fact that a machine managed to do it and therefore it isn’t all that special, but on the other hand I think there’s a lot of room for improvement for machines.  Take the Final Jeopardy! answer on Tuesday for example.  I don’t remember exactly what the answer was anymore, but the category was “U.S. Cities” and the answer was something along the lines of “One of this city’s two airports is named after a World War II flying ace; the other, a World War II battle”.  Watson answered (queried?) “What is Toronto?”  To me, this shows a significant defect in the machine’s semantic representation of the answers and questions.  First of all, what human would provide that answer?  I’m sure that anyone, no matter how smart or dumb they are, would at least provide an answer that is a U.S. city.  If you had a highly advanced semantic map, you would realise that the answer has to be a really big city; even a city as large as Toronto only has one international airport (unless maybe you count Hamilton airport as serving Toronto).  New York?  No, its airports are named after a president and a mayor.  Los Angeles?  No.  So you might get to Chicago, and just stumble on the right answer that way.  It appears that Watson answers questions in a way that is incredibly different from how humans do, and I think that could be a significant disadvantage for it.

During part of the “infomercial”, some IBMer suggested that people might want to use this technology for intelligent agents who answer people’s questions online or whatever.  I would doubt it.  First, how much will IBM want you to pay for this sort of technology?  If history is any precedent, this won’t come cheap.  It’s probably a lot cheaper to hire people in India to chat with a website’s users.  Second, it’s a lot easier to get people to understand search engine syntax and semantics than it is to get machines to understand people semantics.  Why would anyone want to type “What’s the best resource on the web about mathematical paradoxes?” or whatever subject matter you’re interested in, when it’s a lot easier, and most people know, to just type “mathematical paradoxes”.

One final point:  I think that machines’ accomplishments such as this one can’t be considered equal to that of humans until they are intentional.  In other words, until the computer chooses to show up to the Jeopardy! match, I don’t think that the accomplishment can be considered to be equal to that of a human.  We don’t crown a pitching machine as Cy Young Award winner or a cheetah as a gold medallist in sprinting, and I think the significant thing is that these objects cannot choose or appear to choose to attend the sporting events, so similarly, Watson’s accomplishment is not complete without Watson actually choosing to show up to Jeopardy!