The Singularity is Near: Understanding Trivia, Not Trivial

Comments

Isn't it more impressive to say that Ken Jennings is almost as smart as a computer?

And isn't the technology applied here everything we wish Google was but is not? Where do I go to ask Watson questions?

From what I have read, he gets a live feed of the words as they are spoken. The impressive part is understanding the more complex worded questions that were not on display. Watson also searches in real time until he gets to a certain confidence level (what the number at the bottom was). Other videos I've seen on Watson show him playing jeopardy and getting some things ridiculously wrong as well. Sometimes it just comes off funny.

I agree with Andrew R.

I also think it's really, really impressive. As Jim said, it's everything we wish our computers could do for us. This is still a massive undertaking, and while it's a publicity stunt, it drives innovation and research. Maybe next year IBM will launch Watson.com, the next Google.

He did better on more "fact"-based questions -- still, impressive

Very impressive. I assume the computer is getting a live feed of the text of the question, though, rather than using speech recognition. It would also be interesting to know whether the human contestants were still (on those occasions when they answered first) edging it out on speed, or whether it couldn't answer those questions at all; is it always faster when it can answer?

Jim - the problem is that google has an 'adversary'. I presume Watson is only fed factual information. Google is most certainly not; it is also designed to be more broad than just finding facts (concepts!). If you want that, try WolframAlpha. Watson is like WolframAlpha 2.0.

Google or bing will not give you an exact answer to clues like this. They might lead you to a page with them, but they aren't going to speak them out loud.

The video has been removed. Since Engadget claimed copyright, I went to their website and found this link: http://www.engadget.com/2011/01/13/ibms-watson-supercomputer-destroys-all-humans-in-jeopardy-pract/

I don't know if it's the same, but hopefully it give you an idea if you didn't see the original.

PhysicsTeacher is right. Jeopardy is about fraction-of-a-second timing as much as knowledge, and computers have been better than us at timing for decades. Winning in a game without time constraints, such as Millionaire, would be more impressive.

Google already provides simple answers, as long as you ask the right question. (Try googling "temperature New York City" or "GDP Belarus" for example.) Watson merely expands on what Google does. It's a difference in degree, not kind. While it's impressive, it's no harbinger of the singularity.

Watson has to buzz in. If you go to IBM's website you can watch a lot more videos and they show a physical buzzer for Watson and mechanism that actually pushes the button on the buzzer when Watson wants to answer. I am sure they have the proper algorithm down for when to buzz in. However, I remember Ken saying after his run that he was really good at timing the buzzing and that it was a significant factor in his winning.

Also, the problem is incredibly, incredibly challenging for a computer. Jeopardy has puns!! How do you write a program to recognize a pun, in context? Or innuendo? The category titles in Jeopardy are usually a big hint in answering them but when you see them the first time you might not know what they are about. A category recently was "FRAN"s where every answer started with FRAN. How do write a program to be able to learn that? Jeopardy is filled with these one-off weird questions and it is very impressive that Watson is able to understand them and find the correct answer.

Comments for this post are closed