Tuesday, February 22, 2011


A very interesting e-conversation between my husband and my son transpired this weekend, all brought about because my husband had some concerns about Watson, the computer who beat out the contestants on Jeopardy a few days ago.

With Sean and Jerry's permission I'm transmitting those e-mails to you folks, who may have wondered the same things. My son Sean is, among other things, by profession a computer programmer, but you need to remember that here he is explaining things to an 81 year old man and his father-in-law to boot! Here goes

On 2/18/2011 12:54, Jerry wrote to Sean:

All right, so how does Watson work? Is it programmed with all of the day's categories, questions, and answers before hand?

Does it "react" by voice recognition? If so does Trebek have to give it a voice signal of some kind? Otherwise, how does it know the question is over and it's time to buzz. It's probably a given that it can usually buzz faster than any human, so the key is how does it know when to do it?

Incidentally, if you're not aware of it, Watson is the last name of the founder of IBM.


On 2/21, after returning from out of town, Sean explains:

Watson is nothing more or less than a very high powered computer. There are some 2800 processors adn 16 Terra Bytes of memory - your computer probably has 2 Giga Bytes

giga = 10 to the 9th power
terra = 10 to the 12th power

So the first thing to know is that it was an extremely powerful computer with one hell-of-a-lot of memory.

The second thing to know is that aside from some *very* clever programming, it basically used very basic computer science techniques from a class of problems called "searching" (there are textbooks written on searches alone, and of course, google is nothing more or less than a search company). Not to take away from the clever programming involved, Watson's programming was intended to allow it to focus on "key words" in a question and to try to relate those key words to other pieces of data in it's huge databanks. It built a number of "probable answers" and ranked them based on a "confidence" value, and if that value was of a certain strength, Watson would ring in.

Third, Watson was given neither the questions nor the answers ahead of time - it had a huge memory-bank of thousands and thousands of facts, trivia, and such. Nor did it "listen" to Alex - it couldn't hear. The questions were typed at a keyboard as Alex was reading them. So it was essentially competing on an even footing with its human competitors.

Fourth, while we never see it, there is a light somewhere on stage that turns on when the players are allowed to "ring in". Until that time, it makes no difference how many times you push your button, it's ignored. Ken Jennings, one of the two competitors was widely regarded as tremendously fast on the button, which explains why he won so many games (well, and he knew the answers too.) The fact that Watson dominated the games the way that he did raises the one flaw in the process that I believe doomed the human players - there is that fraction of a second between "knowing" that it's OK to buzz and when the thumb actually is depressed. While Jennings may be fast, he is not speed-of-light fast, which is how fast the signal that Watson receives travels - that gave it a HUGE advantage.

At the end of the day, the Jeopardy match was a publicity stunt, just like the chess game with Gary Kasparov and Deep Blue many years ago.

The real value of Watson will make itself known in other venues such as medicine. One of the blogs I follow was talking about that the other day. There is SO much that a physician must know these days, that to have access to something with the vast resources of a tool like Watson that can take input from a patient and correlate that to the HUGE number of possible causes will give your average GP a far better chance of coming up with a better diagnosis.

Ken Jennings wrote a piece about his experience: http://www.slate.com/id/2284721/

Since Jerry forwarded these posts to me, I got into the act:

On 2/22/2011 07:17, I wrote to Sean:

Morning, sweets! Jerry forwarded this exchange on to me. He and I had discussed Watson and Ken Jennings earlier and he was sure the whole thing was rigged like the old TV game shows were. I see by his e-mail to you that I hadn't convinced him and so he went to you for the definitive answer.

So, I think the e-mail exchange would be a delightful post for a future blog. Do you mind me using your explanation -- and/or would you like to revise it in any way before I do? I understand that ways you explain computers to aging parents might not be the image you would like to leave for the world (although the world is NOT reading my blog -- mostly just old people like me.) But I did find the whole thing interesting and if there is a way to use it, I'd like to.

Give me some feedback, ok? xoxoxox Mom

And finally, the reply from Sean. It will end the blog, and I give him a tip of the hat for being such a help to his old mom and pop!

Of course you can use it. I would add that much of my reply to Jer was discussed during the games themselves in some of the clips in which Alex and the IBM staff discussed what was actually going on, so it's not as if I have a great insight. Perhaps the best that could be said of my post was that it distilled things down to the essence.

I should also point out that the "clever programming" which I allude to was in fact *very* clever. One of the hardest problems that remains to be solved (and I don't think that Watson truly solves it yet) is that of natural language processing. There is *so* much that goes into carrying on a conversation; the syntax and semantics of language is trivial when considering the art of the pun and double entendre, for example. That Watson was able to do as well as it did in some of those categories, well, to simply call that "search" (which it certainly was) is to say that space flight is similar to paper airplanes - while true at some level, it doesn't capture the vast differences.

Watson is a HUGE improvement over the machines that came before it. That said, it still occupied an air conditioned room and would not have fit on the stage on which the event was taped. Had that air conditioning failed, Watson would have died a very HOT death. Compare that to the couple of pounds of gray matter inside the heads of the two human contestants and I think you'll see that it was advanced, yes, but it still has a ways to go to truly "beat" a human being. S

1 comment:

marciamayo said...

It's a whole new world out there.