OSDev.org

What do you guys think about IBM's Watson? He seems to be doing pretty good so far, fairly impressive technology as well.

Read a nice article on spiegel.de about it just yesterday. They interviewed one of the "fathers" of the system, who explained that the technology behind it is mostly brute force and statistics, having nothing to do with "real" AI.

Quite similar to chess computers, who "only" can do statistics on positions real fast, but have no idea of how a move or position "looks" or "feels" to a human opponent. (World class chess players still play chess software very much different from how they would play a human opponent.)

The funny thing is that most of the information Watson is using is basically "the internet", i.e. what the internet community as a whole has written. You could say Watson is little more than an automated Google client.

I just saw part one, and while I am somewhat impressed, I really think that the time and money could have been better spend elsewhere.

At least it should be better explained what Watson actually is, which I should think is a "answer interpretation machine". While Watson isn't connected to the internet, that's probably of little importance. In other words, if my fingers were fast enough, I could probably beat Watson at any time with a computer and internet connection without it being unfair, and I really suck at Jeopardy.

Just saw the second part. Watson did quite a bit better.

When a reporter asked Ken Jennings if he'd change anything, he answered: "If I were to play against Watson again, I'd probably focus my strategy around Industrial Strength Magnets."

"I for one welcome our new computer overlord."

Not a very inspired comment, but it made me laugh just the same.

Watson did quite well I think, but I do hope that the next time the developers will focus their energy on something more useful. I mean, even if this system worked flawlessly, it would still take a supercomputer to make use of it.

Why would something like Watson be a waste of time. Even though his method of answering the questions is portrayed here as kinda brute force (which it mostly is) the fact that it can actually understand the questions in the first place is quite the achievement, and the knowledge gained in developing that might one day be used in other applications.

davidv1992 wrote:Why would something like Watson be a waste of time. Even though his method of answering the questions is portrayed here as kinda brute force (which it mostly is) the fact that it can actually understand the questions in the first place is quite the achievement, and the knowledge gained in developing that might one day be used in other applications.

Who said it was a waste of time? Personally I think the time could have been better spend, but it's certainly not a waste.

davidv1992 wrote:...the fact that it can actually understand the questions in the first place is quite the achievement...

It would be - if Watson were actually able to do so.

Watson does not "understand" the question in the meaning of the word. What Watson does is turning the spoken question into text (i.e., simple voice recognition), then it "googles" the question in its database, followed by a statistical evaluation as to which phrase close to the question might be the answer.

It does neither "understand" the question nor the answer, as could be seen by its bad performance when it came to puns, playing on words, or anything creative.

(Example: "Derived from the latin word for 'earth', this is a place where trains end." (terminus) Watson answered "finis". It also repeated another candidate's wrong answer...)

Solar wrote:

davidv1992 wrote:...the fact that it can actually understand the questions in the first place is quite the achievement...

It would be - if Watson were actually able to do so.

Watson does not "understand" the question in the meaning of the word. What Watson does is turning the spoken question into text (i.e., simple voice recognition)

Actually Watson don't have voice recognition. The text is presented digitally and all Watson does is recognising the important part and such after which "he" will look up a probable answer in the database.

At least that is how I understand it.

Even less to get excited about. (Not that voice recognition software was that new. Well, the fact that it worked might have made the headlines. )