Page 2 of 2
Re: IBM Watson beats human champs in first public test
Posted: Thu Feb 17, 2011 5:46 pm
by Phelan
hyperpape wrote:It's what we are--the human type of chimp

.
I seem to recall there being a biologist or primatologist who referred to us as the third species of chimpanzee.
Pan Narrans, the Storytelling Chimp.
Re:
Posted: Thu Feb 17, 2011 6:22 pm
by Fedya
EdLee wrote:fwiffo wrote:I wonder if he'd still win if they eliminated the buzzer as a factor.
How do you mean? I thought they built a mechanical buzzer for Watson so Watson had to buzz just like all the human contestants.
The same buzzer rules applied to Watson, too.
Watson's relay presumably operates at whatever speed electrons in a current can move, which is relatively close to
c. That would be much faster than human reflexes.
Re: IBM Watson beats human champs in first public test
Posted: Thu Feb 17, 2011 6:28 pm
by fwiffo
Watson's mechanical solenoid finger has better reflexes than a human finger. That doesn't tell us anything about AI - just that we can build mechanical things that are better than humans at some tasks, which we already knew.
I'm suggesting that you could ask every clue in the format of Final Jeopardy, where all the players answered and the buzzer isn't a factor. That would be a more interesting test of its language understanding and answering ability.
Re: Re:
Posted: Thu Feb 17, 2011 6:34 pm
by Bill Spight
Fedya wrote:EdLee wrote:fwiffo wrote:I wonder if he'd still win if they eliminated the buzzer as a factor.
How do you mean? I thought they built a mechanical buzzer for Watson so Watson had to buzz just like all the human contestants.
The same buzzer rules applied to Watson, too.
Watson's relay presumably operates at whatever speed electrons in a current can move, which is relatively close to
c. That would be much faster than human reflexes.
A built in delay of around 0.1 sec. seems reasonable to me.
Posted: Thu Feb 17, 2011 7:00 pm
by EdLee
Bill Spight wrote:A built in delay of around 0.1 sec. seems reasonable to me.
Yes, this is nice. Like finding the correct komi. They'd have to do some experiments to find out the correct delay.

Perhaps they had already added a delay?
Re: IBM Watson beats human champs in first public test
Posted: Thu Feb 17, 2011 7:22 pm
by hyperpape
It would be nice to know how Watson would perform in a final jeopardy like format. But from the perspective of AI research, isn't the really important thing that Watson can answer Jeopardy questions on a level roughly comparable to (very very talented) humans? Winning is just icing on the cake.
P.S. Are any of Jennings, Watson or R.whatshisname? good enough that they could just buzz in on every question and end up with a positive score? I read speculation that Jennings was intentionally buzzing in before he'd confirmed that he knew the answer, just for the sake of beating Watson to the punch.
Re: IBM Watson beats human champs in first public test
Posted: Thu Feb 17, 2011 8:03 pm
by jts
hyperpape wrote:It would be nice to know how Watson would perform in a final jeopardy like format. But from the perspective of AI research, isn't the really important thing that Watson can answer Jeopardy questions on a level roughly comparable to (very very talented) humans? Winning is just icing on the cake.
P.S. Are any of Jennings, Watson or R.whatshisname? good enough that they could just buzz in on every question and end up with a positive score? I read speculation that Jennings was intentionally buzzing in before he'd confirmed that he knew the answer, just for the sake of beating Watson to the punch.
In buzzer competitions you usually want to buzz in before you know the answer. Then you figure it out while they call on you.
Re: IBM Watson beats human champs in first public test
Posted: Thu Feb 17, 2011 8:11 pm
by hyperpape
That's true, but there's still some variable under the player's control here. There are questions where none of the players buzz in. They're obviously waiting for some information, at least in some categories (perhaps Jennings decides he's going to buzz in every time on the category "Biographical Facts Concerning Jennings").
Re: IBM Watson beats human champs in first public test
Posted: Fri Feb 18, 2011 11:55 am
by ethanb
hyperpape wrote:That's true, but there's still some variable under the player's control here. There are questions where none of the players buzz in. They're obviously waiting for some information, at least in some categories (perhaps Jennings decides he's going to buzz in every time on the category "Biographical Facts Concerning Jennings").
IIRC if somebody buzzes in before the (human) person behind the switch decides the question has fairly been asked, they get a 3 second penalty before they can buzz in again. If they try again before the penalty is up, they get another second or two added. Maybe the man behind the curtain was a fraction slow activating the buzzers for a question and everybody's buzzer got locked down?
Actually, I think I remember reading that Watson has an unfair advantage there too; he senses the current when his buzzer is live.