Jeopardy! Champs vs. Computer

He was a contestant when you got 5 days, then left, aside from any tourneys you were invited to.

He won his five games, then he won all the tournaments. (3 of them.)

The Toronto answer that amuses so many people strikes me as similar to one of Watson’s suggestions on the question about an art theft from a museum from “this city”, the one in Spain, not in Ohio.
The correct answer was Toledo, a major city in Ohio, but Watson preferred the selection of Madrid.
The thing is, there is a New Madrid, OH, though it’s a comparatively minor city. And Madrid was certainly a reasonable guess on the Spain side.

Similarly, there is a Toronto IL in the US. There may be other cities in the US by that name that I am unaware of.

This may be an example where having too many facts in ones knowledge is actually a disadvantage. It requires Watson to recognize that two similarly named entities are actually distinct (often a tricky matter in AI programming) and then to link all of its individual facts about Toronto/Madrid to the appropriate instance. If Watson were blissfully unaware of the very existence of the less-famous namesakes, it would not have faced this difficulty.

Thanks! That was really fascinating as well as entertaining. I see Jennings thought a lot of the same things my wife and I did, down to the skynet reference. (Like Tangent, we expected one of the Connor’s to come in blasting, though we had our money on Sarah.)

Watching these provoked a lot of thought and discussion afterward. You’re never going to get a 100% fair competition (like Jennings says in that link). Once you get over the (incredibly high) hump of getting the thing to work, the reaction time aspect has to be either ignored or handicapped. neither way is really fair.

Scary thought: The room-sized computers of the 60’s weren’t as powerful as todays ipods. How long until Watson’s room-sized HW fits in your pocket and costs under $100.

Scary thought #2: Once Watson-like computers can make complex decisions like medical diagnosis like the IBM guys touted, what are humans needed for? (other than finding Toronto on a map. :slight_smile: )

Comforting/amazing thought: With a room full of HW, Watson can do really really well at this ONE aspect of human thought: playing Jeopardy! He can’t understand spoken speech, can’t tell that a picture of a horse is a horse, can’t write a symphony, or control a pair of legs to walk, can’t make a mathematical proof, can’t name that tune, can’t answer the question “how do you feel?”, can’t catch a ball, find themes in literature, make a joke, read, negotiate, or cook… and we can do all that and more with 3 pounds of goo.

Of course, it’s only a matter of time before Watson, a speech-recognition program, an OCR program, a computer vision program, and most of the others can sit together in an Asimo body. :slight_smile:

Apparently the number of question marks gave some indication that it wasn’t very confident about the answer - Watson may not normally have buzzed in on that, but was compelled to because it was FJ. And again - straightfoward to you and me means something else entirely in this context.

[Hijack]

Going into this week, Rutter is "undefeated’'. But is he truly undefeated? Theoretically he could have lost one of the quarterfinal matches, and advanced to the semis via the back door. The Top 4 money winners that didn’t win their Qtr final match advance to to the semi finals. And he could have lost either day of the two day final. So I am just curious if the first time he lost a game was his first match against Watson.

[/End Hijack]

Back to the thread, apparently Alex, Ken, and Brad changed clothes between the Jeopardy and Double Jeopardy rounds.

I don’t think anyone has pointed this out about searching for the Daily Double clue early in the Jeopardy round which Watson did. It is important to realize that if you find it early, it means that your opponent cannot find it late. The first priority is to win, not to maximize your total. and a viable strategy for Watson is to take away the randomness of the Daily Double.

Per Wikipedia’s Jeopardy! footnote, “those who ring in too early are penalized 250 milliseconds (1/4 second) each time they jump the gun.” So I propose replacing Ken with a Watson clone, for 2 computers vs. Brad, giving the computers’ programmers the additional challenge of deciding when exactly to ring in, because we’ve already seen that a machine is quicker than the best humans at ringing in in the first possible millisecond.

Then the programmers can fix the problems like the list of cities of a country. And for vaguer trivia, like gymnasts of 107 years ago, Watson could first give a vaguer response, like “missing a limb”, before deciding–if still needed–on an appropriate limb.

Each of the first 2 programs looked to me like 10 minutes of commercials wrapped around a 10-minute IBM infomercial, between trivia that the audience knows 99% of the correct questions for. I’d be more entertained by Watson sitting in for a US Supreme Court Justice, or the next State of the Union address. And yet, I still prefer to watch Jeopardy!

It is about response time. Most of the questions were not that hard and I am sure the humans knew the answers.

It probably did, but for some reason, it didn’t have a viable alternative that matched better, and it had to answer. So if his search algorithms provided him with three possible answers, ‘Toronto’, ‘beef jerky’, and ‘I am the Walrus’, he’s gonna go with the one that matches the most criteria, even though it’s not a good match.

First of all, my very first thought on seeing Watson’s answer was that though Toronto is a well known city in Canada, there are also American cities named Toronto, beyond others things that would have confused the computer. Here is a statement from an IBM programmer, from A Smarter Planet Blog. Ellipses are, obviously, mine.

I just wonder if the person inputting the Final Jeopardy category put in “US Cities” (Us Cities) as opposed to “U.S. Cities”.

My main question, after watching the first two episodes, is how does Watson determine wager amounts?

Anytime you access Google, you are talking about a complex with millions, not thousands of processors, so anyone with a smart phone has access to a lot more processing power and information than what Watson has.

I bet Google is watching Watson very closely, since this really cuts to the heart of their business. Ditto for Oracle. This Jeopardy thing is a publicity stunt, but IBM has fired a shot over the bow of everyone in the database business.

True. Cloud-based computing does a lot to make actual size moot. But cost is always a factor. How will things change when there are millions of Watsons instead of one?

That interview made me really like Ken Jennings.

I was hoping for some discussion about the Final Jeopardy ‘What is Toronto???’ answer. Alex quipped about Toronto being in the US but no discussion.

Rather than the IBM infomercial, I would have like to see seen discussion on why Watson made the mistakes it did. Again today, it came it with a couple of wrong answers that made me go WTF? That answer doesn’t match the category.

I did note on one of the questions that Ken go correct, Watson’s percent on the top answer went from below 50% to above 60% while Ken was answering.

It uses game theory. See here

I Guess the final outcome was no surprise but Watson took a very big risk on final Jeopardy. He could have won by betting 0. Instead he bet $17k.

He’s actually very funny. I’m going to have to buy his book.

No, Ken Jennings had over $17,000. If Watson had bet $0, he could have lost.

He could have won the 2 day total with a 0 bet. The 2 day total is all that matters.