I think the toughest part will getting the computer to understand the question being asked. Once that’s established, scanning huge amounts of data for the correct entry is something computers are great at. But understanding natural language will be difficult.
Would a wireless link to the internet to “Google” the answers be considered cheating? Id be curious to see that in action.
zoid, the computer will not be connected to the internet. I thought that it would have a huge advantage in researching, but without the net, the programmers will have to load every scrap of information it needs to answer, that’s a tall task.
I think it should have to stock itself with information before taping. Give it access to all previous Jeopardy shows and their answers/questions, glean what type of information it may need, connect it to the internet, and let it compile its own database in order to compete. That would be interesting.
One reasonable restriction that might be placed on the computer is that it can’t buzz in unless and until it has the answer ready. The time allotted for the person who’s buzzed in to answer the question is more than enough for any modern computer to search its database. It could very easily buzz in the instant the light goes on, then take its time to come up with the answer, which really puts the human contestants at a disadvantage.
Eh, all of the text in the english wikipedia is probably a few hundred gigs. Add a couple of similarily sized databases, and you’ve got all the information that’s likely to come up in 99% of Jeopardy questions on a few harddrives.
But people buzz in before they know the answer as well, anticipating they’ll be able to work out the answer in the allotted time. And even if you imposed the suggested restriction, it’s not that big of a handicap - this exercise isn’t to test the computer’s speed, it’s to test its ability to process the question.
Actually, thinking about this some more, IBM is probably much better off just searching a limited database like wikipedia then they are querying google. It’s faster, basically everything that’s likely to be in a Jeopardy question is already in there, everything on wikipedia is, at least in theory, factual, while there’s plenty on the internet at large that isn’t even supposed to be factual and the format is at least somewhat standardized.
Plus there’s less risk of the computer coming up with an answer from a spammer web page that has managed to game google’s pagerank algorithm to make it look important and offering to enlarge Trebek’s penis as an answer.
When I was on, I almost always knew the answers that I knew before Alex finished. After all, you can read them faster than he says them. This is even easier if the text version is presented at the same time the question appears on the screen.
There are two tricky things here. The first, as you said, is natural language processing. The second, however, is representing knowledge in a way it can be accessed quickly in unusual ways. For instance, the Final Jeopardy question when I was on hinged on you know who the co-stars of Miami Vice were, and that Don Johnson was on another show. Neither of us got it, but a computer making that connection would be very impressive.
would be simple for the computer, since it can find the answer by looking up the show’s director without know about Chris Reeve’s accident. For people, the accident part is the useful clue.
On the other hand
would be a bit tricky for a computer (but simple for people.) The category is Korean War, but would both the beginning and end of prisoner exchange be in the database? For us, just the word exchange gives the answer.
These are both $100 questions from the first round, and pretty easy.
This $400 poetry question
is also fairly easy for people, but searching through entire poems of all female poets might be a bit tough to do in 20 seconds.
On some questions (answers?) I think that the computer will definitely have an advantage – these are the ones that are totally fact-based. Examples would be words that fit a certain pattern and definition, historical events and figures, and the more obscure the fact the greater the advantage. I think that questions that are punny or very topical (real recent current events) may give the computer problems. In any case this is a very interesting experiment.
Maybe, but the nice thing about Jeopardy, is often the answers to all but the hardest questions are one of the few most famous members of the given category. For example, I guessed the answer to your question I quoted was Emily Dickenson even though I’ve never heard the poem, just because she’s the first person that pops into my head when I hear “female poet”. Similarily, the answers to questions about magicians tend to be “Harry Houdini”, famous celloists are usually “Yo-Yo Ma”.
So the computer doesn’t have to search through every poem ever written to find a match. If it starts with just the most famous hundred poets, I bet it will get a hit in 90% or more of cases. And if it doesn’t then it can move on to a larger and more complete database.
I’m working off two assumptions, one of which is wrong as your last sentence says, and the other may well be wrong:
If it’s being placed in a competitive environment, it should be set up so it provides competition to the other contestants.
Its search speed is equivalent or nearly so to a person’s reflex speed.
A computer’s “reflexes” are so fast that they may as well be instant. If it operated off the idea of “buzz first, rack brain second” then its reflexes would always beat out the humans’. Making it think first adds enough delay to make it competitive. Perhaps a conditional: if nobody buzzes in after 1 second or something, it buzzes in while still searching for the answer.
If it’s capable of getting the answer before Trebek finishes speaking, then sure, it can buzz in instantly. I wouldn’t want to hamstring it so the human contestants are able to best it, I’m just interested in keeping the playing field level. If it’s just a test of the processing ability, then don’t make it a competition with other players at all.
Pfft. Rather it sounds like an excuse to bring back Larissa Kelly. Not that they should ever need an excuse to bring her back, but by god, if you have one, better make use of it!