http://www.bizjournals.com/albuquerque/stories/2009/04/27/daily1.html
What do you think? I think that it’ll lose, as it won’t be very good at buzzing in on time against humans. I am very curious as to how it will bet on Daily Doubles and in Final Jeopardy.
I also wonder if they’ll bring in Jennings to compete against it.
Seems pretty easy, actually. Getting something to ring in a millisecond after the buzzer, and loading it up with a jillion facts are easy, anyways. Getting it to understand natural speech questions, not so much, but I’ve seen examples of such technology that work sort of well.
Plus, if its a hit, maybe they’ll spoof it on SNL’s celebrity Jepordy.
Buzzer? There’s no buzzer on Jeopardy.
It doesn’t even need to really understand natural speech questions. As long as it can pick out key terms, the rest can be pure gibberish. For example:
Alien Trebek: Blurble splatz florg “Through the Looking-Glass” murgle durgle bloop glug.
IBM: Who is Lewis Carroll?
It doesn’t hurt that Jeopardy questions have a tendency to give twice as much information as you need. To use the gibberish example above, it could’ve easily been something like: This “Through the Looking-Glass” author was born Charles Lutwidge Dodgson.
That being said, the article makes it sound like the program actually attempts to understand the questions. But I’d argue that such a technique is not an absolute requirement for a passable trivia answering bot.
If you want it to answer the question asked, and not just spout trivia about keywords, then it does need to be able to answer the question.
Will it have an OCR reader hooked up to the screen, a voice dictation module to pick up Alex’s voice, or both?
And - how long after you ring in and Alex cues you do you have to provide the correct answer?
No, but my understanding is that there is a light that indicates when it’s safe to ring in, and an audible indicator is used for visually-impaired players.
I think there’s actually a light that goes off to tell the contestants they can ring in.
Jeopardy questions seem both good and bad for this. On the good side, their somewhat formulaic, and phrased so that its pretty easy to pick out what they’re looking for: “this author…, this ship…, etc.”. On the downside, a lot of the clues have puns or jokes, which I imagine would be hell to try and program a computer to understand.
Yes, though a trivia spouter will sometimes (often?) hit on the right answers.
“Is the answer Japan?”
“Yes!” :smack::mad:
For usual questions, does the ‘safe light’ have any delay versus when the question first appears on the screen and Alex begins reading out loud?
They should take care that the program takes loosing reasonably well and doesn’t enter ‘destroy all humans’ mode when trying to win at all costs. On the other hand, a compulsion to answer all trivia questions might just be the flaw to exploit when trying to stop the inevitable army of killer robots.
I’m fairly certain they can’t ring in until after Alex has finished reading the question. Indeed, I suspect that’s the reason for locking out the buzzer till the light goes off, to keep contestants from cutting Alex off.
A few more details are in this New York Times article by John Markoff. The computer will get the questions in electronic text and may appear as an on-screen avatar. And they may invite Ken Jennings back to compete against it.
Heh, I tried grabbing some Jeopardy questions off an online database, changing them into the “form of a question” and putting them into MIT’s natural langage question answering machine. It was going 0 for 6 when I got bored.
This is a joke right? I don’t see how a human could possibly beat a computer at ringing in on an electronic system. That almost seems like a bigger issue (everyone having a fair chance to answer the question) than whether the computer can come up with the right answers or not.
I asked it “what is the meaning of life?” and it gave me the dictionary definition of the word. Cute.
Okay, I may have been thinking of another trivia game show, or long ago - where the contestants could ring in early, but then had to take their chances that they’d missed something by not hearing the full question.
That sort of depends on what the ‘ring in criterion’ is for the electronic system. If you ring in and cannot answer correctly, after all, you lose the money, so you’d want to program it to do some sort of a confidence check against the database, at least, before ringing in. And trying to get a sound confidence check like that to run faster than a human’s response time might get tricky.
Will it answer in the form of a question?
Next time, ask it the meaning of life, the universe, and everything.
Because I was under the impression that you rang in once Alex was done talking (which has been corrected). Getting a computer to recognize that (versus, say, any time Alex might pause during the middle of the question, ringing in between the time Alex is done and the time that the buzzers are enabled, etc.) I think would prove difficult.
The computer won’t even make it through the first round before John Connor shows up to destroy it.