Why Ken Jennings’s ‘Jeopardy’ Streak Is Nearly Impossible to Break
fivethirtyeight.com
fivethirtyeight.com
This is just a binomial distribution. Did they seriously estimate the answer instead of solving it analytically?
I'm gonna guess that they did solve it analytically, but for the sake of exposition and readers that might not know what a binomial distribution, they bring up the coin-flip as an illustrative example.
I played on a trivia team at the high school level, and most questions were structured like:
"This American architect, born in 1867, is known for works such as the Robie House, the Guggenheim Museum, and Fallingwater."
So they typically feed you information from least to most well known, and you can buzz in at any time. This rewards players who know the most trivia. Everyone who played knows Frank Lloyd Wright did Fallingwater, but a very good player would know his approximate date of birth and would buzz in immediately on hearing 1867.
I understand that this might make Jeopardy less fun for the audience, but I feel it would be a better test of trivia instead of buzzer timing.
With the wait-to-buzz-in rule, the audience gets an chance to understand and maybe think of the answer before the contestants speak.
Which is lots of fun and very competitive, but a little bit less tele-savvy.
Q: What is the capital of L-(BUZZ)-ithu-- [stops reading].
A. Vilnius.
But if they had had really good reaction time and stopped on L--, I'd have been hosed. Latvia, Lithuania, Liechtenstein, Libya, Liberia, etc...
Q: What is the capital of (BUZZ) Lithuania?
A: Vilnius.
One solution would be to display the text of the question slightly beforehand to the audience, but not the contestants.
I think one of the reasons Jeopardy! is so popular is that it makes the audience feel smart. I'm not amazing at trivia, but I know most of the Jeopardy! answers and having the contestants wait to buzz in creates the illusion that I'm doing as well / better than they are. Even though I'm intellectual aware of the importance of buzzer timing, it still feels good to watch and give the answers before the contestants have a chance to demonstrate they know. Compare that to a trivia show like University Challenge in the UK, where not only are the questions remarkably hard, the contestants buzz in before I've had a chance to parse, let alone answer, the questions.
I think allowing contestants to buzz in early, even if Alex finished reading the question before they could answer, would make Jeopardy! a better trivia game but a worse TV show.
Being first to buzz is a huge part of the game, and no matter when you turn the buzzers on, there will be people trying to buzz at that moment.
My wife, and my neighbor's husband, feel like it's the height of rudeness.
This is kind of the canonical example people can understand for the platinum vs the golden rule, as it happens.
There's another side, too, which is if folks could signal early, reading fast would become an even stronger optimization. It already is, but at least everybody gets an equal chance the way it is (even if they got the answer fairly quickly from reading the clue while ignoring Trebek).
On the other hand, I wouldn't be surprised if the target demographic of Jeopardy is often doing something else and listening without actually watching.
Other game shows work that way as well, and it has its own degree of fun for the audience. For instance, the contestant buzzes in and answers after a few words of the question, gets the answer wrong, the host is grinning like a possum, and after they read the next part of the question, it makes the contestant's answer even more hilariously wrong in context.
[1]: This has always struck me as an idiotic gimmick, especially since the wording rarely makes sense. No one would ever respond to the question "Who was George Washington?" with the answer "This man served as the first President of the United States."
[2]: https://en.m.wikipedia.org/wiki/Jeopardy!#CITEREFTrebekBarso...
I actually played Watson on the Jeopardy set they filmed the Watson episodes on. That "set" was just the auditorium here at IBM Research in NY, and they left the set intact for about six months after taping. Colleagues of mine hosted an academic conference in our building during that time.
Since Watson was on everyone's mind at that point (and we were literally looking at the set for it during people's talks), my colleagues asked the head systems researcher if he could give a special talk at the conference. He graciously accepted, and offered to do a full demonstration. There was only one volunteer from the rest of the conference - I think that if you're not a native English speaker, and not familiar with American pop culture and history, Jeopardy questions can be mystifying - so I gladly volunteered to play.
To answer the obvious question: I lost. Badly. In fact, not only did I lose, I lost a full-cheating version of Jeopardy where the audience (participants from the conference) shouted out answers, and I buzzed in using answers I liked. Not only that, but some of the answers that the audience was shouting were taken from Watson, looking at the readout it showed of candidate answers. (Some of the people were able to immediately recognize the correct answer, even though Watson was unsure of it. But keep in mind Watson came up with those answers.)
I would have lost to even a decent Jeopardy player, so losing to Watson is kind of overkill. But I think I learned a few things about playing Jeopardy, and how the human experience differs from Watson's experience.
I spent almost all of my cycles when the host was asking the question trying to figure out when he would stop talking. That is, I spent the entire question-asking period timing my buzz. I could not think of the answer and time my buzz simultaneously. The few times I thought I did know the answer, I did not know the answer when I buzzed in. I knew I knew the answer. The difference is that I did not have the answer in mind when buzzing, just that given a few seconds - which you get, by the basic protocols of the game - I could come up with the answer. I think other human players are the same: they spend the question-asking period timing their buzz, and are going by an intuitive feel for whether or not they know the answer. The actual answer recall does not happen until after they buzz in.
Watson is inverted. It only buzzes in when it is confident of its answer. In that sense, the game is not just about Watson winning the buzzer timing. Watson definitely could win every buzzer, but then it would lose the game; it's not accurate enough to get every question correct. So if Watson won every buzzer, it would answer incorrectly many times, and ultimately lose.
With that in mind, I hope it restores some of the romance. During the test run, the category was "Presidential Rhymes" with the clue "Barack's pack animals". Watson answered "Obama's llama's" immediately. Me and the other contestant just stared at each other, slack-jawed. I had seen it done on tv, but there was something far more immediate and impressive, since the head systems designer was ten feet from me, and the Power7 systems just a few rooms away.
Can you expand on why it was so shocking or maybe impressive by the Obama response. I'm confused what the significance of it is.
We were just shocked because it answered before we could even think the question through, and it happened in front of us. It's an irrational thing, but being physically present for the act makes it more impressive.
Can you comment on further Watson developments since that time?
They also did some other tricks, like giving Watson a parsed ASCII stream of the clue, rather than making it see the screen and parse it out. I think it also got an electronic feed of the "you can buzz in now signal" rather than making it parse out the same light the contestants see.
Also, the ASCII feed should have matched human reading speed.
So the computer's most effective weapon against Jennings was a 555 timer? That's cute.
This is incorrect and overstates his dominance because (1) it assumes Jennings will be ahead, something that won't always happen against Rutter/Jennings/Collins/Chu/Craig level players and (2) the lock up and opponent knowing Final Jeopardy conditions are not independent.
Pretty sloppy basic probability by 538 and make me think I'm not missing much by not subscribing (they didn't offer full RSS last I checked).
Just checking my understanding.
A. He cheated B. He's lucky C. He's a genius D. It's written
There may be a diversity in palette for this matter, even among the same people at different times.