Computer to play Jeopardy against past champions

[Lorizael]

As you said, you pulled some numbers out of your ass, making it your opinion. My point is that judging whether it's worth any risk to upgrade your chance of winning from 90 to 95 is still a judgement call. Or even from 50 to 55. The programmers provide an algorithm for the computer to weigh it. In my opinion the machine was too conservative and I would have programmed it to be a touch more aggressive. It's a judgement call based on what I observed. I would be willing to bet that generally you're more conservative than I am and that helps to feed our impressions of us watching the same event.

It's not a judgment call. It has to do with which scenario is more likely to produce victory. For some reason, you seem to think being aggressive there is more likely to produce victory, except that you don't have any statistics to back up that assertion. Watson, on the other hand, has all the statistics.

[rah]

No, the programmers provided the algorithms for Watson do use to do the calculations and the parameters for making decisions. The computer is following instructions.

[Lorizael]

Christ, rah, it's like you've never heard of importing data.

[Wezil]

Who are 3 people that have never been in my kitchen?

22000 big ones.

Cliff Clavin.

Cliff was robbed.

[rah]

Christ, rah, it's like you've never heard of importing data.

Christ, it's like you've never heard of programming. Computers don't think for themselves yet. They're only as good as their programmers. DUH

[rah]

example. You saw the projections for it's top three responses I assume. Now how do you think it decided whether to buzz in a normal question. I couldn't tell exactly but I never saw it buzz in when it's top answer was less then 50% confidence. Some programmer gave it a threshold. If they said 90% that would be kind of wussy. If they said 80% that would be more aggressive. Now I'm sure they did quite a few simulations to compare the impact of the confidence level on the final score and game circumstances but a threshold is set none the less. It probably even varies based on many factors. But the fact remains that a programmer influenced it. And by tweaking it could make Watson appear more aggressive or more passive.

[Lorizael]

What you're missing here is it's not the programmers that are running the simulations but Watson itself. Watson has access to past Jeopardy results, its own record, and the ability to run real-time simulations. Based on this data, it is able to calculate probabilities of success and make the decision most likely to produce victory.

[Lorizael]

rah, this isn't a traditional game show situation where you're trying to win the most money possible. Watson is simply trying to win, period. If you're trying to win as much money as possible, then your level of risk aversion matters when it comes to the correct decision. But if you're just trying to win, then risk only matters in so far as it affects one thing: the probability of winning. And because there is only that one variable, there are objectively correct and incorrect answers as it concerns achieving that goal.

[rah]

When the game isn't over yet. the amount of money still matters. The best way to win is to have so much money that another playing getting a few timely daily doubles can't catch you. (even though watson was controlling the board so well, the humans getting them was a low probability. But the probability wasn't 0)

And yes, I know Watson is running the simulation but it's constrained by the code. Watson did not write the code, programmers did. The code influences decision making. Yes, Watson is data mining to produce answers and projections but the code is the decision maker in the end.

[Wezil]

I think the other players should cheat.

[Asher]

Anyone who has played Civ knows the computer is quite excellent at cheating.

[Wezil]

Maybe we should follow the money?

Has anyone looked at Trebecks bank statements lately? I suspect a payoff.

[rah]

Anyone that has ever played CIV V know that ****ty computer programming leads to ****ty AI decision making.
I have been quite impressed with Watson though. But then they probably put 10000x the amount/dollars of effort into it.

[Lorizael]

When the game isn't over yet. the amount of money still matters. The best way to win is to have so much money that another playing getting a few timely daily doubles can't catch you.

No, the best way to win is to aim to have $1 more than twice your nearest competitor come the final Jeopardy round. Every single decision you make should be geared toward that end - nothing more, nothing less. Anything in excess of that amount is a waste and represents unnecessary risk.

[Jon Miller]

See, that is why a person is different than a computer.

A person doesn't just want to win, they want to win a lot of money.

JM