At 06:10 PM 9/2/96 -0500, correspondent wrote:

If there are more than one accurate prediction for an even, how will CP Server will allocate the prizes? What made me think of that is that, only in the case of famous peoples, a very smart individual might make a prediction simply out of being smart and informed.

To a first approximation, they could split the reward among the correct guessers. This is an issue that I considered in some detail a year ago, but which (surprisingly enough) hasn't been raised by others. As I pointed out in AP part 1, it is necessary to reduce "shotgun guesses" among people who simply guess a date (or many dates) and make a bet. That's why I included the system of requiring the predictor to include a payment with the prediction, in such a way that the prediction remains anonymous while the included money is always paid to the AP organization. This might initially appear to be a burden to the potential "guesser"/killer, but in fact if he understands the reason for it he'll want the system in place: It's one of the main ways to avoid the situation where multiple people make guesses for a particular person and date, and thus it would tend to ensure that the successful "guessor" is the only one to make that guess. Some of my analysis in AP part 1 was superficial and not really accurate, because in the example it gave, it suggested that the amount of money that would have to be included might be as little as 1/1000th of the reward money. However, chances are it'll usually be somewhat higher for this reason: Let's say a person does something REALLY AWFUL, such as Jeffrey Dahmer or another multiple murderer. Suddenly, he goes from completely anonymous to Public Enemy #1. One would suppose that if AP was operating "efficiently," he'd be dead in a week. However, if indeed he's dead in a week, and you only have to pay 1/1000th per day to place your bet on his demise, you could (hypothetically) make your bet for all seven days, paying a total of 7/1000th of the prize, and assuming the death occurred any day that week you'd be guaranteed to share in that prize. But that's the problem: You'd only _share_ in that prize, because many other people would get the same idea and they'd make similar bets, and the thing would simply turn into a Lotto-type game. The potential killer out there, aware of this problem, would hesitate to make his bet under those circumstances, because it is almost certain he'd lose at least part of the prize to others. The result would be a great deal of suspense, because nobody would know when somebody is actually going to place a prediction and carry out a killing. The AP organization would rake in the money from all those bets, but the killing would be delayed and the betting public would become unhappy. At some point the "prediction" donations would slow down, and perhaps a killer would take this as a cue to actually make his prediction. One way to avoid this is to carefully adjust the amount of payment that's required with the "prediction," raising it _just_enough_ to deter all but "informed guessing" among people who know what's going to happen. Or, at least, to reduce "uninformed guessing" to a level which doesn't dramatically affect the fulfillment of the donations. The problem with this is that this price-setting would be a lot of work, and is not likely to produce the "right" price. It would be somewhat akin to the kind of central planning that the communists never did very well. The problem with trying to set a price like this is that to do it right, requires WAY too much knowledge, knowledge that will often only be known by a small number of people that you can't identify. Perhaps the most obvious solution is to allow the free market to decide how much a given prediction is worth. In other words, the Invisible Hand of Adam Smith. Instead of asking for some specific amount of money along with any given prediction, simply announce that along with a prediction the predictor ought to include some portion of digital cash, although there would be no minimum amount required. Assuming the associated prediction turns out to be true, the reward fund will be distributed on a pro-rated basis, divided up based on the amount of prediction. For example, if you're the killer and you include a dollar with your prediction, and I'm a random, guessing predictor and I include a dime, you get 1/1.1 of the reward and I get 0.1/1.1. Had you included $10 with your prediction, you'd have won 10.00/10.10, and I'd get 0.1/10.10. Sure, the amount a predictor included might be as little as a dime, and if that's the only correct prediction he'll get the entire amount of the reward. But a killer would be stupid to ONLY include a dime, because somebody else could, likewise, include a dime per day for a prediction for a given person, and then he'd get half of the reward if nobody else did the same thing. And since it would only cost him $0.10 per day or $36.50 per year for a given person, he'd be dollars ahead to do this. It should be clear that a person who really KNEW that the target would die on a particular day would want to include enough digital cash to help ensure that he's the beneficiary of a good fraction of the reward, ideally most or all of it. On the other hand, he won't want to include so much that it's "too much" a proportion of the reward itself, since the payment is non-refundable and it reduces his net reward. The random guesser likewise wants to maximize his share, but unlike the killer does not have the specific knowledge that the death will occur on that particular day. With this system, the market is responsible for finding its own equilibrium point. The AP organization need not decide how likely a given death is, and how much money to ask for. Its job is made substantially simpler. Jim Bell jimbell@pacifier.com