Secret Goals

[flwyd]

One of the basic distinctions in game theory is between games with perfect information, like chess or tic-tac-toe, and games with hidden information, like poker (each player's cards and the deck) or Monopoly (the dice). But I was this evening thinking about games with not only hidden information, but hidden goals like Aquarius (hidden goal card).

Does anyone know about game theory thinking about hidden goals? In some cases (like Aquarius) it may be easy to treat it as ordinary hidden information. But in other situations (like politics), it may change the understanding quite radically. It seems like it would be very hard to develop a predictive model of a player's actions if you don't know what he's going for (e.g., somewhere in emotion-money-ideals space).

Comments

[vvvexation.livejournal.com]

If you don't know someone's goals, then the only way to predict their actions is by inductive rather than deductive reasoning, right? Which is indeed slower and likely more error-prone.

[flwyd.livejournal.com]

I was thinking of a situation where you might know what sorts of goals someone might have, but not in any detail. rubicantoto's example of Chrononauts is good: you know their goal involves collecting a particular set of three cards, but you don't know which three. And as the game goes on, you may be able to guess based on their actions thus far.

In a business or political situation, I was thinking of something like multi-variable bargaining where each side doesn't disclose the relative importance of each variable.

[rubicantoto.livejournal.com]

Or (to sidetrack a bit) in a personal situation, like dating.

I'm being somewhat cynical here, but also sort of serious. I think the vast majority of people have some kind of calculus for determining when to drop bombs of various types, but not until the n'th date. It's common advice on Savage Love, for example, to bring up most fringe kinks later and later.

In general, this is a very practical treatment of it as "ordinary hidden information". The most-specific/lowest-weighted behavior-affecting variables are the last ones you need to track to build an accurate model.

[vvvexation.livejournal.com]

Huh. That last actually reminds me of Deborah Tannen's The Argument Culture, which I've just been reading -- specifically the part about how, especially (perhaps) in this country, we've developed this prevailing notion that in business or legal negotiations you have to demand more than you really want just to have any hope of getting anything, and that just makes it much harder for people to come to a mutually satisfactory arrangement even when one does theoretically exist.