Related to what I said here (https://www.lesswrong.com/posts/G2SQzLnoBnmYhPygE/divia-s-shortform?commentId=wSce6sGx9LLqQGAQ4), I also have a general beef with game theory typically being a misleading abstraction (outside of the context of certain formal games (when they are played by economists??)). I think the most common objection I’ve heard to game theory as a paradigm is that it’s much more common for “games” to be iterated than one-off, but I think maybe my bigger complaint is that it’s quite rare for stuff to be anywhere near as turn-based as the paradigm implies.
That said, I don’t have this sort of objection to decision theory as a paradigm—I basically just like it—and it at least superficially it shares the turn-taking abstraction that I’m objecting to, so I’ll keep trying to clarify what’s up for me.
Something that’s come up for me since our last chat was (starting to) read Elinor Ostrom’s Governance of the Commons, which leans very heavily into “the actual games people are actually playing have very little in common with the simplified games people use as their core metaphors.” Writing a good book review of that is on my list of things to do.
(Elinor’s book is basically a very long, methodical rant at all the people either saying “Tragedy of the commons, therefore, Government Ownership Of Things” or “Tragedy of the commons, therefore, private property rights.”)
And I guess also in real-life situations the menu of options is typically really large instead of small? I don’t actually know if that’s central.
As much as I would like to already have a clean objection, it’s probably more fruitful at this point for me to generally poke around and try to articulate whatever in the general space I have some traction on.
For me, turn-taking isn’t a problem—game theory addresses simultaneous turns as well (including the famous prisoners’ dilemma). My objection is the massive oversimplification into one-dimensional known (or known probability) payouts. Almost every real interaction has highly-variable and diverse impact on both short-term outcome and positioning for future interactions.
it’s much more common for “games” to be iterated than one-off, but I think maybe my bigger complaint is that it’s quite rare for stuff to be anywhere near as turn-based as the paradigm implies.
Iterated games are very common in game theory (especially in LW discussions), and even the most basic games (like prisoner’s dilemma) don’t tend to be turn based. I think you’re critiquing a naive version of game theory, that would indeed deserve your critique if it was the norm, but I don’t think it is. It probably has been at the start, but now I think more realistic assumptions are the norm—like iteration, reputation, mistakes, temporal discounting, diminishing returns, limited information, etc.
As an example, thesethreeposts show some of this complexity applied to prisoner’s dilemma
As another example, this post of mine explains a theory based on a game theoretic model (or “game”), that’s far more complex than the basic games like prisoner’s dilemma. It has 5 types of players, their utility is based on 4 different things, they need to make decisions about multiple things, they need to take future states into account, they have temporal discounting and diminishing returns—and all that gives something that makes lots of very good predictions about politics.
I don’t “want you to acknowledge” anything. You mentioned a few things that bother you with game theory, I responded saying these things are addressed. If that’s interesting/helpful, great, and if not, that’s fine too.
That games like prisoner’s dilemma, coordination games, and other basic games in game theory aren’t turn based, the players makes their decisions simultaneously. The table in the link has a column for whether the game is sequential or not.
I think that’s still “a turn” in some sense. Things still happen in discrete steps (i.e. people make their decision, then reveal their decision), instead of a continuous back-and-forth.
Related to what I said here (https://www.lesswrong.com/posts/G2SQzLnoBnmYhPygE/divia-s-shortform?commentId=wSce6sGx9LLqQGAQ4), I also have a general beef with game theory typically being a misleading abstraction (outside of the context of certain formal games (when they are played by economists??)). I think the most common objection I’ve heard to game theory as a paradigm is that it’s much more common for “games” to be iterated than one-off, but I think maybe my bigger complaint is that it’s quite rare for stuff to be anywhere near as turn-based as the paradigm implies.
That said, I don’t have this sort of objection to decision theory as a paradigm—I basically just like it—and it at least superficially it shares the turn-taking abstraction that I’m objecting to, so I’ll keep trying to clarify what’s up for me.
Something that’s come up for me since our last chat was (starting to) read Elinor Ostrom’s Governance of the Commons, which leans very heavily into “the actual games people are actually playing have very little in common with the simplified games people use as their core metaphors.” Writing a good book review of that is on my list of things to do.
(Elinor’s book is basically a very long, methodical rant at all the people either saying “Tragedy of the commons, therefore, Government Ownership Of Things” or “Tragedy of the commons, therefore, private property rights.”)
And I guess also in real-life situations the menu of options is typically really large instead of small? I don’t actually know if that’s central.
As much as I would like to already have a clean objection, it’s probably more fruitful at this point for me to generally poke around and try to articulate whatever in the general space I have some traction on.
For me, turn-taking isn’t a problem—game theory addresses simultaneous turns as well (including the famous prisoners’ dilemma). My objection is the massive oversimplification into one-dimensional known (or known probability) payouts. Almost every real interaction has highly-variable and diverse impact on both short-term outcome and positioning for future interactions.
Iterated games are very common in game theory (especially in LW discussions), and even the most basic games (like prisoner’s dilemma) don’t tend to be turn based. I think you’re critiquing a naive version of game theory, that would indeed deserve your critique if it was the norm, but I don’t think it is. It probably has been at the start, but now I think more realistic assumptions are the norm—like iteration, reputation, mistakes, temporal discounting, diminishing returns, limited information, etc.
As an example, these three posts show some of this complexity applied to prisoner’s dilemma
As another example, this post of mine explains a theory based on a game theoretic model (or “game”), that’s far more complex than the basic games like prisoner’s dilemma. It has 5 types of players, their utility is based on 4 different things, they need to make decisions about multiple things, they need to take future states into account, they have temporal discounting and diminishing returns—and all that gives something that makes lots of very good predictions about politics.
Seems like you want me to acknowledge that various people, including on LW, have been doing pretty sophisticated stuff with game theory lately?
I don’t “want you to acknowledge” anything. You mentioned a few things that bother you with game theory, I responded saying these things are addressed. If that’s interesting/helpful, great, and if not, that’s fine too.
What do you mean by “even the most basic games [...] don’t tend to be turn based”?
That games like prisoner’s dilemma, coordination games, and other basic games in game theory aren’t turn based, the players makes their decisions simultaneously. The table in the link has a column for whether the game is sequential or not.
I think that’s still “a turn” in some sense. Things still happen in discrete steps (i.e. people make their decision, then reveal their decision), instead of a continuous back-and-forth.