What about when there are agents with difference source codes and different preferences?
Set aside different preferences, because I have no idea how to deal with that. It seems like just having different source codes makes things very difficult.
You used “1” and “2″ as your indexical labels, which come with a very obvious order. But suppose that Omega labeled one agent ♠ and the other agent ♣. If you and your opposite have different source codes, you have no idea how your opposite internally represents the symbol that they received. For example, you certainly have no guarantee that the two of you order these in lexicographically the same way. So how could you possibly coordinate on a tie-breaking strategy?
Set aside different preferences, because I have no idea how to deal with that. It seems like just having different source codes makes things very difficult.
Where do you think preference lives? It’s but a property of the source code, or rather, a way of looking at the source code. If you change the source code, preference changes as well (to some extent).
Where do you think preference lives? It’s but a property of the source code, or rather, a way of looking at the source code.
I agree that preferences live in the source code. Having different preferences implies having different source codes. I also agree that preferences are but a way to interpret source code, so that any two source codes might be said to have the same preferences with respect to some preference scheme.
So my comment was assuming that we already have an implicit way of mapping source codes to preference schemes, and that that mapping is not injective. That is, different source codes might have the same preference scheme. But these different source codes might still use different internal representation schemes for representing their input.
I think that my issue disappears if each agent has access to the other’s source code.
I also agree that preferences are but a way to interpret source code, so that any two source codes might be said to have the same preferences with respect to some preference scheme.
I don’t understand what the above means (and so can’t readily be in agreement with it). (“Preference scheme”?)
So my comment was assuming that we already have an implicit way of mapping source codes to preference schemes, and that that mapping is not injective.
I expect preference mapping to be “essentially” injective (possibly disregarding only stuff that doesn’t play any role in the algorithm, including the way it computes as well as the externally visible behavior). Preference is what the program does, and it’s everything that the program does. (But what the program does isn’t necessarily preferable according to that same program seen as preference.)
I don’t understand what the above means (and so can’t readily be in agreement with it). (“Preference scheme”?)
At this date almost four years later, I’m not exactly sure what I meant, which is a pretty strong indictment of my ability to write clearly.
I think that I meant the following: Take the set of all possible outcomes, partition this set of outcomes into equivalence classes, and then put an order on the resulting set of equivalence classes. The ordered set of equivalence classes is what I was calling a “preference scheme”. (… I think.)
ETA: On further thought, I don’t think that that’s exactly what I meant, but I honestly don’t know what I did mean. I’d like to think that I could have explained myself better at the time, but I didn’t write enough so that even I-now could reconstruct what that explanation would have been.
Huh, that surprises me. I was going purely by personal experience, plus the fact that bidding goes from low to high in Bridge; that’s the order I would expect people to list the suits in, because that’s the order they occur in.
Oh, I see the problem. I was talking about Tyrrell_McAllister’s question upthread, in which the assumption of identical source code (i.e. copying) is dropped.
If you don’t know much about the other agent—except that it is also trying to win—I figure you should also probably just do the best you can to pick the most mutually-obvious ordering, hoping that they will be doing much the same. Sometimes, it won’t work out—but that is doing as well as you can.
That’s assuming linear utility. If the most important thing is to consistently get at least a few points, then randomness may be a better strategy.
So how could you possibly coordinate on a tie-breaking strategy?
I’d use alphabetical ordering, and assume my opposite would too. So clubs would come before spades, black before white, etc.
Now, if your opponent doesn’t use the same language (or doesn’t use a language, only unicode symbols), it almost becomes a game of chance, like “here’s an aarvark and a wapiti, the same animals have been shown to some (non-english speaking) Chinese guy, you each get to choose one, if the two of you choose different ones you get a hundred bucks”.
For example, you certainly have no guarantee that the two of you order these in lexicographically the same way. So how could you possibly coordinate on a tie-breaking strategy?
You can’t, unless you can communicate in some way, such as by talking to eachother or reading eachother’s source code. In that case it’s easy.
Set aside different preferences, because I have no idea how to deal with that. It seems like just having different source codes makes things very difficult.
You used “1” and “2″ as your indexical labels, which come with a very obvious order. But suppose that Omega labeled one agent ♠ and the other agent ♣. If you and your opposite have different source codes, you have no idea how your opposite internally represents the symbol that they received. For example, you certainly have no guarantee that the two of you order these in lexicographically the same way. So how could you possibly coordinate on a tie-breaking strategy?
If you are incapable of consistently putting things into order your winnings on this type of problem go down.
Resource-limited agents should not stress too much over this—such problems are mainly of interest to philosophers.
Where do you think preference lives? It’s but a property of the source code, or rather, a way of looking at the source code. If you change the source code, preference changes as well (to some extent).
I agree that preferences live in the source code. Having different preferences implies having different source codes. I also agree that preferences are but a way to interpret source code, so that any two source codes might be said to have the same preferences with respect to some preference scheme.
So my comment was assuming that we already have an implicit way of mapping source codes to preference schemes, and that that mapping is not injective. That is, different source codes might have the same preference scheme. But these different source codes might still use different internal representation schemes for representing their input.
I think that my issue disappears if each agent has access to the other’s source code.
I don’t understand what the above means (and so can’t readily be in agreement with it). (“Preference scheme”?)
I expect preference mapping to be “essentially” injective (possibly disregarding only stuff that doesn’t play any role in the algorithm, including the way it computes as well as the externally visible behavior). Preference is what the program does, and it’s everything that the program does. (But what the program does isn’t necessarily preferable according to that same program seen as preference.)
At this date almost four years later, I’m not exactly sure what I meant, which is a pretty strong indictment of my ability to write clearly.
I think that I meant the following: Take the set of all possible outcomes, partition this set of outcomes into equivalence classes, and then put an order on the resulting set of equivalence classes. The ordered set of equivalence classes is what I was calling a “preference scheme”. (… I think.)
ETA: On further thought, I don’t think that that’s exactly what I meant, but I honestly don’t know what I did mean. I’d like to think that I could have explained myself better at the time, but I didn’t write enough so that even I-now could reconstruct what that explanation would have been.
Conventionally, spades comes before clubs:
Bridge: spades, hearts, diamonds, clubs;
Poker (generally) spades, hearts, clubs, diamonds.
Except that generally you list them going up—clubs, diamonds, hearts, spades.
Reference? They are not generally listed that way on the internet:
Google “spades, hearts, diamonds, clubs” − 4,090
Google “clubs, diamonds, hearts, spades” − 2,950
Huh, that surprises me. I was going purely by personal experience, plus the fact that bidding goes from low to high in Bridge; that’s the order I would expect people to list the suits in, because that’s the order they occur in.
Those numbers are not hugely different—it might be more accurate to say that “there is no reliable consensus on order”.
To deal with this kind of problem, you want the best ordering you can find.
Not everyone has to agree on it.
The other agent, who has different source code to you, has to agree on it. If it were you and Sniffnoy playing the game …
In the post, it said:
“Suppose Omega appears and tells you that you have just been copied”.
Oh, I see the problem. I was talking about Tyrrell_McAllister’s question upthread, in which the assumption of identical source code (i.e. copying) is dropped.
If you don’t know much about the other agent—except that it is also trying to win—I figure you should also probably just do the best you can to pick the most mutually-obvious ordering, hoping that they will be doing much the same. Sometimes, it won’t work out—but that is doing as well as you can.
That’s assuming linear utility. If the most important thing is to consistently get at least a few points, then randomness may be a better strategy.
I’d use alphabetical ordering, and assume my opposite would too. So clubs would come before spades, black before white, etc.
Now, if your opponent doesn’t use the same language (or doesn’t use a language, only unicode symbols), it almost becomes a game of chance, like “here’s an aarvark and a wapiti, the same animals have been shown to some (non-english speaking) Chinese guy, you each get to choose one, if the two of you choose different ones you get a hundred bucks”.
You can’t, unless you can communicate in some way, such as by talking to eachother or reading eachother’s source code. In that case it’s easy.