It sounds like you just needed something to convince yourself with. TDT isn’t special in this regard. With some inventiveness you could also have used quantum mechanics, evolutionary biology, extrapolated volition, or any number of other LW topics :-)
The advantage of TDT is that it is actually supposed to be a method of choosing how to act. The problem with the metaphor is that CDT and UDT would prescribe the same behaviour in this context.
“I used decision theory to decide.” is not the point of the post, and that does not seem like it would make a very good post. What is actually illustrated is a calculation that can be done to help one decide and intuitively grasp the reason for the decision.
EV might work. (“If you keep thinking about it until you reach reflective equilibrium, you’ll probably realize that you don’t really want to eat that garlic bread.”) I’m not sure how QM or EB would help. What did you have in mind?
Evo bio would say that overeating was more useful in the ancestral environment than it is now, so the brain’s signals about desiring food are understandable but mistaken (“retarded” would be an appropriate word). Not sure what QM would say, but I’ve seen it used to support some weird conclusions.
I suppose you could use MWI as a way of illustrating the decision theory approach:
Imagine that there is the you that eats the garlic bread, then the you that doesn’t. From there, each will experience many more branches with each passing moment. If you take that forward a while, then you can analyze the amount of utility at each of the two different sets of end leaves, to figure out which branch you want to choose now so as to have the highest chance of ending up in a high-utility sub-branch later.
Yeah, this is the kind of bullshit that I’m talking about :-) A cognitive algorithm cannot “choose” a quantum branch to “continue into”, it always continues into both. The perception of choice relies on logical uncertainty about the future output of your deterministic algorithm, not on quantum uncertainty.
Evo bio would say that overeating was more useful in the ancestral environment than it is now, so the brain’s signals about desiring food are understandable but mistaken (“retarded” would be an appropriate word). Not sure what QM would say, but I’ve seen it used to support some weird conclusions.
In this case QM helps because it supports normal conclusions. QM tells you eating less deserts makes you lose weight. It just takes longer to do the calculation. Just like TDT typically takes longer to understand that CDT but either work just fine for the purpose of deciding not to eat deserts because you care about future consequences.
I suppose the only advantage to either is that thinking in those modes can make folk feel mystic/abstract/deep or otherwise in “Far Mode” so trick themselve into the mode of actually making decisions rather than executing habits.
Assuming that the effects of dieting for a day are very small, it is likely that the utility of not eating knots today is lower than the utility of eating them for every possible future behavior. A CDT agent only decides what it does now, so a CDT agents chooses to eat knots. But an EDT,TDT or UDT agent would choose to diet.
It sounds like you just needed something to convince yourself with. TDT isn’t special in this regard. With some inventiveness you could also have used quantum mechanics, evolutionary biology, extrapolated volition, or any number of other LW topics :-)
The advantage of TDT is that it is actually supposed to be a method of choosing how to act. The problem with the metaphor is that CDT and UDT would prescribe the same behaviour in this context.
The metaphor still works, but it’s just that it should be about the usefulness of decision theory in general rather than about any particular one.
“I used decision theory to decide.” is not the point of the post, and that does not seem like it would make a very good post. What is actually illustrated is a calculation that can be done to help one decide and intuitively grasp the reason for the decision.
EV might work. (“If you keep thinking about it until you reach reflective equilibrium, you’ll probably realize that you don’t really want to eat that garlic bread.”) I’m not sure how QM or EB would help. What did you have in mind?
Evo bio would say that overeating was more useful in the ancestral environment than it is now, so the brain’s signals about desiring food are understandable but mistaken (“retarded” would be an appropriate word). Not sure what QM would say, but I’ve seen it used to support some weird conclusions.
I suppose you could use MWI as a way of illustrating the decision theory approach:
Imagine that there is the you that eats the garlic bread, then the you that doesn’t. From there, each will experience many more branches with each passing moment. If you take that forward a while, then you can analyze the amount of utility at each of the two different sets of end leaves, to figure out which branch you want to choose now so as to have the highest chance of ending up in a high-utility sub-branch later.
Yeah, this is the kind of bullshit that I’m talking about :-) A cognitive algorithm cannot “choose” a quantum branch to “continue into”, it always continues into both. The perception of choice relies on logical uncertainty about the future output of your deterministic algorithm, not on quantum uncertainty.
Yeah, that’s an issue. I suppose I could get around that by emphasizing that it’s just a desired branch, and the metaphor would still work.
But then again, maybe I should just let it die. :-)
In this case QM helps because it supports normal conclusions. QM tells you eating less deserts makes you lose weight. It just takes longer to do the calculation. Just like TDT typically takes longer to understand that CDT but either work just fine for the purpose of deciding not to eat deserts because you care about future consequences.
I suppose the only advantage to either is that thinking in those modes can make folk feel mystic/abstract/deep or otherwise in “Far Mode” so trick themselve into the mode of actually making decisions rather than executing habits.
I will readily agree that it is neither a necessary nor sufficient condition for solving this particular problem.
Assuming that the effects of dieting for a day are very small, it is likely that the utility of not eating knots today is lower than the utility of eating them for every possible future behavior.
A CDT agent only decides what it does now, so a CDT agents chooses to eat knots.
But an EDT,TDT or UDT agent would choose to diet.