I think the wiki does contain a written down definition:
Timeless decision theory (TDT) is a decision theory, developed by Eliezer Yudkowsky which, in slogan form, says that agents should decide as if they are determining the output of the abstract computation that they implement. This theory was developed in response to the view that rationality should be about winning (that is, about agents achieving their desired ends) rather than about behaving in a manner that we would intuitively label as rational.
I think what Sophronius describes in the paragraph would is what’s “intuitively labeled as rational”.
I think that’s sort of the problem with the post. It’s a list of 10 things that intuitively feel like they are the things rational people should do.
It’s not a list that tries to describe what reasoned principles about rationalism Lesswrong did come up with. TDT is sort of the LW house decision theory. It’s about moving beyond the intuitive idea of rationalism that popular out there. LW rationality is on the other hand supposed to be about winning.
I think the example of reacting when fear comes up is a good example. A nurse should follow the algorithm that if she feels a given patient is in a critical condition the patient gets extra supervision.
The intuitive rational belief that the nurse should have good reasons that she can explain to other people about why a patient needs supervision. The intuitive rational belief is that there should be reasons besides the emotions of the nurse to give the patient extra supervision.
We do have studies that validate the abstract heuristic that the nurse should let her feeling overrule her intellectual analysis of the situation.
If you read the original paper from two decades ago that introduces the concept of evidence-based medicine you find that it’s about getting medical professionals to read more scientific papers and deemphasized intuitive decision making.
We learned something in those two decades. We decided that rationality should be about winning. We don’t know everything but we can at least make an effort to be less wrong. We know that specific choices are well made with intuition than it would be stupid to not go the winning way and instead try to analyse the situation intellectually. Of course the nurse should still learn medical science but she should also listen to her intuition.
We are in the 21st century and not anymore in the 20st. End 20st century ideology is outdated and it’s useful to update. To get less wrong.
Is TDT the best way to think about making decisions? It’s still in it’s infancy and there still room to refine it. Let’s run CFAR workshop to see what heuristics are actually practical when you teach them to humans.
I think the wiki does contain a written down definition:
I am sorry, but that is not specified at all. If I give you a specific problem (I have a list of them right here!), will you be able to tell me what “the TDT answer” should be? The way people seem to use TDT is as a kind of “brand name” for a nebulous cloud of decision theoretic ideas. Until there is a paper and a definition, TDT is not a defensible point. It has to be formally written down in order to have a chance to be wrong (being wrong is how we make progress after all).
If it’s a set of related decision theories, fine—tell me what the set is! Example: “naive EDT” is “choose an action that maximizes utility with respect to the distribution p(outcome | action took place).” This is very clear, I know exactly what this is.
I think the wiki does contain a written down definition:
I think what Sophronius describes in the paragraph would is what’s “intuitively labeled as rational”.
I think that’s sort of the problem with the post. It’s a list of 10 things that intuitively feel like they are the things rational people should do.
It’s not a list that tries to describe what reasoned principles about rationalism Lesswrong did come up with. TDT is sort of the LW house decision theory. It’s about moving beyond the intuitive idea of rationalism that popular out there. LW rationality is on the other hand supposed to be about winning.
I think the example of reacting when fear comes up is a good example. A nurse should follow the algorithm that if she feels a given patient is in a critical condition the patient gets extra supervision.
The intuitive rational belief that the nurse should have good reasons that she can explain to other people about why a patient needs supervision. The intuitive rational belief is that there should be reasons besides the emotions of the nurse to give the patient extra supervision.
We do have studies that validate the abstract heuristic that the nurse should let her feeling overrule her intellectual analysis of the situation.
If you read the original paper from two decades ago that introduces the concept of evidence-based medicine you find that it’s about getting medical professionals to read more scientific papers and deemphasized intuitive decision making.
We learned something in those two decades. We decided that rationality should be about winning. We don’t know everything but we can at least make an effort to be less wrong. We know that specific choices are well made with intuition than it would be stupid to not go the winning way and instead try to analyse the situation intellectually. Of course the nurse should still learn medical science but she should also listen to her intuition.
We are in the 21st century and not anymore in the 20st. End 20st century ideology is outdated and it’s useful to update. To get less wrong.
Is TDT the best way to think about making decisions? It’s still in it’s infancy and there still room to refine it. Let’s run CFAR workshop to see what heuristics are actually practical when you teach them to humans.
There are a bunch of folk rationality beliefs.
I am sorry, but that is not specified at all. If I give you a specific problem (I have a list of them right here!), will you be able to tell me what “the TDT answer” should be? The way people seem to use TDT is as a kind of “brand name” for a nebulous cloud of decision theoretic ideas. Until there is a paper and a definition, TDT is not a defensible point. It has to be formally written down in order to have a chance to be wrong (being wrong is how we make progress after all).
If it’s a set of related decision theories, fine—tell me what the set is! Example: “naive EDT” is “choose an action that maximizes utility with respect to the distribution p(outcome | action took place).” This is very clear, I know exactly what this is.