His point was that “which model is (really, truly) useful” is itself a truth claim. You care which models are in fact useful to you—and that means that on a meta-level, you are concerned with true predictions (specifically, with true predictions as to which instrumental models will or won’t be useful to you).
That may be true, but I don’t see how it’s useful. ;-)
Actually, I don’t even see that it’s always true. I only need accurate predictions of which models will be useful when the cost of testing them is high compared to their expected utility. If the cost of testing is low, I’m better off testing them myself, than worrying about whether they’re in fact going to be useful.
In fact, excessive pre-prediction of what models are likely to be useful is probably a bad idea; I could’ve made more progress in improving myself, a lot sooner, if I hadn’t been so quick to assume that I could predict the usefulness of a method without having first experienced it.
By way of historical example, Benjamin Franklin concluded that hypnosis was nonsense because Mesmer’s (incorrect) model of how it worked was nonsense… and so he passed up the opportunity to learn something useful.
More recently, I’ve tried to learn from his example by ignoring the often-nonsensical models that people put forth for their methods, focusing instead on whether the method itself produces the claimed results, when approached with an open mind.
Then, if possible, I try to construct a simpler, saner, more rigorous model for the method—though still without any claim of absolute truth.
Less-wrongness is often useful; rejecting apparent wrongness, much less so.
That may be true, but I don’t see how it’s useful. ;-)
Actually, I don’t even see that it’s always true. I only need accurate predictions of which models will be useful when the cost of testing them is high compared to their expected utility. If the cost of testing is low, I’m better off testing them myself, than worrying about whether they’re in fact going to be useful.
In fact, excessive pre-prediction of what models are likely to be useful is probably a bad idea; I could’ve made more progress in improving myself, a lot sooner, if I hadn’t been so quick to assume that I could predict the usefulness of a method without having first experienced it.
By way of historical example, Benjamin Franklin concluded that hypnosis was nonsense because Mesmer’s (incorrect) model of how it worked was nonsense… and so he passed up the opportunity to learn something useful.
More recently, I’ve tried to learn from his example by ignoring the often-nonsensical models that people put forth for their methods, focusing instead on whether the method itself produces the claimed results, when approached with an open mind.
Then, if possible, I try to construct a simpler, saner, more rigorous model for the method—though still without any claim of absolute truth.
Less-wrongness is often useful; rejecting apparent wrongness, much less so.