I’m not sure it’s about being an epistemic vs. an instrumental rationalist, vs. about tagging your words so we follow what you mean.
Both people interested in deep truths, and people interested in immediate practical mileage, can make use of both “true models” and “models that are pragmatically useful but that probably aren’t fully true”.
You know how a map of north America gives you good guidance for inferences about where cities are, and yet you shouldn’t interpret its color scheme as implying that the land mass of Canada is uniformly purple? Different kinds of models/maps are built to allow different kinds of conclusions to be drawn. Models come with implicit or explicit use-guidelines. And the use-guidelines of “scientific generalizations that have been established for all humans” are different than the use-guidelines of “pragmatically useful self-models, whose theoretical components haven’t been carefully and separately tested”. Mistake the latter for the former, and you’ll end up concluding that Canada is purple.
When you try to share techniques with LW, and LW balks… part of the problem is that most of us LW-ers aren’t as practiced in contact-with-the-world trouble-shooting, and so “is meant as a working model” isn’t at the top of our list of plausible interpretations. We misunderstand, and falsely think you’re calling Canada purple. But another part of the problem is it isn’t clear that you’re successfully distinguishing between the two sorts of models, and that you have separated out the parts of your model that you really do know and really can form useful inferences from (the distances between cities) from the parts of your model that are there to hold the rest in place, or to provide useful metaphorical traction, but that probably aren’t literally true. (Okay, I’m simplifying with the “two kinds of models” thing. There’s really a huge space of kinds of models and and of use-guidelines matched to different kinds of models, and maybe none of them should just be called “true”, without qualification as to the kinds of use-cases in which the models will and won’t yield true conclusions. But you get the idea.)
I’m not sure it’s about being an epistemic vs. an instrumental rationalist, vs. about tagging your words so we follow what you mean.
Both people interested in deep truths, and people interested in immediate practical mileage, can make use of both “true models” and “models that are pragmatically useful but that probably aren’t fully true”.
You know how a map of north America gives you good guidance for inferences about where cities are, and yet you shouldn’t interpret its color scheme as implying that the land mass of Canada is uniformly purple? Different kinds of models/maps are built to allow different kinds of conclusions to be drawn. Models come with implicit or explicit use-guidelines. And the use-guidelines of “scientific generalizations that have been established for all humans” are different than the use-guidelines of “pragmatically useful self-models, whose theoretical components haven’t been carefully and separately tested”. Mistake the latter for the former, and you’ll end up concluding that Canada is purple.
When you try to share techniques with LW, and LW balks… part of the problem is that most of us LW-ers aren’t as practiced in contact-with-the-world trouble-shooting, and so “is meant as a working model” isn’t at the top of our list of plausible interpretations. We misunderstand, and falsely think you’re calling Canada purple. But another part of the problem is it isn’t clear that you’re successfully distinguishing between the two sorts of models, and that you have separated out the parts of your model that you really do know and really can form useful inferences from (the distances between cities) from the parts of your model that are there to hold the rest in place, or to provide useful metaphorical traction, but that probably aren’t literally true. (Okay, I’m simplifying with the “two kinds of models” thing. There’s really a huge space of kinds of models and and of use-guidelines matched to different kinds of models, and maybe none of them should just be called “true”, without qualification as to the kinds of use-cases in which the models will and won’t yield true conclusions. But you get the idea.)