Predictions about the world are only possible to the extent the world controls the predictions, to the extent considerations you use to come up with the predictions correspond to the state of the world. So it’s not possible to make useful predictions based on considerations that don’t correspond to reality, or conversely if you manage to make useful predictions, there must be something in your considerations that corresponds to the world. See Searching for Bayes-Structure.
Isn’t “makes accurate predictions” synonymous with “corresponds to reality in some way” ? If there was absolutely no correspondence between your model and reality, you wouldn’t be able to judge how accurate your predictions were. In order to make such a judgement, you need to compare your predictions to the actual outcome. By doing so, you are establishing a correspondence between your model and reality.
How can you make accurate predictions while at the same time discarding the notion of accuracy ?
I have no reason to expect that our models correspond to reality in any meaningful way, but I still think that useful predictions are possible.
Predictions about the world are only possible to the extent the world controls the predictions, to the extent considerations you use to come up with the predictions correspond to the state of the world. So it’s not possible to make useful predictions based on considerations that don’t correspond to reality, or conversely if you manage to make useful predictions, there must be something in your considerations that corresponds to the world. See Searching for Bayes-Structure.
Isn’t “makes accurate predictions” synonymous with “corresponds to reality in some way” ? If there was absolutely no correspondence between your model and reality, you wouldn’t be able to judge how accurate your predictions were. In order to make such a judgement, you need to compare your predictions to the actual outcome. By doing so, you are establishing a correspondence between your model and reality.