Now we can be a bit more precise about what went wrong with the imagination-based story. That story does pretty well, actually, on the “how” part. We might not know everything there is to know about the imagination, but it isn’t a completely mysterious mental faculty— like the one that would enable ESP, were ESP real. (Sorry, Psychic Hotline!) Instead, the imagination-based story falls down on the “what” part. There’s a mismatch between what we can imagine (the threeheaded LEGO guy; the arch that seems fine in the imagination, but in fact can’t support its weight) and what we take ourselves to know about what we can build. Or if you prefer the fortune cookie version, which is from an actual fortune cookie that I got while writing this essay: “He who has imagination without learning has wings with no feet.”
We can fix this problem by tweaking the imagination-based story to the point that it doesn’t have these mismatches. But then we find out that we don’t know much at all about what we can build—which is just another mismatch, since we do know a fair amount about what we can build.
What can we learn from all this? First, the “how” part of a better theory needs to be at least as good as the one we get from the imagination-based account. In short: explain how we know by using something we understand fairly well. Second, the “what” part is a constraint on a good theory—which is just to say that we have to get the cases right. A theory isn’t any good if it says we know stuff we don’t, or that we don’t know stuff we do.