For example, I'm currently writing a Lisp program that uses an LSTM neural network to compose Scottish fiddle music. As a Scottish fiddler (and composer) and AI enthusiast, this is an intensely interesting project to me. And it's certainly an insane idea with a tiny market and no way to monetize. And that's how it will remain. Scottish fiddling will always be an obscure niche, even within that community there will never be much demand for new music (and there's so many buried gems from the 18th century that this will virtually never change), and any existing demand can be easily fulfilled by the scores of people--including myself--who compose fiddle music the old-fashioned way.
Unlike Altair BASIC, this isn't a case of somebody creating a product in a growing industry people don't realize they need yet. Even I--the "founder"--don't see any potential and realize I'm providing free labor. I'm just in it because it's a great hack.
tl;dr - Distinguish between "anticipating public demand" and "doing something only for yourself".