The problem of taste is philosophically deep and relates to issues in epistemology that have only recently been resolved with new ideas from information theory and machine learning.
Part of the problem is about complex inductive processes. Human knowledge has encompassed simple deduction (Aristotelian syllogism), simple induction (P values, N=200 medical studies), and complex deduction (operating systems, compilers, etc). We are comfortable with how these systems of reasoning work.
But art relates to the fourth quadrant - complex induction, with billion-parameter models trained against enormous datasets. When a human says "I think this painting is beautiful, but that one is ugly", they are expressing something about the response of their visual cortex (a multi-billion parameter learning system) to a rich stimulus (say, 1 million pixels). With the advent of research like GPT-3, we can now build these systems, though we don't understand their properties very well. There is a lot more for us to understand here, both conceptually and technically.
Part of the problem is about complex inductive processes. Human knowledge has encompassed simple deduction (Aristotelian syllogism), simple induction (P values, N=200 medical studies), and complex deduction (operating systems, compilers, etc). We are comfortable with how these systems of reasoning work.
But art relates to the fourth quadrant - complex induction, with billion-parameter models trained against enormous datasets. When a human says "I think this painting is beautiful, but that one is ugly", they are expressing something about the response of their visual cortex (a multi-billion parameter learning system) to a rich stimulus (say, 1 million pixels). With the advent of research like GPT-3, we can now build these systems, though we don't understand their properties very well. There is a lot more for us to understand here, both conceptually and technically.