Aesthetic judgement is a two-place function: “X likes Y.” But for human Xi’s, “X1 likes Y”, “X2 likes Y”, “X3 likes Y” etc. tend to correlate with each other. So one could in principle draw a network like Network 1 in “How An Algorithm Feels From Inside”, with nodes labelled “X1 likes Y”, “X2 likes Y”, etc.; but it would be computationally infeasible to use such a network for anything, so one uses a network like Network 2 instead, with the central node labelled “Y is beautiful”. (But in reality, if you knew whether X1 likes Y, whether X2 likes Y, whether X3 likes Y, etc., there would be no question whether Y is beautiful left to ask.) This is a useful approximation, but breaks down with things lots of people like and lots of people dislike, e.g. Justin Bieber’s music. (Even then, it may be useful to use a network like Network 2 but only including a certain subgroups of humans, e.g. musicians, or people like lukeprog who’ve heard lots and lots of different music, or people with IQ above 130, or people in my social circle, or people who wear leather jackets and long hair, etc.)
Of course, the reason why how much X1 likes Y is correlated with how much X2 likes Y is not telepathy—it’s that certain causal influences act on both. So, even if you know how much Xi likes Y for all i, there are questions left to ask.
Aesthetic judgement is a two-place function: “X likes Y.” But for human Xi’s, “X1 likes Y”, “X2 likes Y”, “X3 likes Y” etc. tend to correlate with each other. So one could in principle draw a network like Network 1 in “How An Algorithm Feels From Inside”, with nodes labelled “X1 likes Y”, “X2 likes Y”, etc.; but it would be computationally infeasible to use such a network for anything, so one uses a network like Network 2 instead, with the central node labelled “Y is beautiful”. (But in reality, if you knew whether X1 likes Y, whether X2 likes Y, whether X3 likes Y, etc., there would be no question whether Y is beautiful left to ask.) This is a useful approximation, but breaks down with things lots of people like and lots of people dislike, e.g. Justin Bieber’s music. (Even then, it may be useful to use a network like Network 2 but only including a certain subgroups of humans, e.g. musicians, or people like lukeprog who’ve heard lots and lots of different music, or people with IQ above 130, or people in my social circle, or people who wear leather jackets and long hair, etc.)
Of course, the reason why how much X1 likes Y is correlated with how much X2 likes Y is not telepathy—it’s that certain causal influences act on both. So, even if you know how much Xi likes Y for all i, there are questions left to ask.