Discussion about this post

User's avatar
Seth's avatar

What a lovely and intelligent essay, to which I will append a deeply stupid example.

Many statistical inference algorithms are based on drawing repeated samples from a distribution. You only know what the distribution is like by looking at the samples. A bad sampling algorithm will get stuck basically sampling the point over and over again. It will look like it drew 100 samples, but actually you only got, like, 2 pieces of useful information.

A good sampling algorithm will sample independently. The next sample has nothing to do with with the last sample. If it looks like it drew 100 samples, you got 100 pieces of information.

A GREAT sampling algorithm will draw samples that are *anti-correlated*. It will draw 1 sample, then it will fly off into the distance and draw the next sample from as far away as it can go. If it looks like it drew 100 samples, you actually got like 200 pieces of information.

Shakespeare, at his best, writes like a GREAT sampling algorithm. Every idea is attacked from 20 different directions. The full space of interpretations, associations, implications is smashed into a few dozen lines. Sometimes it looks like it is losing coherence, losing its grip on the original problem; but no, actually, the problem is just much bigger and much weirder than it first seemed.

Qasim Hameed's avatar

I really enjoyed this essay, Nabeel—thank you for sharing your very eloquent thoughts.

I’d consider a fourth point, which I believe follows from the three you've already mentioned: part of what makes "great" art meaningful to us is our sense that it was hard for the artist to bring all its great elements together. The struggle of making it gives it a considerable part of its value. AI art, since it's designed to be created with "ease", lacks this necessary element of struggle, and feels inferior even if the output may appear compelling.

53 more comments...

No posts

Ready for more?