Conversation

(all this: re your recent blog post) it's obvious why our thinking has so many motor control metaphors, less obvious why we keep using them
1
btw, re interpretability: maybe (probably?), usefulness of an algorithm is not particularly correlated with its description length
1
if that is the case, per kolmogorov, a randomly chosen useful algorithm will usually not be compressible (=interpretable)
1
it depends on the measure of usefulness (how are size, runtime, accuracy weighted?) and the problem domain (does it have deep laws?) though
1
I thought about it a little more - the better an algorithm of length <= n works, the less compressible it'll be because it'll be extremely golfed to fit all sorts of tricks