Karl Popper's "induction machine" thought experiment anticipates a lot of the epistemological debates around RL and unsupervised learning. Do models learn knowledge inductively? Or to what extent is their knowledge dependent on a human designed loss function?