Many of the perceptual deficits of autism — difficulty recognizing familiar faces when a hat or beard is added or removed, stress and meltdown when facing departures from routines or minor changes in environment such as a book out of place on a bookshelf, memory of perceptual details neurotypicals don’t notice or consider unimportant, difficulty distinguishing signals in the presence of background noise or environmental chaos, inflexible behavior — while commonly attributed to executive-function deficits, bear a suggestive similarity to a pathology in machine learning and in particular artificial neural networks called “overfitting”, in which a learner adapts itself to particular examples from its training set at the expense of generalizability. Perhaps, at the human neural level, overfitting is a general mechanism that causes many of the different characteristic manifestations of autism.