Another frame for the rejection of hypotheses on the whole is the embrace of instrumentalism, where tools such as AI (or physics or whatever) are only useful as tools (chiefly, as engines of prediction). I’ve contrasted this difference between conjectural scientific knowledge and instrumental prediction as a massive gap to AGI. To your point, I don’t think the difference is generally appreciated, perhaps because the depth of inter-theoretic scientific knowledge makes it elusive.

To your criticism that Computer Science is not really a science, this retreat from explanations into instrumentalism certainly captures it. But as the limits of instrumentalism in CS are generally becoming more understood (pragmatically, if not explicitly), things are changing. Interpretability might be one positive pressure. But for researchers, there’s an operational need to be more explicit about their underlying explanations (as you are here) to criticize and drive that research forward.

Thanks for a great article. Popper is generally ignored in CS, but his epistemology is central to understanding why AI works.