If AI Is a Parrot, What Are We? Rethinking the Criteria for Human Understanding in Light of Artificial Intelligence

This paper breaks the question "Does AI understand?" down into four distinct questions: world model, meaning, access consciousness, and phenomenal consciousness. Reviewing evidence from mechanistic interpretability (internal representations of the Othello board, maps of place and time, multi-step reasoning in language models), it shows that the radical version of the "stochastic parrot" critique, which assumes no internal structure at all, is no longer consistent with the evidence, even though the philosophical version of that same critique (that representation is not the same as meaning) remains unresolved. The paper then applies the very criteria used to condemn the machine (post hoc rationalisation, problem-solving through heuristic shortcuts, the absence of a reliable measure of "raw intelligence") to humans themselves, and shows that these criteria condemn humans just as much. One genuine difference remains, however: biological self-generating organisation (autopoiesis), which no non-biological system has yet reproduced, though nothing indicates that this gap is metaphysical rather than engineering-related. By drawing an explicit line between functional convergence and the question of experience, which stays open, the paper concludes that parrot-like behaviour, in machines and in humans alike, is a spectrum rather than a strict binary; and perhaps the greatest danger is not that machines remain parrots, but that we do.