We are, as a species, compulsive askers. The toddler’s incessant “Why?” is not merely endearing, it is a form of epistemic insubordination against adult complacency. But somewhere between primary school worksheets and committee meetings, the question gets tamed. Neutered. Reduced to a polite gesture of clarification. If we are honest, most of us stop asking altogether.

AI now learns to ask. More precisely, it learns to prompt. Prompt engineering, the art of crafting inputs that elicit optimal outputs from large language models, shares uncanny DNA with complex question-asking. Both require clarity, creativity, context awareness, and the intuition to anticipate response structures. Raz and Kenett hint at this parallel: the better we train humans to ask, the better we will train machines to respond, and, potentially, to ask in turn. But this mutual bootstrapping carries its own paradox. As humans become more adept at crafting precise prompts for AI, an act that reflects the formulation of well-structured questions, they hone their own epistemic strategies.

In turn, AI systems respond with increasingly sophisticated outputs, some of which model, even if imperfectly, the heuristics of inquiry. The more we train these models to ask and answer, the more we are forced to refine what we mean by a ‘good’ question. And yet, the machine’s question does not arise from anxiety or awe. It does not grieve its ignorance. We do. That is the irremediable difference.