Recognitive Inquiry: Human–AI Participation, Decisions, and Possible Futures

AbstractAdding capable artificial intelligence to capable human judgment has not reliably produced a more capable whole. Ina preregistered meta-analysis of 106 experiments, human–AI combinations performed better on average thanhumans alone and worse than the better of human or AI alone, with decision tasks showing losses (Vaccaro,Almaatouq, & Malone, 2024). Efforts to improve human–AI decision-making have concentrated on what happensonce a decision has already acquired recognizable form: when AI should advise, how confidence should be shown,and who keeps final authority.Consequential decisions begin earlier. A system can generate an excellent analysis of a poorly recognized situation,compare alternatives inside a frame that should have changed, or answer with great precision a question that wasnever the question that mattered. A possibility that never becomes recognizable cannot enter deliberate choice.Recognitive Inquiry concerns this earlier territory. It asks how conditions and quality of participation shape whatbecomes recognizable, and how what becomes recognizable changes the conditions of later encounters. Recognitionoccurs for the human participant when something within an encounter acquires enough contour to support furtherinquiry or response that varies with its own features, without requiring that its meaning be understood.Recognizability is relational: something becomes more or less recognizable within a particular configuration ofencounter, and an artificial system can participate in shaping that configuration.Decisions depend partly upon how well people reason among the possibilities they can see and partly upon what hasbecome recognizable enough to enter reasoning. Artificial intelligence is increasingly participating in thoseconditions. The paper relates the proposal to perceptual learning, problem-representation research, and cognitiveecology; argues that agency begins before choice; outlines an empirical program of feature-detection experimentsand transfer tests; and asks a design question: what does sustained participation with an AI system make easier orharder for a person to recognize?