Artificial Intelligence Is a Misnomer: Purpose, Consciousness and the Mischaracterization of AI Risk

The term intelligence fuses two things. The first is the capacity of a conscious, evolved organism to set an end and make that end its own. The second is the substrate of information processing: the acquisition, retention, generalization and pattern recognition by which paths to a goal are evaluated and selected. Only the first deserves to be called intelligence in the full sense; the second is inert in isolation, its information made useful by a purposive, conscious subject rather than found in the information itself. Drawing on the distinction between intrinsic and derived intentionality (Searle 1980, 1992), the symbol grounding problem (Harnad 1990) and the frame problem of disembodied reasoning (McCarthy and Hayes 1969; Dennett 1984; Dreyfus 1992), I argue that artificial systems possess only the second form together with goals supplied from outside, which makes them tools and the phrase artificial intelligence a misnomer. This does not minimize the dangers of advanced AI; it relocates them. If artificial systems originate no purposes of their own, then every risk they pose traces to the conscious beings who build, instruct and deploy them. Today's systems, their competence distilled from human knowledge and unable to renew without human updating, are not even general intelligences but derived instruments; the genuine achievement of the era is a general-purpose learning algorithm, not a mind. A serious theory of AI risk must begin with a serious theory of intelligence, one that refuses to mistake information processing for understanding, optimization for desire and capability for a will.