We have been debating what intelligence is for two and a half thousand years, and we are no closer to a definition that generalises. Psychometrics, cognitive-developmental theory, pluralism, the computational and connectionist paradigms, and the embodied and enactive paradigms each capture something real, and each fails as a general account. I argue that this is not a failure of effort, but a sign that we have been looking for the wrong kind of thing. Intelligence resists definition, but it does…
Read moreWe have been debating what intelligence is for two and a half thousand years, and we are no closer to a definition that generalises. Psychometrics, cognitive-developmental theory, pluralism, the computational and connectionist paradigms, and the embodied and enactive paradigms each capture something real, and each fails as a general account. I argue that this is not a failure of effort, but a sign that we have been looking for the wrong kind of thing. Intelligence resists definition, but it does not resist recognition. What we do, in practice, is recognise it - and that recognition is graded, observer-relative, and dependent on task and context. Drawing on cases from across the biological continuum, and on the behaviour of current AI systems, I set out a description rather than a definition: a profile of six components, assessed as a shape rather than collapsed into a score. The framework I describe is deliberately medium-neutral: I treat the question of whether intelligence requires a biological substrate as open, rather than settled in either direction. That intelligence is something we recognise rather than something we define points to a more pragmatic resolution. For most practical purposes - deciding what to trust a system with, and under what oversight - "Is it intelligent?" is the wrong question. The better question, more modest and answerable, is "Does the system do what we need it to do, well enough, in the setting where we need it?"
