Ungrounding intelligence: From biological minds to artificial cognition

This paper proposes a two-dimensional framework for classifying artificial cognitive abilities. The first refines the traditional distinction between narrow and general intelligence, arguing that narrow intelligence can still constitute genuine intelligence when properly defined. The second, drawn from the grounded cognition paradigm, distinguishes between grounded and ungrounded intelligence depending on whether cognitive abilities are linked to embodiment, perception, and intrinsic motivation. We maintain that ungrounded systems - although disembodied and mindless - can nonetheless display adaptive, intelligent behavior. To formalize this distinction, we introduce the Value Grounding Problem, which asks whether an agent's values are internally generated or externally imposed. This framework yields a fourfold taxonomy that clarifies how different forms of artificial and biological intelligence relate to cognition, agency, and motivation, while outlining the prospects and limits of both grounded and ungrounded intelligence.