Whose “Intelligence” Are We Building? On Exclusion in Conceptual Frameworks of Artificial Intelligence

Abstract

Every definition of intelligence differentiates between haves and have nots, yet the exclusionary premises embedded in different conceptual frameworks of intelligence are rarely examined. This paper argues these premises matter both normatively and practically, as they are value-laden and operationalized in AI benchmarks used to evaluate “intelligent” systems. Drawing on Science and Technology Studies (STS) and intelligence research, we examine three dominant conceptual frameworks, i.e., anthropocentrism, pure ability, and generality, and show how they entail distinct speciesist and ableist assumptions that partially manifest in current AI evaluation practices. Rather than advocating for a universally inclusive definition of intelligence, we argue for epistemic reflexivity. To support this, we propose a structured set of questions for AI developers to make the normative assumptions embedded in benchmark design explicit.