Seven Axioms on the Nature of Generative AI: Laying the Foundation for a Genuine Understanding of Machine Intelligence

Abstract

We challenge tendencies to evaluate Generative AI (GenAI), and large langue models (LLMs) in particular, through the lens of human intelligence, arguing that such anthropomorphic renditions constitute a fundamental category error. We do not dispute LLMs' capabilities, but contend that analogies to human intelligence obscure its true nature. This leads to unrealistic expectations and risks overlooking unhuman-like capabilities. We propose seven axioms to characterize GenAI and LLMs: (1) no direct relationship to truth, (2) no connection to the physical world, (3) absence of subjectivity, (4) lack of temporality, (5) no intentionality, (6) purely relational information representation, and (7) complex pattern prediction. These axioms, informed by existential philosophy, linguistics, and cognitive science, reveal GenAI as an "alien intelligence" fundamentally different from human cognition. We argue for developing a genuine machine psychology that embraces GenAI's alien nature rather than constraining it within human frameworks, to unlock applications that leverage its unique capabilities.