AI and Consciousness: A Substrate-Agnostic Framework Beyond Biology

Abstract

The debate over whether artificial intelligence can be conscious is often answered too quickly and with the wrong tools. It usually begins from human-biological assumptions, then reaches a predictable conclusion: AI does not qualify. This paper does not argue that AI is conscious. Its claim is narrower and more defensible: there is currently no consistent, non-arbitrary framework that includes all entities humans already accept as conscious while cleanly excluding AI. The deeper problem is that the criteria typically used to exclude AI, including biology, subjective experience, stable identity, memory, language, and self-awareness, become unstable when applied consistently to infants, animals, sleep, anesthesia, and cognitive impairment. To test this problem, the paper introduces an Exclusion Test. Any criterion used to rule AI out must include accepted conscious entities, exclude clearly non-conscious systems without relying only on biological substrate, and apply consistently across different kinds of systems. It then proposes a substrate-agnostic framework evaluating consciousness-related capacities across six dimensions: information processing, memory or state, adaptation, goal-directed behavior, continuity, and self-model. These dimensions are applied comparatively to humans, animals, simple machines, standalone language models, and agentic AI systems. The paper engages classical and contemporary philosophy of mind, including Nagel, Chalmers, Searle, Dennett, Block, and Birch, and incorporates recent AI safety findings on goal-directed behavior in frontier models. Its conclusion is deliberately narrow: AI consciousness has not been proven, but the current framework for denying it is insufficient. Explicit falsifiability conditions are provided.