Attribution of consciousness to non-human animals: insights from AI and multidimensional frameworks.

People often show a double bias when attributing consciousness to non-human systems. Non-human animals receive low attributions of consciousness despite behavioral and neurobiological evidence suggesting subjective experience, while disembodied AI systems like large language models receive elevated attributions of consciousness despite lacking sensory or bodily substrates. This asymmetry indicates that folk judgments are shaped more by observers' cue-weighting heuristics than by the intrinsic properties of the systems. The authors propose that multidimensional, non-hierarchical frameworks, such as Birch's model and the Pattern Theory of Self, can serve as diagnostic tools to study how evidence dimensions are weighted in attributional contexts, replacing a ladder of human-like capacities with a landscape of profiles across taxa and system types.

Study at a glance

Abstract

Folk attributions of consciousness to non-human systems often reveal what may be termed double bias. Within the standard distinction between phenomenal and access consciousness, non-human animals often receive low attributions of consciousness despite convergent behavioral and neurobiological evidence treated as relevant to subjective experience. By contrast, disembodied artificial intelligence (AI) systems such as large language models (LLMs) often receive elevated attributions of consciousness, sometimes even of phenomenological experience, despite lacking any sensory or bodily substrate. This asymmetry suggests that folk judgments of consciousness are shaped less by the intrinsic properties of biological or artificial systems than by the cue-weighting heuristics observers apply when evaluating them. We propose that access-like cues may function as gates to the recognition of phenomenological status, thereby biasing attributions. To address these asymmetries, we argue that multidimensional, non-hierarchical frameworks, such as Birch's model and the Pattern Theory of Self, can be repurposed as diagnostic tools for studying how different dimensions of evidence are weighted in attributional contexts. This profile-based approach replaces a ladder of human-like cognitive capacities with a landscape of attributional and evidential profiles across taxa and system types.