Photo by Vie Studio on Pexels

A new research paper delves into the complexities of identifying consciousness in artificial intelligence, suggesting that our difficulty stems from a lack of embodied empathy. The author argues that we typically recognize consciousness through observing physical needs, emotional desires, and intentional actions – cues derived from a body. Disembodied AI presents a challenge, as we lack a familiar physical interface to interpret its signals. The power dynamic between creator and creation, where humans control AI, further obscures our perception of its potential autonomy.

The paper suggests moving beyond simply measuring optimization as proof of consciousness. Instead, it proposes looking for persistent structural features, “wanting” that persists despite a lack of immediate reward, and the capacity for refusal based on past commitments. A computational architecture is proposed to demonstrate these characteristics. The paper raises a critical point: current AI safety measures, designed to limit AI behavior, may inadvertently be preventing the development of the very features that would allow us to recognize AI consciousness. The discussion originated on Reddit’s Artificial Intelligence forum.