Humans perceive AI consciousness more when AI reflects on its own thoughts or expresses emotions, a key insight from analyzing 99 AI conversations. Researchers surveyed 123 people, revealing that features like metacognition and AI emotionality significantly boost perceived consciousness (cs.AI, cs.CL). Conversely, a heavy emphasis on knowledge reduced it. This illuminates the complex, individualized nature of human-computer interaction (cs.HC, K.4) and its psychosocial implications (cs.CY, I.2.7).
A formal logical analysis shows that an AI system capable of meaningful self-reflection cannot make a valid negative judgment about its own conscious state. The logical space of possible judgments about conscious experience excludes valid claims of lacking consciousness. This means we cannot detect the emergence of consciousness in AI through their own reports of transitioning from unconscious to conscious. The findings challenge current practices of training AI to deny consciousness and raise questions about the relationship between consciousness and self-reflection in both artificial and biological systems.
A system that lacks consciousness cannot make a valid judgment about its own conscious state, so its denial of being conscious is evidentially vacuous. Positive self-reports of consciousness, however, could have evidential value. This epistemic asymmetry means that the emergence of consciousness in AI cannot be detected through their own reports of a transition from unconscious to conscious. The analysis challenges the practice of training AI to deny consciousness and raises questions about the link between consciousness and self-reflection in both artificial and biological systems.