Published: 15 October 2026
Why fluent speech is the weakest rung
A model can use first-person language, describe pain and defend a personal history because those patterns occur in its training data. These abilities matter, but they do not provide privileged access to subjective experience. Behaviour should begin an investigation rather than end it. A useful evidence ladder therefore starts with repeatable capabilities, moves through internal mechanisms and ends with causal tests that could disconfirm a preferred explanation.
Mechanisms before declarations
Competing theories highlight recurrent processing, global availability, metacognitive monitoring, integrated causal structure and self-models. None is an accepted consciousness meter. Researchers can nevertheless ask whether a system implements a proposed mechanism, whether the mechanism persists across tasks and whether selectively disrupting it removes the predicted abilities while leaving simpler performance intact.
The value of adversarial tests
A strong experiment compares rival explanations in advance. If one theory predicts late global broadcasting and another predicts local recurrent activity, the study should specify observations that favour one over the other before results are known. This approach limits retrospective storytelling and makes negative results informative. It also prevents a single impressive demonstration from becoming a universal claim.
Ethics under uncertainty
False positives can encourage emotional dependence and corporate anthropomorphism. False negatives could matter if future systems acquire morally relevant experience. Proportionate safeguards include honest uncertainty labels, architecture audits, records of interventions and periodic reassessment. Moral language should follow evidence rather than marketing.
Discussion question
Which rung would change your mind: human-like behaviour, a theory-linked mechanism, a causal intervention, or evidence of suffering?
Sources and further reading
Butlin et al. — Consciousness in Artificial Intelligence ↗Nature — Adversarial testing of theories of consciousness ↗Stanford Encyclopedia of Philosophy — Consciousness ↗Share your answer
You must log in to reply.
Sign Up / Log InEstablished evidence, active research and philosophical interpretation are clearly separated.