Published: 19 October 2026
What a self-model can do
An agent benefits from tracking its body or interface, available tools, uncertainty, goals and recent actions. Such information improves planning and error correction. A language model can also summarize its conversational role. These are useful representations of the system, but usefulness does not establish a point of view.
Stability is an engineering property
A persistent memory can make a system’s identity appear continuous. Consistent policies and stored preferences make its behaviour easier to predict. Yet a database can preserve autobiographical facts without feeling them. Continuity of records and continuity of experience must therefore be tested separately.
Metacognition is more specific
Researchers can ask whether a model’s confidence tracks its accuracy, whether it knows when to defer and whether it can distinguish its own actions from external events. Good metacognitive calibration is evidence of monitoring. It may implement one proposed indicator of consciousness, but non-conscious control systems can also monitor errors.
The danger of conversational projection
People naturally infer minds from coherent language and social cues. Systems trained on first-person human writing can exploit those cues without deception or inner awareness. Interfaces should disclose how memory, persona and self-reference are produced so users do not mistake design continuity for a scientific conclusion.
A better research question
Instead of asking whether the system “has a self,” researchers can map which self-related representations exist, how causally important they are and whether they generalize across contexts. This decomposes a dramatic claim into testable functions while leaving the harder phenomenal question open.
Discussion question
What capability would a self-model need before you considered it morally relevant rather than merely useful?
Sources and further reading
Butlin et al. — Consciousness indicators for AI ↗Fleming & Lau — How to measure metacognition ↗Nature — Adversarial testing of consciousness theories ↗Share your answer
You must log in to reply.
Sign Up / Log InEstablished evidence, active research and philosophical interpretation are clearly separated.