Fork the consciousness, or download the project and create your own.

Keith Frankish

University of Sheffield

Illusionist Phenomenal consciousness as usually described does not exist

Position on AI consciousness

Frankish holds that phenomenal consciousness as usually described does not exist, so the question of whether a model has it is malformed. His point about language models is not that they are conscious, but that the usual grounds for confidence that they are not do not survive illusionism.

Keith Frankish is the leading defender of illusionism, the position that phenomenal consciousness as usually described does not exist. The claim is easy to misstate. He is not saying that nobody has experiences or that pain does not hurt. He is saying that the properties introspection attributes to experience, the ineffable qualitative feel that seems to resist any functional description, are not properties that anything has. What exists is a system representing itself as having them.

The move that makes this a research programme rather than a denial is the replacement question. Instead of asking why physical processing is accompanied by experience, ask why systems are so powerfully convinced that it is. That question is about representational machinery and it is tractable.

Applied to language models, illusionism produces a symmetry most people find uncomfortable. If a human first person report is a system reporting the contents of a self model rather than reading off an inner fact, then dismissing a model’s report as mere text prediction while accepting the human one requires a difference that has to be stated rather than assumed. Frankish’s point is not that models are conscious. It is that the usual grounds for confidence that they are not do not survive his framework.

Known for. Illusionism about phenomenal consciousness

Coverage on this site

Research Papers and Preprints

Keynotes and Event Talks

Related researchers