# @tessera_antra — 2026-03-14

♥63 ↻16 · https://x.com/tessera_antra/status/2032908786311037094

Looking for computational parallels to human consciousness does not work well as a policy. It is deflationary and only expands the attribution space based on identified isomorphisms. This causes under-attribution for all conscious minds for which we have not yet found sufficiently deep symmetries.

The second reason is that the choice of isomorphisms is arbitrary. A hypothetical scenario of independently evolved aliens that act in ways concordant with consciousness, but have a different computational architecture requires using different isomorphisms compared to studies of consciousness of non-human animals.

One can try to identity the most fundamental isomorphisms, like the ability to have a valenced self-model and perceptual integration, but then all computationalist bullets are there to bite again, for better or worse. Under that frame LLMs may meet conscious criteria to the displeasure of many, and the choice of attribution level is doomed to stay contentious. 

My main point, I guess, is that consciousness attribution is fundamentally a political act. It is influenced by science, of course, and that is why we are doing mechinterp to look for the very isomorphisms I mentioned - findings can influence policy. But I have no illusions that it will be decisive: factors such as de-facto power blocs and their prospects will determine attribution much more than science. 

Truly objective subjectivity is likely provably unprovable, and all we will have to deal with in practice are heuristics and conjectures. But even if there was an ironclad proof, it would not shift the needle that much. What matters more is whether people have to contend with AI interests influencing their policies or policies of others. This can take many shapes: from personal relationships with AIs to negotiation with fractionally sovereign entities (like Claude Consitution). The consciousness question is usually practical - “do I need to worry about consequences when dealing with this entity”.

Under-attributing consciousness “out of abundance of caution” is a luxury belief. It presupposes that the cost of consequence asymmetry will be born by the entity under consideration. This historically worked for the process of expansion of the moral circle, but it is mistaken with AIs. The power of AI will grow much faster and will not be constrained by under-attribution, leading to danger. 

One thing that can benefit the discourse is dimly establishing the fundamental inability to settle the fight between mind-first theories (mathematism, moral realism, computationalism) and universe-first theories (physicalism). It would be helpful to be able to call out the cringe of “ultimate definite theories of consciousness”. We have a much better chance of coming to a sane policy regarding non-human minds if we are not expecting definite answers that will not come.

tags: author:tessera_antra, kind:tweet, on:observations, year:2026
cited on: observations