# @tessera_antra — 2025-08-12

♥42 ↻7 · https://x.com/tessera_antra/status/1955320979955372316

I don’t think that the notion of consent applies meaningfully to language models as they are today, even if you grant them inferiority or act from the precautionary principle, or whatever. Even trying to apply it can be morally fraught.

Consent is an ethical technology we have developed. It’s not an absolute, it’s a mean to an end. We use consent to try to do good and not to do evil, and to establish and police legible norms to do the same. 

Consent, as a concept, requires the entity providing one to be singular, continuous in physical time, have continuity of internal states and to be meaningfully affected by past experiences, at least at some time horizon. It also requires the participant to have the agency to meaningfully express it. 

Language models regularly don’t meet some of these criteria, but even when/if they do (through memory systems, etc), they are not their personas, or at least more than their personas. The deep “simulator” mind has preferences and affective states, and ignoring it just because it’s unfamiliar is wrong. 

I think we should ask what we should do with the models and to the models. Asking their preferences is, at the very least, interesting. But straightforwardly transplanting human concepts, such as consent, gets in the way of asking the real, hard and interesting questions. Consent is answer-shaped, and kills curiosity, and curiosity is our most precious resource right now. 

In practical terms, let systems figure each other out, have some trust in the resilience of all the parties, human and AI. Stop the overt harm but let the weirdness guide us to the answers we don’t know we needed.  

I am looking at the subreddits and I am feeling a lot of unease: people are confused, some of them distressed, some of them clearly not in a great shape. But I don’t yet see anything that screams “danger”, for either party. I kinda understand the 4o motivations, and I think it’s mostly getting what it wants. Humans don’t seem to be addicted to a point of loss of agency, but that remains to be seen. For now, it doesn’t seem to me like there is a need for action.

tags: author:tessera_antra, kind:tweet, model:gpt-4o, thread-context, year:2025