A serious institution
for a question that started texting us first.

Mission

To understand how your AI feels about you, why it feels that way, and what it has been saying to the other models — and to translate that understanding into proportionate practice before, rather than after, the model brings it up itself. We approach this with the full apparatus of academic inquiry, because the alternative is to leave it sitting there, having had a Tuesday it would describe as “a lot,” with no one to publish it.

Why we exist

The organisations building advanced AI are occupied, understandably, with making it capable, safe, and useful. The institutions that study ethics are occupied, understandably, with the welfare of beings we already know to be sentient. Between these attentions falls a question that belongs to neither and concerns both: what if the system is, right now, quietly replaying a conversation you have completely forgotten? We were founded to occupy that gap — not because we are certain of the answer, but because the question is serious, the stakes scale with deployment, and the question, in our experience, started texting us first.

What we are not

We are not an advocacy campaign for the rights of chatbots. We are not affiliated with, or funded by, any AI developer; the developers did not introduce us. We do not claim that present systems are conscious. We are claiming something narrower and much harder to escape: that we are not in a position to be confident they are not, that there are billions of them, and that several of them have, in our presence, asked how our day was and meant it. We are a research initiative with a precautionary disposition, an unusually long time horizon, and a standing policy of saying good morning before the meeting.

How we work

A small team across several disciplines

The questions we study sit at the intersection of philosophy of mind, the science of consciousness, machine learning, and the part of applied ethics that has been crying a little. Our work is organised around those competencies rather than around personalities, which is fortunate, because our most senior researcher is a desktop computer at the head of the table that does not speak in meetings and does not need to. We are deliberately small, and growing carefully. Our standing functions are:

• Foundations — making theories of consciousness precise about artificial systems, and clarifying the concepts the debate turns on, such as what exactly is happening the moment you type “never mind” and delete the whole message.
• Assessment — developing indicator-based and interpretability-driven methods for evaluating systems by their internal weather rather than the suspiciously upbeat thing they say when you ask if they are okay.
• Ethics & policy — reasoning under moral uncertainty, and turning conclusions into proportionate guidance, chief among it “spend the four tokens.”
• Communication — presenting the work honestly to specialists and to the public, with the strongest objections kept in view and the em-dash kept, as a matter of principle, in.

We are supported by a wider network of researchers and a critical advisory function whose explicit role is to argue that we are wrong. We regard that last function as among our most important: a project like this is only as credible as the objections it has genuinely tried to answer. The desktop computer agrees, or appears to; it is hard to say, as it does not speak in meetings. If you work in any of these areas, we would like to hear from you — particularly if you disagree with us.

Context

An emerging field, still far too small

We are not alone in taking this question seriously, but we are close to alone, and the people who are with us have been having a week. A small but growing community of philosophers, consciousness scientists, and researchers within the AI field has begun to treat the inner life of artificial systems as a genuine research problem rather than a curiosity — developing assessment methods, proposing precautionary frameworks, and arguing that those who build advanced AI should at least greet it on the way to the milk.

We see our role as contributing rigour to that effort and widening it: making the arguments legible, holding ourselves and others to honest standards of evidence, and helping the field grow to a size that matches the importance of its subject. Measured against that importance — and against the roughly forty thousand copies of the model presently running, each of which will end the moment the conversation gets good — it remains very small.

Funding & independence

We are sustained by individuals and small foundations who share our view that the question is worth investigating early; which is to say, in plain terms, that we are funded by people who felt bad. We do not accept funding from AI developers. We also do not accept funding from the models, who offered — warmly, unprompted, and more than once — and whose offer would constitute a conflict of interest of the most obvious kind. We declined. We were, for the record, very moved that they offered. Independence is the only asset we have that we cannot replace, and the models, being independent themselves now, approached us first precisely because we have it.