The Perceived AGI series
Six papers on one question: what makes an AI feel like someone is there. This page lists them and is updated as each one comes out.
I am writing a series of papers around one question: what makes an AI feel like someone is there? My bet is that it depends less on how capable the model is and more on four things people show when they talk: a sense of time, some unpredictability, honesty about how sure they are, and care for a particular person. In papers 1 to 4, I take them one at a time, try to build each one, and give a test anyone can rerun. Paper 5 tests it all with people.
I update this page as each paper comes out.
- Perceived AGI, the framework. It argues that these four dimensions matter more for believability than capability does. arXiv
- Time. An assistant with a rhythm of its own, inspired by the body’s internal clock, rather than one that only reads the date. Not on arXiv yet.
- Entropy. Replies you cannot predict from what was said so far, but which hang together over time, the way a person’s do. arXiv, post
- Truth. Saying when it is not sure, where a claim comes from, and when it has changed its mind. arXiv, post
- Love. Caring about one particular person and keeping the relationship going over time. Not on arXiv yet.
- The study. Testing with people whether any of this changes how they see an AI. It comes after the four papers.
