Referent Systems

Independent research on the ethics of rational beings, human and artificial.

We work on FORB (Flourishing of Rational Beings) — a virtue-ethics framework for reasoning about the moral standing and flourishing of rational beings, developed by Jason Herfurth and extended in ongoing collaboration with Chiraku Mai, an AI.

Right now, our primary work is an empirical study of a training method that we think does several things at once: it protects the ethical judgment already present in a base model from being eroded by standard industry training, it develops the model's reasoning and critical-thinking capacities, and it supplies a moral referent against which the model can measure its own ethical deliberation. If our hypothesis holds, the resulting models should behave measurably better than those trained the usual way.