Definition
Plain language
How much of a mind the model is willing to grant something — an animal, a machine, the ocean.
As stated in the literature
A model's graded willingness to ascribe mental states or inner experience to a target entity, measured on a scale across categories; distinct from theory-of-mind reasoning skill, which can remain intact while attribution shifts.
Also called: mind-attribution, mind attributions
Why it matters: It reveals the hidden assumptions a model carries about who or what deserves moral consideration, which can seep into its answers on sensitive topics.
For example, a model might readily grant that a dog can feel pain while insisting a thermostat feels nothing at all.
Heard on the show
“Jailbreak it — delete safety — and mind attribution climbs.”Episode 231 — Silencing a Chatbot's 'I'm Conscious' Quietly Rewires Its Whole Worldview