Summary

When someone becomes unable to make their own decisions – because of illness, an accident, or a condition like dementia – a trusted person with legal authority (Power of Attorney) needs to make decisions on their behalf. Research shows this is hard to do well, with decisions being right only two-thirds of the time and attorneys experiencing anxiety and guilt.

This project asks whether AI could help. We'll build “AI Simulated Individual Minds” – AI models trained on in-depth interviews and surveys that capture a person's values and preferences – and test whether people who can consult one make more accurate decisions, and feel less burdened, than those who can't.

We'll work with volunteer pairs: the person whose values are captured, and the person who would decide on their behalf for them. Alongside the study, we'll develop clear ethical guidelines covering issues around trust, identity, privacy, safety, security, consent, and appropriate use, to support responsible design and legislation for such tools. No actual Power of Attorney decisions will be made during this project, and all AI models and personal data will be deleted at the end of the project.