Human review
AI can help organize information and prepare questions. Clinical and legal decisions need qualified professionals and the people involved.
The approach
I’m starting with tasks my family has found difficult and trying out ways AI might help. That includes organizing information, supporting communication, and preparing for conversations with professionals.
Working principles
AI can help organize information and prepare questions. Clinical and legal decisions need qualified professionals and the people involved.
Prototypes can be wrong or incomplete. I want people to be able to see the source of an update and know when something needs checking.
These projects involve sensitive subjects. Public examples should avoid identifying details and personal records, and any testing needs care around consent and access.
I’m thinking about different languages, difficulty speaking or using a screen, and relatives trying to help from another country.
The long-running idea
By self-evolving, I mean a knowledge base that could change as new documents and updates arrive, while keeping its sources and a history of changes. If two updates disagree, the aim would be to flag that for review.
Long-running agents could refer to this shared context between conversations, helping with family updates, appointment questions, requested check-ins, and follow-ups. This would need human oversight; an AI-generated record can contain mistakes.
This remains exploratory. The ongoing agent capabilities described here are not deployed services.
Keep new updates with earlier context.
Sources stay attached to important facts.
Flag conflicting information for review.
Use shared context for ongoing coordination.
Clear limits
These are personal, noncommercial experiments. They do not provide diagnosis, treatment, or legal advice, and no medical effectiveness or regulatory approval is claimed. They are not broadly available. Further testing would need appropriate consent, privacy protections, and professional input.
Why this work began