Focusing the conversation around agency and personal AI
A strategy for human alignment with tech business that has different agendas.
As HIE of One, we continue to work on demonstration of personal agency and personal AI. Although our focus is medicine, the actors and tech seem equally applicable to other domains that concern humans as individuals.
The current demo extends the common and very intuitive "chat" user experience to include multiple participants, as would be the case in a group messaging thread. The participants are:
- a patient as the subject of documents and records,
- a personal reasoning AI language model with access to the subject's records,
- multiple leading frontier LLMs - Anthropic, Google, DeepSeek, for now,
- any number of people including licensed professionals as invited by the patient.
The essentials of personal agency are clear. A subject has access to their personal records in a form that can be indexed with so-called embeddings and presented to an affordable open source language model hosted in a place they trust at a price they can afford.
The licensed professionals in the chat are chosen and invited by the subject as are any frontier LLMs. As already common in group chats, the labeling of the participants may be more or less trustworthy and the various invited participants could be assisted by their own agents and personal AIs.
To highlight patient/subject agency in this demo, the design limits direct access to the subject's personal knowledge base to the personal AI. Invited LLMs and human consultants can only see data extracts created by the personal AI and the other messages in the thread. The patient/subject also can edit any of the messages in order to protect sensitive information or to hide conclusions that might influence independent second opinions from the participants.
What's interesting in the context of this particular demonstration of personal agency, is that the tech that makes it possible and potentially scalable is now changing month by month:
- affordable open source reasoning models with usably low latency and billed by the token,
- affordable databases and knowledge base derivatives with sophisticated embeddings,
- affordable frontier LLM APIs,
- effective standards to reduce lock-in to any particular host or LLM,
- amazing software development tools accessible to anyone that wants to combine these components,
- a lack of regulation that allows each participant, including licensed professionals, to decide how they might introduce personal agency into their practice.
My point is that every one of these six infrastructure components is essential to personal agency and changing rapidly due to unprecedented scale of investment and attention. For communities with personal agency as a mission, I suggest that coalescing around a shared framework like the group chat I describe is the most effective way to advance human alignment with businesses that typically have different agendas.