A realistic patient use case for AI in healthcare

Alice is a 47 y/o female with a history of urinary tract infections (UTIs) and she’s been seen by different physicians under different health plans for this and other reasons. Her current physician has prescribed an antibiotic to be taken as needed. Alice keeps a digital diary of her symptoms, clinician messages, and other health-related events on her smartphone. This simple diary complements the information in her various electronic health records (EHRs).

When a new symptom appears, Alice updates her diary and uses a health navigator service to help her decide whether to take her prescribed antibiotic or contact her physician. The navigator feeds her diary and all of the information from her EHRs, consolidated by her personal digital agent, into the GPT 4 large language model (LLM). With appropriate prompts by the navigator, the LLM creates a timeline showing Alice’s symptoms, prescriptions and self-administered treatments. A link to the updated timeline is added to the diary on her phone. If Alice decides to contact her physician, she can choose to send the timeline by simply adding the link to her message.

The personal digital agent is key to the consolidation of Alice’s healthcare data into a usable form

Personal data like health records or message threads are stored in hundreds of different apps and services so a standard for connectivity is essential to making personal AI practical. A personal digital agent connects private personal AI to a LLM in the cloud. Trustee®, developed by HIE of One, is open source for transparency and customization, but is also standards-based to promote best practices and competition among would-be navigators. Customization of the personal digital agent also benefits from standardization, as when a common language encourages shared learning and trust across communities of users.

A personal digital agent protocol (PDAP) standard serves every participant in the value chain for expert advice and navigating complex situations. For Alice, it makes personalized expert advice more accessible. For her physician, it reduces the time and cost of assessing large amounts of personal data by sharing that job with a LLM. For the LLM operator, a standard reduces the regulatory liability by introducing a professional…in this case, Alice’s physician…as the fiduciary.

A PDAP standard considers three principals (subject, fiduciary, and LLM) with clear separation of their technical and business roles. Reality is more complex. Each subject depends on dozens of service providers and registries to store personal data. Fiduciaries depend on technical systems for messaging, scheduling, record retention, billing, client service, and access to reference materials not available to the LLM. The LLM operator, likely a top-50 corporate giant or a nation-state, has complex capital and business systems of their own. Scoping and chartering a PDAP standard benefits from participation by all three of the principal roles and consideration of the businesses that will support them.

Toward a practical personal digital agent

In theory, Alice has control of her personal data stores and some insights into how to manage access to those data stores. In practice, exercising control over our personal data across messaging, social media, health, finance, and more is tedious, and trusted advice on how to manage it is hard to come by. Professional fiduciaries like doctors, lawyers, and money managers are expensive, if they’re accessible at all. As a result, people frequently seek advice and support from friends, family, and online communities. 

Making expert consultation safe, effective, and accessible will require a new kind of online community that leverages: 

• technology for digital agency like IETF GNAP 

• simplified management of the policies that control one’s digital agent

• trusted, low-cost hosting and support for the agent

• cost-effective, privacy-preserving access to the best LLMs

For almost a decade, the volunteers at HIE of One developing Trustee^® and related open-source demonstrations have kept pace with the evolving technologies and regulations for access to personal data. As personal AI and generalist LLMs become accessible, we are using the IETF process for standards development. Our goal is to empower people in their personal and professional roles by supporting all kinds of communities, including cooperatives, not-for-profits, decentralized autonomous organizations, and for-profit structures.

You can help advance personal digital agency by commenting on this post, joining the pdap mail list at https://www.ietf.org/mailman/listinfo/pdap or emailing info@trustee.ai to participate in the biweekly HIE of One community calls.