Skip to main content

Personal Health Assistants

BINF4070, with Prof. Yuhai “Orson” Xu at CS/DBMI and Prof. Eugene Wu at CS.

Here’s the syllabus/outline and the Course’s Github project.

Personal Health Assistants​

We start with Personal Health Assistants (PHAs) that predate Personal Health Agents (and their use of LLMs) in HCI circles. Here’s the course definition:

A Personal Health Assistant (PHA) combines authorized personal health data and computational tools to help a person inspect patterns, ask questions, and prepare decisions. Its claims remain traceable to evidence and bounded by intended use.

The assistant is what the user works with. They don’t care how it works: we do that with agents but it could be a simple RL model that backs the assistant as well. Think pre-GPT chatbots. Agents are merely an implementation detail.

Desirables/Properties​

PHAs are

  • Personalized to the user using them.
    It’s in the name.
  • Dynamic with respect to the users health data.
    No point in noting that you were as healthy as an ox ten years ago.
  • Inspectable
    Show your reasoning. Be transparent. Especially in case something goes wrong.
  • Safe — “Do No Harm” is of vital importance here.
    Further, privacy and security need to be woven into yoru solution and are not secondary/last-minute/auxiliary concerns.
  • Aligned with users
    Who are they? What do they want? What are their values? Their concerns? What is their lived experience? Center them above all else.