AI Literacy Is Not a Pamphlet

Chris Dymek started in philosophy. She passed her doctoral exams, looked at friends taking one-year appointments in places they didn’t want to live, and moved into computer science instead. Programmer, systems analyst, director of re-engineering at a utility, a doctorate in organizational change, CTO of a small health organization in Tucson, then EHR implementation in Maryland as the HITECH money landed. Research at NORC and Westat. Data infrastructure at ASPE. Finally, Director of Digital Healthcare Research at AHRQ.

The philosophy never left. She paraphrases Wittgenstein, and flags it as a paraphrase: whatever can be thought can be said, and whatever can be said can be said clearly. Her gloss is that digitizing anything forces exactly that discipline, and how you say a thing shapes how you implement it.

Knowing that, knowing how

Her framework splits AI literacy in two. Knowing-that is what a classroom gives you: the difference between a large language model and other predictive modeling, what agentic AI means, that hallucinations happen, that bias and ethical implications are real. Knowing-how is the practiced ability to use any of it.

“I may know that I have to reduce debt and increase savings,” she said, “but to know how to do that is a very different matter.”

She learned it on herself. Classroom first, then prompts, and only through writing prompts against her own real questions did she get any good at it. Read, practice, read again, practice again.

Ask what goes wrong for a clinician with the that but not the how and she doesn’t hedge: “They can be duped.” You cannot audit a system you have never operated.

Why now

The sharpest passage is about her family. Her mother worked for Chrysler, belonged to the UAW, and never had to think about financial literacy, because a pension and Social Security did the thinking. Her daughter has to be very concerned about it. Nothing about money changed. The scaffolding disappeared.

AI literacy, she argues, is the same kind of response. As she describes the landscape: some states are trying things, the FDA reviews AI-enabled devices, and past that there isn’t broad-based regulation. (The nearest thing to a counterexample is narrow and currently contested. ONC’s HTI-1 rule makes developers of certified health IT disclose how their predictive decision support was built, and a pending proposal would strip that back out.) In a federal floor that thin, a workforce that can recognize risk in the tools it uses is not a nice-to-have. “They’re your first line of defense.”

Guardrails and strikes

Which is only half of it. Not many people, she notes, are getting a return on their AI investments, in healthcare or outside it. She reads that as a literacy problem more than a technology problem, and points to what a literate staff does: explore, form tiger teams, re-engineer the work. Her picture for this is a bowling alley. There are guardrails along the sides and there are pins at the end. You want both. As Leon put it back to her, the bumpers are what let you bowl with confidence.

Her model has three constituencies (organization, professional, patient and caregiver) and, for patients, a five-rung ladder: knowing-that, functional, critical, participatory, and finally empowered partnership, where a patient co-creates care using AI-supported information. That top rung draws on Hugo Campos, a previous guest on this show. It is not theoretical for her. A friend with stage-four breast cancer uses large language models to shape the questions she brings to her doctor, and Chris describes what it gave her: great agency.

Monday morning

Tie AI literacy to a strategic goal. Work out the competencies that goal requires. Build the learning platforms. Find the digital-literacy efforts already running that you can attach to. Then measure, collect data against the measures, and adjust: the unglamorous learning-health-system loop. AHRQ spent roughly fifteen years building the evidence that health literacy moves patient safety and outcomes, and producing the toolkits to act on it. Chris wants that same playbook run again for AI.

The barrier she names is not budget. It is trust. People have sometimes walked out of commencement addresses that mention AI and the future of work. An organization that treats literacy as content delivery will fail on trust it never thought to measure.

Listen to the full conversation: https://practicalaiinhealthcare.com/episodes/#dymek

You Might Also Enjoy

• S1E41 with Hugo Campos — The patient advocate Chris credits for the top rung of her ladder, building his own tools for patient-directed AI.
• S1E30 with Amy Price — Participatory medicine and co-production, the practice Chris’s “empowered partnership” level describes.
• S1E42 with Fred Bennett — Health literacy at the point of the doctor-patient conversation, the adjacent literacy problem AI now sits on top of.