By Samuel Mendez, PhD, Lecturer, Graduate Research School, The University of Western Australia
In this edition of Health Literacy Voices, Dr. Mendez takes a deep dive into the audience Q&A from his recent IHA webinar, “AI Literacy for Health Literacy Professionals.”
Can Generative AI Help Patients Prepare for Conversations With Their Provider?
There is still a lack of published research on the concrete benefits, if any, of using generative AI to prepare for medical visits. But that doesn’t make people’s personal experiences any less important. If someone is feeling more prepared or more confident, you can’t really deny that. You might want them to explore whether they could get that from talking with a loved one, meeting with a patient navigator, or using a more secure tool. But that discussion may never happen if you deny their experiences.
And it seems like many people may share these experiences. A Kaiser Family Foundation poll found about 3 in 10 adults use AI as a source of health information at least monthly. Only about one-third of users reported that they usually followed up with a doctor. One reason might be that they don’t see the need to, as users tended to be confident in their own skills. Another reason might be cost, as about 1 in 5 users reported using AI because they couldn’t afford access to care.
It’s important that your patients know popular tools like ChatGPT are not private. Their chats can be used to train models. Even if users toggle a setting so particular chats aren’t used that way, interactions might still be stored and analyzed for marketing purposes, refining the user interface, etc. It really depends on the fine print, which can change suddenly and quietly.
Beyond privacy concerns, the need for critical health literacy skills still applies. Strategies you might use to help people parse through health messages on social media are still helpful.
If Patients Are Uncomfortable With Their Providers Using Generative AI, How Do We Address That?
Can we guarantee that their health information is kept private?
I think it’s worth finding ways to honor patients’ privacy concerns, especially while trust in health institutions remains a hot-button issue. This might mean using a different workflow for specific patients. This might mean proactively raising questions about their comfort level with your use of generative AI tools. It might also mean learning more about patients’ own use of generative AI.
As a health professional, it’s important to be able to have an informed discussion. Your patients might need answers about data privacy, factual errors, or perceived benefits before they can tell you how they feel about the tools you use. You might need to ask IT or other experts in your organization for details specific to your context. Generally speaking, you can use HIPAA compliance as a familiar standard.
Some generative AI tools are part of electronic health record systems. For example, Epic has incorporated generative AI tools for several years now. Tools that are directly integrated into clinical software like this should be HIPAA-compliant. So, data privacy should be on the same level as the rest of a patient’s EHR data. That said, we have seen EHR breaches in the past, so you might not be able to “guarantee” anything exactly.
Outside of these cases where HIPAA compliance is mandatory, you can guarantee even less. The way a tech company treats your patient’s data is up to an app’s terms of service, which can change. If your patient disagrees with data use down the line, they might not be able to erase its influence on future models.
How Do We Weigh the Health Equity Impacts of Data Centers Against the Potential Benefits of Generative AI?
There are health equity issues around the impacts of data centers on public health and the environment. How do we square that with the potential benefits of generative AI in health literacy?
I think it’s good practice to define and measure potential benefits for yourself. Do you want a specific task to take less time? Do you want your writing to more consistently meet health literacy standards? Do you want to think less about the first draft of a message? This will help you define the tradeoffs you’re considering and take stock of what benefits a particular workflow is actually giving you. This will also make it easier to explore your options. And we do have options!
Generative AI does not need to be part of your workflow on a particular task.
It’s always worth mentioning that generative AI does not need to be part of your workflow on a particular task.
Even if you have identified a very important use case for yourself, we’re not stuck with the tech industry status quo. For example, you can download models on your own computer and run them through a program like GPT4All. Likely, it will be slower than using Claude or ChatGPT. The outputs might not be as polished. And yet it still might bring enough of the benefits you’re seeking, without the need to interact with power-hungry industrial infrastructure.
Organizational leaders can experiment with hosting local models and providing access through an intranet. They might also make policy decisions to define what types of problems are worth grappling with environmental justice tradeoffs to solve. It will take a lot of creativity, experimentation, and action on our part. As people who care about health literacy and health equity, we know we can’t count on the status quo to best serve the people who may need it most. So we have to start answering questions like these for ourselves, with the communities we want to serve.
About the Author
Samuel Mendez, PhD, is a lecturer at the Graduate Research School at The University of Western Australia. His research bridges media studies and computational social science to improve health communication and examine the limits of technology in addressing the social dimensions of health care communication.
Join the Conversation
What stood out to you? Share your thoughts or questions in the comments below.