The Future of Healthcare and Generative AI: Navigating Challenges and Possibilities

Image credits: Emiliano Vittoriosi via Unsplash.

Generative AI is all over the news these days. What’s the outlook for AI in healthcare?

It’s a fast-moving and complex innovation space that touches on an important part of the human experience. Rather than rehash the many articles available online, I’d like to share two vignettes about my personal experiences with generative AI in the last month.

I initially explored ChatGPT for entertainment-focused content

Back in December, my early experiments with ChatGPT were mostly entertainment focused. “Please write me a Shakespearean sonnet about DNA.” “I need a limerick about basketball in the voice of a pirate.” Not much different from about a hundred million other users.

But can ChatGPT write up the results of a clinical trial?

My usage changed when my 84-year-old father asked me if ChatGPT could write up the results of a clinical trial. I didn’t know, so we fired up my web browser. Since we didn’t have novel clinical trial data to feed ChatGPT, we tried to reverse engineer a clinical trial report from a peer-reviewed article in the scientific literature. 

Here are the highlights of what happened next:

  • ChatGPT correctly identified the components of a clinical trial report, but I failed to find a prompt that led to that format as the final output. The bot knew what data I wanted, but the output wasn’t the fully digested form that I sought.

  • ChatGPT located the source paper in the scientific literature, but it egregiously mishandled related references. Making up fake articles is a known type of AI hallucination, so I was on the alert. But the fake citations were surprisingly tricky to spot. Sometimes the first few author names would be correct, but the year of the “paper” would be wrong. Or the year was right, but the authors didn’t actually work on the scientific area that was the subject of the fake citation. If I hadn’t been looking closely and been willing to turn to Google Scholar, I would have been fooled.

  • In ChatGPT’s reconstituted abstract about the clinical trial data in the original paper, the results that it identified as being among the most important in the study were not the same results that were highlighted by the authors of the published study. There seemed to be a judgment call at play, over which humans and the AI differed. Since I consider interpretation to be the key function of a scientist, I was particularly intrigued by this gap.

And can ChatGPT interpret my blood metabolomics test results?

Dr. Tiffany Vora shares two experiences with using ChatGPT in a medical context.

Two months later, I got the results back from a blood metabolomics kit that I had ordered from a startup. I looked at the summary report and asked the classic question: what am I supposed to do with all this information?

To my surprise, the answer was “ask the chatbot on our app.” I didn’t have a sense of what I could and couldn’t ask it, so I just asked it … everything. After some trial and error, I realized that I could ask it to stop telling me things that were trivial, annoying, or off base (providers, take note: don’t tell a stressed person that she needs to be less stressed; she already knows). I could tell it my dietary restrictions and ask it to only give me insights consistent with them. And I could tell it to “act like a clinical nutritionist” when it was generating its suggestions. This chatbot was also not great with the scientific literature, but I was on the lookout for those shenanigans. I ended up emailing the company’s CEO asking for a “how to” guide for generating useful and reliable prompts for their system.

How can we get better at using AI for healthcare and everyday life?

In both of these experiences, my primary struggle wasn’t medical: it was in learning how to use generative AI in the most effective way. I’ve been dedicating time each week to working on my prompts for various tasks for which I would like to use generative AI, from research writing to translation of science to image generation (my son's head almost exploded when I told him I had a Discord account). I’m getting better each day.

We're still early at unlocking AI's potential in healthcare

Dr. Tiffany Vora delivering one of her talks on the future of healthcare.

But I feel intense frustration knowing that there is so much AI-enabled territory that I’m missing. I want to be at peak AI! Right now! For everything! More recently, I’ve taken a 1-hour online course on prompt engineering for developers. I’ve picked up some nifty insights, like using text delimiters, asking for structured output (something that I’d failed at earlier), providing successful examples, and strategies for avoiding hallucinations. It’s funny that I would organically use those approaches when talking to a human, but it didn’t occur to me to use them when “talking” to an AI!

How generative AI can help improve healthcare

It should come as no surprise that the broader public seems to have mixed feelings about the use of AI in healthcare in general. In a recent Pew Research Center study, surveyed Americans tended to think that AI would help reduce medical errors and perhaps improve embedded racial and social inequalities in medicine. On the other hand—remarkably—only 13% of respondents said that they think that AI will improve the patient-provider relationship.

AI in healthcare is not about replacing people. It can help healthcare providers reduce medical errors, avoid burnout, and more. (Image credits: lenetstan).

I find that result surprising because many of today’s generative AI use cases in medicine revolve around administrative tasks: filling out electronic health records, writing letters to insurance companies, generating and making sense of clinical notes. These were exactly the tasks pointed out to me when I asked an Indian hospital’s doctors and nurses what they would change if they had a magic wand. They are also major elements of today’s epidemic of burnout in healthcare careers.  I suspect that the Pew’s study was highlighting the larger issue of patient trust, which is foundational in healthcare. How do we balance trust with productivity?

I’m excited by the prospect of my physician looking at me, instead of her monitor, as we’re talking. She could be using touch to assess me and to provide comfort, rather than typing away on her keyboard. In my mind, generative AI has the potential to bring the spotlight back to the human relationships that are at the core of medicine.

That’s why I’m learning to do my part as a patient in this new AI-enabled world. What are you doing to harness the potential of generative AI?


About Tiffany

Dr. Tiffany Vora speaks, writes, and advises on how to harness technology to build the best possible future(s). She is an expert in biotech, health, & innovation.

For a full list of topics and ways to collaborate, visit Tiffany’s Work Together webpage.

Follow Tiffany on LinkedIn, Instagram, and X.


Donate = Impact

After a 19-day voyage to Antarctica aboard The Island Sky in November 2023, Tiffany has many remarkable stories to share & a wealth of insights to catalyze a sustainable future.

You can support her ongoing journey by making a contribution through her donation page. Your support will spread positive impact around the world, empower Tiffany to protect time for impact-focused projects, and support logistical costs for pro bono events with students & nonprofits.



And if you want to explore the topic further, visit the Collections page of my website for articles, podcasts, and videos.
For keynote speaking or long-term projects on biotech, health, and innovation, send me a message. Let’s talk!


Want to dive even deeper? Check out these great reads!

Disclosure: I am an affiliate of Bookshop.org where your purchases support local bookstores. I will earn a commission if you click through and make a purchase.

Previous
Previous

Book Recommendation: The Art of Impossible by Steven Kotler

Next
Next

Are We Prepared for Radically Longer and Healthier Lives?