Using 50,000 Papers to Build MedEdMentor's Theory Suggester

One of the biggest challenges for medical education researchers is selecting an appropriate theory to frame their research project.

Often, this crucial step is either skipped or researchers simply choose from the first few theories they encounter. Of course, researchers can and do ask mentors for help, but this too is imperfect. Conversations with expert mentors reveal that they often have a specific paradigm and set of theories that they regularly employ and may not always be equipped to operate outside of these bounds.

We started with MedEdMentor AI on ChatGPT, and we're thrilled with its initial abilities to recommend theories to frame a research phenomenon.

But we ran into a few problems: it's only available to ChatGPT Plus subscribers, and while the suggestions are surprisingly relevant, none of them are necessarily grounded in literature.

So today we're launching Theory Suggester to suggest theories directly from the literature. (It's free while in testing)

We believe that large language models (LLMs) have many strengths when it comes to suggesting theories for medical education research projects.

Asking an LLM for suggestions, especially MedEdMentor on ChatGPT, returns highly creative suggestions, and we've shown that there is strong alignment with what humans choose for a given research phenomenon. Additionally, you can ask follow up questions to clarify any unfamiliar theories.

But these LLMs are limited by hallucinations.

The LLM either won't have the most recent publication data, or cannot reliably give you a truthful answer.

To solve these problems, we built an complementary tool: Theory Suggester. Not only are all of its suggestions straight from the literature, it's hosted right on MedEdMentor so it's available even if you don't have a ChatGPT Plus subscription.

All you have to do is put in your research phenomenon, and you get back a list of tailored theory suggestions.

For example, using the example of teaching medical students oral presentation skills, we can immediately see that experiential learning, rhetorical theory (persuasion and speeches), and genre theory (learning a specific type of presentation) have been used multiple times before.

Furthermore you can see the number of papers that use that theory. However one caveat is that without directly reviewing the papers, some of the results (especially lower down) may not be true examples.

We'll work on displaying the actual papers for you in a future update.

At MedEdMentor, we aim to augment — not replace — human mentorship.

We envision researchers starting with Theory Suggester to get a sense of what theories have been used before, and then moving to MedEdMentor AI on ChatGPT to see what other creative theories an LLM can come up with.

Next, we recommend following our new Lesson about selecting a theory to start going deeper on each of the theory suggestions. Finally, this would be a great time to approach a mentor with the work you've done so far to get additional input.

If you don't have a mentor available, later this year we'll be working on helping community members find the help they need.

We're hopeful that Theory Suggester will lead to more diverse and innovative research.

Not only can researchers now find the most commonly used theories to investigate their phenomenon, they can also find rarely used theories, and (by process of elimination) find theories that were never used before!

And for the researchers who don't have easy access to mentors, this tool offers a more theoretically sound starting point for your medical education adventures.