Not All Friction Is Good Friction

Learning science has a well-studied name for good friction: "desirable difficulties," a term coined by Robert Bjork in 1994. Conditions that make it look like we are learning quickly (practicing the same thing over and over, in one sitting) often don't stick, while conditions that slow us down and feel harder in the moment (spacing practice out over time, or testing ourselves instead of rereading) tend to produce learning that lasts. And because the harder version feels worse, learners and teachers alike tend to misjudge what is working and choose the easier path.

Robert and Elizabeth Bjork are careful to point out that "the word desirable is important." A difficulty becomes undesirable when the learner doesn't have the background knowledge or skills to respond to it successfully, which means that whether friction is productive depends a lot on who is doing the work.

This is one of the questions our field is struggling with most right now. Medical training was the most rigorous learning environment I have ever been in, and I perceive the knowledge and skills I gained to be proportional to the effort I put in. But AI is changing everything right now, including medical education. What do medical students actually need to know? What do they need to memorize? Which activities will produce the learning they need to function in an AI-infused clinical environment, and which ones are now antiquated? Put in terms of friction, which hard parts are still the learning, and which hard parts has AI made unnecessary?

I have a lot of thoughts about what medical school should look like in an environment with ubiquitous AI availability, but I will hold those for another post. What do you all think? I’d love to include your ideas and questions alongside my own. Email me! (geoffstetson@mededmentor.org)

The same question, “what is productive friction?”, lay underneath the symposium on AI and academic writing at AMEE 2026 that I wrote about in my last post. Dr. Lorelei Lingard's argument that "writing is thinking" is, at its core, an argument about friction: writing is hard, it takes a lot of cognitive energy, and that effort may be what produces understanding. Yu-Che Chang, another panelist, put it more simply (I am paraphrasing from memory): we can all drive to work, but some of us choose to walk or bike because it is good for our health. We could let AI do our writing, but doing it ourselves brings benefits precisely because it is harder.

Scholarly work is a little different from training, though, because the difficulty is rarely designed by a teacher, and the goal is a finished piece of work as much as our own learning. So I am stretching the Bjorks' idea here, but I think a similar test applies: is this the hard part I actually need to do myself in order to learn and grow?

Take the literature search. For me, and for most of the work I do, running a literature search is unproductive friction, which is part of why the literature search in MedEdMentor is so amazing for me (it finds the most relevant articles in my field quickly and efficiently). But I am not a librarian, and I have never written a scoping or systematic review. For someone who is doing that kind of work, the literature search is the work, and doing it themselves is productive friction.

The line can also move for the same person over time. For a long time, writing felt like unproductive friction for me; it was really, really hard, and I was under the impression that I would never be a good writer. Then, back in 2019, I made a friend over Twitter, Dr. Jennifer Spicer, an infectious diseases physician and passionate medical educator who worked at Emory University at the time but has since taken an awesome new job at the University of Colorado Anschutz School of Medicine. Together we started the MedEdTwagTeam (a wild portmanteau of Medical Education Twitter Tag Team). Every week we published "tweetorials," strings of 10 to 15 tweets, each with an image, sharing faculty development materials about teaching in the clinical learning environment. We grew an audience of more than 10,000 people and had a lot of fun doing it, until Twitter became X (womp-womp).

The format gave me a few things: it was a topic I was passionate about, I had control over the medium and could write in my own voice, and the mix of words and images worked really well for the way my brain works. What felt unproductive started to feel productive, and that gave me the confidence and skills to write in other venues, both non-traditional and traditional. All of that writing was fundamental to me learning and internalizing the foundational ideas of HPE.

This is a key principle for us at MedEdMentor. The coaching stance we have given Uni, our AI mentor, is that it will not run off and do your work for you. Whether you type or dictate, it makes you communicate clearly about your project and your ideas, and think hard about them, before it offers suggestions or a path forward. Our whole point is to retain the productive friction, reduce the unproductive friction, and help people go deeper, not faster.

For example, while testing Uni for our Dust Off Your Projects workshop, I played a hospitalist with a stalled peer observation project. When I said my first goal should be getting IRB approval, Uni said, "I want to push back gently on the order," and asked, "What do you want this project to become?" before it would accept my proposed goal. And by the end of our conversation every line of the plan was in my own words. At the same time, when I asked Uni to find the relevant literature, it did a wonderful job on its own (which, for me, is exactly the friction I want removed).

I would love your thoughts on both versions of the question:

1. Which hard parts do our learners still need to do themselves?
2. Which parts of your own scholarly work are friction you need, versus friction you can hand off without losing much?

Write to me at geoffstetson@mededmentor.org.

Thanks for reading, and I'll see you back here next Monday!