Remembering something can make you forget something else. Your brain does it on purpose.

That sounds backwards. Retrieval practice, pulling an answer out of your head instead of rereading it, is one of the best-studied learning habits there is. But in 1994 three psychologists at UCLA found a catch hiding inside it. Once you know about it, you start seeing it in how most of us learn our tools.

Fruit, orange, and the banana that went missing

Michael Anderson, Robert Bjork and Elizabeth Bjork gave students category-word pairs to study. Fruit–Orange, Fruit–Banana, Drinks–Scotch. Eight categories, six words each.

Then came practice. For half of the categories, students were quizzed on half of the words, three times each, with a cue like "Fruit – Or___". The other categories got no practice at all.

After 20 minutes on an unrelated task, everyone got a surprise test. Here's a category name. Write down every word you studied for it.

In the first experiment, with 36 introductory psychology students, practiced words were recalled 73.6% of the time. Words from categories nobody practiced came in at 48.4%. No surprise there.

The surprise was the third group: unpracticed words from practiced categories. Banana, when you'd been drilled on orange. Those fell to 37.5%. Below the baseline. Practicing orange made banana harder to find than if the students had practiced nothing at all.

They called it retrieval-induced forgetting. The title of the paper says it more bluntly: "Remembering can cause forgetting."

Two details matter. The effect hit strong, common category members, the ones most likely to jump in and compete, much harder than rare ones. And in a 2000 follow-up, the team showed the forgetting appeared only when people had to retrieve the practiced words. Seeing them again the same number of times made them stronger too, but left the neighbours alone.

Why would a brain do that?

Anderson's explanation is that it's a feature. When you reach for "orange", banana and apple light up as well and get in the way. To land the right answer, something has to push the competitors down. Do that a few times and they stay a little suppressed afterwards. Less noise next time you want orange.

In 2015, Maria Wimber and colleagues, working with Anderson at the MRC Cognition and Brain Sciences Unit in Cambridge, tried to watch this happen in a scanner. People learned two pictures for each cue word (Marilyn Monroe and a hat, say), then repeatedly recalled just the first one. With each retrieval, the brain pattern for the competing picture faded in visual cortex. People with more activity in prefrontal regions linked to resolving competition showed more of that fading. And the more a picture's pattern was suppressed, the more likely it was to be forgotten on a later test (Nature Neuroscience).

Now the honest part. In 2018 Kevin Potter, Lucas Huszar and David Huber at UMass Amherst reanalysed that data and found the key result hinged on how missed responses were scored. They ran the behavioural task twice more and found no forgetting at all (Cortex). So the brain-scan story is contested. Inhibition is the leading explanation, not a closed case.

The basic effect is on firmer ground. A 2014 meta-analysis in Psychological Bulletin by Kou Murayama, Toshiya Miyatsu, Dorothy Buchli and Benjamin Storm pooled 512 samples and found a reliable but modest effect (g = 0.35). Big enough to notice. Small enough that nobody should throw out their flashcards.

Glowing cyan neural pathway thickening while neighbouring branches dim to violet

Where developers run into it

Think about how we learn a tool. You hit a problem, search, find the one command that fixes it, and then use it forty times. git reset --hard. git stash. The same three flags on docker run.

Every time you reach for the familiar one, its siblings are the competitors. git restore, git revert and git switch live in the same mental drawer, and the one you keep pulling out may be making the drawer stickier for the rest. I can't point to a study on git commands specifically, so treat this as my reading. But the setup is close to the lab version: one category, a few strong members, repeated retrieval of only some of them.

It can even spread between people. William Hirst's group at the New School found that listeners show the same forgetting when they follow along closely while someone else selectively recalls (Cuc, Koppel & Hirst, 2007). In their third experiment it showed up in ordinary, free-flowing conversation. The teammate who always answers "just use X" might be quietly shrinking your memory of Y.

The fix is in the same research

Here's the part I like. Retrieval doesn't always hurt the neighbours.

Jason Chan, Kathleen McDermott and Henry Roediger at Washington University in St. Louis tested this with prose passages, the connected kind of material you actually study, across three experiments with 210 people. Being tested on some facts improved recall of related facts that were never tested, 24 hours later. Extra study without a test didn't do that (Chan, McDermott & Roediger, 2006). They called it retrieval-induced facilitation, and their third experiment showed that people's strategies could switch it on or off.

The Wimber paper has a telling line in its methods. Participants were told not to link the two pictures for each cue into one mental image, because integration between competing memories is known to limit this kind of forgetting. In other words, the researchers had to keep the memories separate to make the effect appear.

Isolated items in a list compete. Things that hang together can pull each other up.

Watercolour constellation of stars linked by thin gold threads into one connected web

Try this today

Pick a tool you use every day and write down, from memory, the commands you almost never use. Skip the one you always type. Go for the siblings. For git, that could be: what's the difference between reset, revert and restore? Answer it closed-book, then check.

When you recall something, recall it with its neighbours. Ask why this one and not that one. git revert makes a new commit that undoes an old one. git reset moves the branch pointer. Recalling the contrast turns rivals into a connected set.

And quiz yourself across the whole topic now and then, not just the bit you needed today. Spread the retrieval around so no single answer gets to crown itself king of the drawer.

Where Recl fits

This is part of why I built recall in Recl around whole saves rather than loose facts. You save a talk, an article or a post, it pulls out the key takeaways, and later it brings that save back as a quick question, so you're retrieving an idea with its context attached. Some of the public summaries are on library.recl.app.

Recl screen showing a saved video with its summary, key takeaways and a Recall tab

Your memory is always deciding what to keep within reach. Give it the whole drawer to work with, not just the thing you grabbed last time.

I'm Asutosh. I write about how the brain learns and remembers, and I build Recl, a tool for remembering what you watch and read.

Sources