PhD thesis: Weakening memories by half-remembering them

[Back to Research]

For my thesis, I worked on a series of behavioral and fMRI memory experiments to understand a little more about forgetting, called ‘Weakening memories by half-remembering them’ (PDF).

Here’s the gist.

The act of remembering something – dredging it up from your memory and letting it blossom fully into a recollection – is one of the best ways to strengthen a memory.

We were interested in what happens when you only half-remember something, i.e. when a memory activates in your mind but doesn’t quite wake up completely.

We had a theory (from our earlier work with neural networks, and from other researchers) that memories that only partially activate will get weakened, i.e. forgotten just a little bit.

We figured this could be happening all the time in the real world, e.g.:

  • When you’re driving quickly past a billboard advertising milkshakes, say, and it half-reminds you that you’re supposed to pick up the milk on your way home, but then you get distracted by a red light. The ‘pick up the milk’ memory didn’t quite get a chance to activate fully before you moved on, and we think you’ll actually be less likely to remember it now than if you’d never been half-reminded of it.
  • When you’re failing to recall a weak memory because a stronger but irrelevant memory is getting in the way. Where did I park my car this morning? I have no idea. All I can think about is where I parked it last week, which is no use at all! The weak memory (today’s parking spot) being occluded is only partially activated, and will now be even harder to remember later.
  • When you’re focusing on a book and doing a good job of ignoring an audible conversation in the background. Related memories half-triggered by that conversation were activating in your mind, but only a very little bit.
  • When a name or a word is on the tip of your tongue, and you’re straining to poop it out of your mouth. The name or word you can’t quite remember is only partially activated.

The common theme here is that there are memories that are triggered but don’t have the juice to fully wake up in your mind, and as a result, will be even harder to remember later.

Initially, we set out to design experiments to capture these kind of real world phenomena in the lab.

  1. We’d start by teaching people some simple paired associations: whenever you see this word, think of this image.
  2. Then (and this is the key part) we’d try and partially activate some but not all of those paired memories.
  3. Finally, we’d test to see how well you remember all of the paired associations.

Our hope was that you’d remember the paired associations that had been partially activated less well than the ones that had played no role in the middle section of the experiment.

We tried dozens of versions of this, but it turned out to be very hard to get the experiments quite right. A lot of the time, the memories seemed to be activating much too much or too little. It was hard to activate them at just the goldilocks level we thought would cause forgetting.

So we tried something a little more ninja.

We put people in fMRI brain scanners, and used a special analysis we’d helped develop (called MVPA) to try and measure how strongly a memory was activating based on patterns of neural activity. We could then make the prediction that the memories that had activated a little but not much would be harder to remember than memories that hadn’t activated at all.

Sure enough, we found that:

  1. memories that activated highly (according to our brain scan memory-o-meter) were strengthened.
  2. memories that did not activate at all were unaffected.
  3. memories that activated partially were weakened.

1) and 2) are pretty uncontroversial. It’s 3) that’s interesting.

This isn’t the first study to make this claim, nor even the first from our lab. But it was the first to use fMRI brain scanning to provide a ‘covert, neural measure of a memory’s activation in this kind of experiment, and thus to predict whether the memory would be easier or harder to remember later on’.

In other words, we could predict whether you were more or less likely to remember something, just based on how strongly the memory had activated when we reminded you of it. This provided support for the theory that partially activating a memory could weaken it.

In the long run, this could inform therapies for patients suffering from post-traumatic stress disorder or phobias (though a lot more work would be needed). Here’s the plan: ever-so-gently trigger the painful memories, but distract or interfere with those memories so that they don’t fully activate, weakening them little by little. This isn’t so different from some existing treatments, but we could potentially use fMRI to know how much to activate the offending memories, and to back off if they’re getting too active before they get counter-productively strengthened.

This work will be published later this year - Detre, Natarajan, Gershman & Norman (submitted), after massive extra work on the analyses from Malai, Sam and Ken.

I’ve simplified a lot here. Chapter 1 of the dissertation provides a much better background on the theory. Chapter 2 describes some of the behavioral experiments that didn’t quite work, and tries to understand why. Chapter 4 covers the final fMRI experiment, especially the methods. However, the analysis has been considerably revamped since then, so drop me a line if you’d like to read our submitted paper.

Huge and heartfelt thanks to Ken Norman, Malai and Sam, the members of my reading and oral committees, and everyone in the lab and wider scientific community who helped.