When I was very young, my family visited Disney World, and for years after, I had a fairly vivid memory of the Dumbo ride: the elephants spun around vertically like a Ferris wheel. When I returned there decades later as an adult with my own family, I was stunned to see that the elephants spin around like a merry-go-around, also moving up and down. The Ferris wheel was a false memory, a lie my brain told me. But why?
This is exactly what researchers like Sophie Siestrup of the University of Münster are seeking to understand: Why are we sometimes tricked by our own brain to remember something differently than it actually was? In a new study published in the Journal of Cognitive Neuroscience that looks at how episodic memory might change over time, Siestrup and colleagues mapped out the role of different types of “prediction errors” in the brain.
I talked with Siestrup about the new work, including the novel “toy stories” the researchers used to track changes in episodic memories, and what the future of this work may hold.
CNS: How did you become personally interested in this research area?
Siestrup: I joined the field of episodic memory research as a little bit of an outsider. I am a trained biologist with a background in behavioral biology. However, I was always interested in understanding the human brain, something that cannot be achieved with mere behavioral observations. When I learned about this huge interdisciplinary project on episodic memory, I decided to give my career a little twist and was lucky enough to be invited to join.
Changes in episodic memory got me hooked from the start because this topic relates so well to our experiences in everyday life. I think we all know these moments when we are sure to remember something well, let’s say the wonderful cheesecake that our mother baked for our last birthday, but are surprised to find out, for example by seeing a photo, that we actually had chocolate cake.
CNS: What was the insight you were seeking with this new study?
Siestrup: In previous research, we and others already demonstrated that mild surprises about memory contents, so-called mnemonic prediction errors, can lead to changes in episodic memories. But not much is known about how this is actually accomplished by the brain and which factors determine whether a memory will be changed or not. For this reason, we investigated how different types of mnemonic prediction errors influence brain activity and episodic memories. We expected that not all prediction errors will influence memories in the same way, but that the degree to which they can influence memories might depend on the relevance of the surprise, or prediction error, for the episode. Thus, in the current paper, we investigated two types of prediction errors: some that were irrelevant for the memorized episode (surface-level prediction errors) and some that challenged their entire storyline (gist-level prediction errors).
CNS: Can you explain a bit more what “mnemonic prediction errors” are?
Siestrup: The idea of a mnemonic prediction error comes from the so-called predictive coding framework. According to this, the brain constantly predicts what is going to happen next on the basis of previous experiences. For example, when we pass by a bright yellow car parked on the roadside on the way to work on Monday, Tuesday, Wednesday and Thursday, our brain will expect to see that car on Friday as well. However, when we unexpectedly do not see the same car on Friday, there will be a mismatch between prediction and reality, i.e., we experience a prediction error. Such prediction errors are highly important for learning because they signal that there is some new, interesting information that our brain did not know yet. As a consequence, it can also happen that a memory is “updated” with new knowledge that was not yet present at the time the memory was originally encoded. Importantly, such prediction errors are often not consciously experienced. This means that even if we personally do not feel surprised, our brain may.
CNS: What were you most excited to find?
Siestrup: The most exciting finding was that while surface-level prediction errors led to substantially weaker brain responses than gist-level prediction errors, only surface-level prediction errors gave rise to changes in episodic memories. Additionally, it was a little bit surprising that gist-level prediction errors did not lead to memory changes at all: When prediction errors challenge the storyline of an episode, the brain seems to become aware of the distinctiveness of both episodes, and memories for truly experienced events remain unaltered.
CNS: Were there any novel approaches in your study that you would like to highlight?
Siestrup: Many studies on episodic memories use rather simple and/or artificial materials and tasks. For example, participants are supposed to remember words from a list or random picture pairings. Our approach was different: to us, it was most important to create a naturalistic situation of remembering an episode that one has truly experienced in a controlled laboratory setting. To accomplish this, we developed a new experimental protocol that features episodes composed of short toy stories.
CNS: Would you please explain how these toy stories work?
Siestrup: For the toy stories we used PLAYMOBIL® toys to create many different stories inspired by real-life situations. For example, in the encoding session participants watched a video about two kids planting a sunflower. In the video, the story unfolds as one toy after another appears in the scene. In this example, we first see some soil in a garden. Then, a boy with a watering can and a girl with a spade appear. They work on the soil a little bit, until a big sunflower appears on the soil. Both kids are happy about their success and celebrate with a little dance.
CNS: And then you change the story, right?
Siestrup: Yes, in the current study, we induced prediction errors by introducing small changes into these toy episodes by exchanging one of the toy items. For example, the girl does no longer hold a spade but a garden fork. As you can see, this change is not really relevant for the storyline: the kids will still be able to accomplish their goal of planting a sunflower no matter what exact tools they use. In our study, these irrelevant changes are called surface modifications. Additionally, there were also changes which had a large impact on the storyline, so-called gist modifications. In this example, the sunflower is exchanged for a raven. Thus, the kids do not accomplish their goal, as the flower cannot grow because a bird stole the seeds.
Participants watched modified videos during an fMRI session to analyze brain responses to mnemonic prediction errors. Afterwards, we showed participants modified videos again and asked whether they knew these episodes from the encoding session. We observed that experiencing prediction errors during the fMRI session (seeing manipulated videos) increased the participants’ tendency to falsely believe that these episodes had been truly experienced compared to a control condition without prediction errors. We think this finding represents a change in episodic memory.
CNS: What do you most want people to understand about this work?
Siestrup: When memories do not exactly match our past experiences, this does not have to be a bad thing. In fact, when your memory has changed, it might actually be due to a very useful mechanism: updating old beliefs through prediction errors when new information is available. So next time you misremember something don’t get annoyed by your faulty brain, but instead tell yourself “Good job learning something new!”
CNS: What’s next for this line of work? And how best can these findings be applied to future studies?
Siestrup: Even though we already learned a lot from the present and previous studies, many questions about the influence of prediction errors on episodic memory changes remain unanswered. Specifically, we and other researchers are still puzzled by the role of the strength of prediction errors, as we do not know yet whether moderate or strong prediction errors are most potent in triggering memory changes. Luckily, we will be able to target those questions in future research with the help of two highly motivated and smart PhD students who recently joined our team. We will also extend our naturalistic protocol to a different type of stimulus material, namely professionally recorded dialogues. In fact, data collection for the first fMRI study that extends on the present publication was just finished, so stay tuned for the next paper!
-Lisa M.P. Munoz