A fundamental understanding of memory continues to elude psychologists. The cognitive and neurological processes that underlie the encoding of lived experiences is astoundingly complex, and not limited to a single pathway or mechanism. Indeed, one well-documented but still poorly understood dichotomy is that of episodic vs. statistical memory, two types of encoding that relate respectively to specific details about events (memory) vs. abstract knowledge extracted from multiple experiences (learning).
Thanks to researchers from Yale University, though, we’re one step closer to understanding how competing memory pathways are organized and prioritized in the brain. Their study, published in the Proceedings of the National Academy of Sciences, married behavioral tests and brain imaging to explore the interaction of learning and memory when they compete for resources in the hippocampus.
“There’s a rich tradition in psychology of studying different kinds of memory and how they interact with one another,” explained study author Brynn Sherman, a graduate student in the psychology department at Yale University.
“In particular, there’s been a strong focus on how we form memories for specific, discrete experiences (episodic memory) vs. how we form more abstract memories about what tends to happen (statistical learning). A lot of work has focused on how these two forms of memory are different from one another, but more recent work in neuroscience has suggested that they might actually be related to one another. Here, we bridged this gap, using both psychology and neuroscientific methods to ask how these two memory systems interact with one another.”
The study involved three experiments, with a total of 181 participants. The researchers pitted episodic memory against statistical learning by subtly manipulating the presentation of different scenes on a computer screen, such that certain scenes (like a waterfall) were invariably followed by certain other scenes (like a first). This Category A → Category B relationship was only true for half of the 12 scene types; for the other 6 scenes (Category X), there was no pattern.
Participants were not forewarned of the relationship between certain scenes; the authors left it up to their hippocampi to naturally learn the associations, whether consciously or not. Thus, when tasked with recalling, for example, whether a particular pair of scenes felt more or less familiar (Experiment 1) or estimating approximately when they had seen a particular image (Experiment 2), participants would demonstrate greater or weaker statistical or episodic memory.
The results of the study show that, when episodic and statistical encoding compete for resources, the latter wins out. That is, individuals tended to recall Category A images more poorly than Category X images. As Category X images were random and the brains of participants were unable to detect any pattern to learn, greater attention was given to encoding the episodic memory.
Category A images, which invariably led to Category B images, however, were recognized as belonging to a pattern — more resources were diverted to understanding this pattern than to encoding the image itself.
The final experiment recreated these results with the addition of fMRI brain scans. It was shown that neural evidence for prediction of an upcoming image (Category B), as demonstrated by activity in the hippocampus, was associated with poorer encoding of the current scene (Category A).
“When people think about memory, they often think about rich memories for specific experiences,” Sherman told PsyPost. “But, we also have memory for much more mundane aspects of life: memories for what usually happens in our day-to-day experiences. These kinds of memories are useful for making predictions about the future, and our work highlights how these automatic predictions that we make can actually have a powerful influence on how we remember those specific experiences later on.”
The authors note some limitations, most notably that Experiment 3 failed to reproduce behavioral differences at a group level, as did Experiment 1. Second, certain statistical acrobatics were necessary to compensate for the lack of temporal and physical granularity of fMRI imaging.
Memory and learning are complex. They both underlie and are strongly influenced by cognition and emotion, and beyond the practical (e.g., pedagogic) implications of understanding memory and learning, no model of the human brain will ever be complete without it.
“This study opens a lot of exciting avenues for future research. We’re really interested in the specific mechanisms by which prediction influences memory, and we’re running multiple follow-up studies to try to dig deeper into this,” Sherman said.
The study, “Statistical prediction of the future impairs episodic encoding of the present,” was authored by Brynn E. Sherman and Nicholas B. Turk-Browne.