Prediction Errors Disrupt Hippocampal Representations and Update Episodic Memories

Abstract

The brain supports adaptive behavior by generating predictions, learning from errors, and updating memories. Prediction error, or surprise, is a known trigger for memory updating; however, the mechanisms that link prediction error, neural representations, and naturalistic memory updating remain unknown. In an fMRI study, we elicited prediction errors by interrupting familiar narrative videos immediately before an expected conclusion. We found that prediction errors reversed the effect of post-video univariate hippocampal activation on subsequent memory: hippocampal activation predicted false memories after prediction errors, but protected memories from distortion after expected events. Tracking second-by-second neural patterns revealed that prediction errors disrupted the temporal continuity of hippocampal representations. This disruption of signal history led to memory updating after prediction error. We conclude that prediction errors during memory reactivation prompt the hippocampus to abandon ongoing predictions and neural representations. Following prediction error, the hippocampus switches to an externally-oriented processing mode that supports memory updating.

Building bridges: the hippocampus constructs narrative memories across distant events

“…In contrast with the present study, other studies have suggested that boundary-evoked hippocampal activity supports storage of events that precede an event boundary (Baldassano et al, 2017;Ben-Yakov and Henson, 2018;Sinclair et al, 2020). These findings are not necessarily incompatible with findings of the present study.…”

“…Similar to the idea of reminder-evoked retrieval (described above), it is possible that some kind of boundary-evoked memory retrieval could support the integration of temporally-distant events into a larger narrative. However, it is also possible that boundary-evoked hippocampal activity supports event segmentation (32)(33)(34)60), as well as memory encoding for a preceding event (60,61). Our findings suggest that, even if boundaryevoked activity in the hippocampus can reflect a "wall" between temporally adjacent events, it might also reflect the construction of a "bridge" between temporally distant events.…”

“…Future investigations should determine whether the correspondence of hippocampal activity at event boundaries with memory for preceding and subsequent events reflects one common, underlying process (e.g. updating representations in response to high prediction error; Antony et al, 2020;Bein et al, 2020;Franklin et al, 2020;Lu et al, 2020;Sinclair et al, 2020) or two overlapping processes (e.g. memory storage vs retrieval).…”

“…An additional possibility is that the hippocampus might integrate events into a coherent narrative, but only at specific moments during experience. Several recent studies (Baldassano et al, 2017;Ben-Yakov & Dudai, 2011;Ben-Yakov & Henson, 2018;Reagh et al, 2020;Sinclair et al, 2020) have found that the hippocampus becomes particularly active when people perceive transitions between events, or event boundaries (Newtson, 1973;Zacks, 2020). Although the significance of boundary-evoked activation is unclear, it might indicate that the hippocampus supports retrieval of information from past events in order to inform encoding of a new event (Griffiths & Fuentemilla, 2020;Stawarczyk et al, 2020).…”

The limited reach of surprise: Evidence against effects of surprise on memory for preceding elements of an event

“…Surprising occurrences, like rewarding or novel ones, lead to dopaminergic activity (Barto et al, 2013;Horvitz, 2000;Ungless, 2004), suggesting surprise may also retroactively enhance memory. Moreover, a recent preprint found that episodic prediction error (a mismatch with a previous presentation of the stimulus) leads to retroactive memory enhancement (Sinclair et al, 2021). Other studies have found that aversive stimuli too may retroactively enhance, rather than impair, memory (Anderson et al, 2006;Dunsmoor et al, 2018;Dunsmoor et al, 2015;Smith & Beversdorf, 2008).…”