Predicting the future comes at a cost to encoding the present

Poster Session C - Sunday, April 14, 2024, 5:00 – 7:00 pm EDT, Sheraton Hall ABC

Craig Poskanzer¹ (crp2170@columbia.edu), Hannah Tarder-Stoll^1,2, Raheema Javid¹, Mariam Aly¹; ¹Columbia University, New York, NY, ²Rotman Research Institute, Toronto, ON

Forming new memories requires a focus on the external world, whereas retrieving old memories requires attention to our internal world. The hippocampus plays a central role in resolving the tension between encoding new information and retrieving existing memories to make predictions about the future. Computational models suggest the hippocampus balances these processes by alternating between states in which it prioritizes memory encoding vs. memory retrieval. Although previous studies have examined switching between these hippocampal states, they have not concurrently probed encoding and prediction in behavior. Here, we ask whether the discrete encoding and retrieval states in the hippocampus can be detected as a behavioral trade-off between forming new memories vs. using old memories to make predictions. Participants learned a sequence of scene categories (e.g., beaches, castles, forests). After sequence learning, they completed a simultaneous encoding and prediction task. They were shown trial-unique category images and asked to make predictions about upcoming scene categories. Finally, they were given a surprise memory test for the trial-unique images. This allowed us to measure whether there was competition between encoding of the trial-unique images and prediction about upcoming images. As hypothesized, individuals (n = 31) showed a trade-off between encoding and prediction: memory encoding of trial-unique images suffered when prediction was correct vs. incorrect. These preliminary findings suggest that the trade-off between encoding and prediction in hippocampal computations can be observed even at the scale of behavior unfolding over many seconds.

Topic Area: LONG-TERM MEMORY: Episodic