Olya Bulatova^1,2 (), Mahveen Salman Mubarak², Keisuke Fukuda^1,2; ¹University of Toronto, ²University of Toronto Mississauga

Explicit prediction alone, even without statistical regularity, is sufficient to implant visual memories of predicted yet unexperienced events (Bulatova & Fukuda, 2025). This prediction-based false memory is extremely robust: it cannot be eliminated when a contradicting event immediately rebuts the prediction, and when predictions were highly unreliable. However, it remains unclear whether the prediction-based false memory is encoded at the time of prediction or inferred at the time of retrieval. To test this, we recorded EEG as participants 1) made predictions of upcoming visual events that were either confirmed, violated or not verified, and 2) reported their memories of the actual events. Specifically, during encoding, participants saw a colourless silhouette of an everyday object and they predicted what colour the silhouette would be. A thousand ms afterwards, the silhouette turned into either the predicted colour, the opposite colour, or remained colourless with equal likelihood. Importantly, participants were instructed to remember the actual, rather than predicted, colour of the silhouette. After encoding 360 silhouettes, participants’ memory for the actual colour of the silhouettes was tested. First, participants saw a colourless silhouette and indicated whether they remember seeing it 1) definitely in colour, 2) maybe in colour, 3) definitely not in colour, or lastly that 4) they did not see the silhouette. Afterwards, participants reported the remembered colour on the colour wheel along with their confidence in their report. Our results so far (N = 12/30) reveal that the midline theta (3-7Hz) power during encoding was significantly stronger when prediction turned into false memories than when they did not. Furthermore, the response-locked Late Posterior Positivity was significantly larger when indicating the presence of colour memory for the actual events than for predicted events. Taken together, our results demonstrate that both encoding and retrieval processes contribute to the prediction-based false memory.