Fig. 1: Fragmentation of World Models.
From: Fragmentation and multithreading of experience in the default-mode network
a Any experience can be divided into models of states (abstract contexts), agents (others’ beliefs/goals), and actions (temporal paths through state space). The midline prefrontal cortex can be viewed as assembling such a model space, configuring the best model for each world abstraction. It generates top-down predictions in a tripartite organisation, with domain-specific belief updates recruiting each region selectively. Consider visiting a friend’s home for lunch. You walk into what you assume is the bedroom, only to discover a fully equipped work PC. The context then updates from “bedroom” (State 1) to “home office” (State 2). You observe the friend’s apparent unhappiness during cleaning, possibly due to not offering help (Agent frame 1), changing your representation of their mood (Agent frame 2). Consequentially you consider alternative dining options than eating in, e.g., grabbing food from a nearby food truck (Action path 1) or restaurant (Action path 2). The experience itself appears fused, but its deeper compositionality is implicit in the narrative structure of human experience (and later memories). The authors have applied the CC-BY 4.0 license to illustrations generated by Biorender: Created in BioRender. Yasin, F. (2025) https://biorender.com/f0o8ol8. b Design schematic for obtaining belief update time-courses by aggregating reported updates over multiple participants. c Smoothed, group-level belief update time-courses peaking when participants signalled their predictions were being updated in each domain. (Inset) Movie scenes as text descriptions depict different domain-updates (Due to copyright restrictions of the movie “Bang! You’re Dead”, the screenshots are replaced with text descriptions of the scene here; participants watched the full movie inside the scanner). d Interrater reliability in update time-courses computed through split-half correlation (100 times) for State (n = 18), Agent (n = 21) and Action (n = 19) updates. Error bars denote the standard error of the mean.
