Empirically test clone-based disambiguation and separate B representations in the ABCB task

Establish whether animals performing the ABCB sequential navigation task, in which two B goals share the same spatial location, employ a goal-instance disambiguation mechanism (clone-based or similar) by testing for the predicted behavioral signatures and for separate neural representations of the two occurrences of goal B when encountered twice.

Background

The paper introduces schema-based hierarchical active inference (S-HAI) and uses simulations to predict behavioral and neural signatures in structured navigation tasks. For the ABCB task—where two goals (B) coincide spatially—the model requires a clone-based mechanism to disambiguate goal instances and predicts distinct neural representations for the two B occurrences. The authors explicitly state that these predictions remain to be empirically tested.

References

One such prediction concerns the behavior and neural representations that might be observed in animals performing the ABCB task (Figure~\ref{fig:abcb_reward}). Our simulations suggest that correctly solving this task requires a mechanism (clone-based or similar) capable of disambiguating between distinct instances of the same goal (e.g., goal B). This, in turn, should produce specific patterns of behavior and Level 2 neural representations, including separate representations of the same goal when it is encountered twice (see Figure~\ref{fig:task3}). These predictions remain to be tested in future experiments.

— Schema-based active inference supports rapid generalization of experience and frontal cortical coding of abstract structure  (2601.18946 - Maele et al., 26 Jan 2026) in Discussion