Lukas Maninger¹, Lina Eicke-Kanani¹, Anna-Lena Eckert², Constantin A. Rothkopf^1,3, Thomas S. A. Wallis^1,3; ¹Technical University of Darmstadt, ²Marburg University, ³Center for Mind, Brain and Behavior (CMBB), Universities of Marburg, Giessen, and Darmstadt

When observing collision-like interactions between two objects, humans can automatically attribute causality, determining whether one object (A) caused another (B) to move. For such launching displays, participants report weaker causal impressions when B's outgoing trajectory deviates more from Newtonian dynamics; however, there is little quantitative work on subjective causality perception. Here, we hypothesize that in the face of uncertainty, participants may construct a counterfactual initial configuration that is in higher agreement with the observed outcome. We collected data from 32 participants, each performing three tasks: 1. predicting B's outgoing trajectory after seeing A's movement, 2. estimating A's start position after viewing a launching display where B had a movement direction of varying degrees of implausibility, and 3. rating causality subjectively for similar launching displays as in 2. We developed a Bayesian model to jointly account for behavior across all three tasks by incorporating uncertainties over observations, biases in orientation perception, and decision noise. Reported angles in 1 and 2 were generally accurate but showed repulsion away from the cardinals. Causality ratings tended to decrease with collision implausibility for all but two participants. Our probabilistic model of trajectory prediction and causality rating involving sensory uncertainty was able to capture behavioral data on a subject-by-subject level. Our data shows that causality perception is based at least in part on physical plausibility. Such a mechanism would be consistent with the notion that causality is attributed based on the compatibility of an observation with an internal prediction. However, we note that we found a small tendency to base reports of A's start position on B's implausible movement direction, indicating the possibility of a retrospective component to reconcile unexpected outcomes. While this bias was minor, we speculate that counterfactual simulations could also contribute to causality judgments in real-life scenarios under uncertainty.

Acknowledgements: This work was supported by the European Research Council (ERC; Consolidator Award 'ACTOR'-project number ERC-CoG-101045783) and the Deutsche Forschungsgemeinschaft (German Research Foundation, DFG) under Germany's Excellence Strategy (EXC 3066/1 "The Adaptive Mind", Project No. 533717223).