Nathaniel Braswell¹, Chaz Firestone², Nicolò Cesana-Arlotti¹; ¹Yale University, ²Johns Hopkins University

Whereas logical inference is typically associated with symbolic notation and laborious proofs, it also arises intuitively in everyday reasoning. Previous work shows that human infants deploy basic disjunctive inferences to infer occluded objects’ identities (Cesana-Arlotti et al., 2018). In two prior studies (Braswell et al., VSS 2023), we discovered that this developmentally basic logical computation arises spontaneously when adults recognize objects in visual scenes. In particular, we showed adults visual events wherein objects are hidden and then revealed in ways that either follow or violate the logically predicted outcomes; subjects responded faster when a revealed object’s identity was consistent with the inference’s prediction than when it violated it. Here, we explore whether such inferences are automatic and even “irresistible”, arising in circumstances where the subject receives statistical evidence contrary to the inference. In Experiment 1, participants had to identify two kinds of concurrent objects: ones logically predicted by the events in the scene and logically unrelated ones. Strikingly, participants recognized objects predicted by the logical inference faster than identically looking objects that were logically unrelated to the scene, suggesting that logical inferences were facilitating and expediting visual processing. In Experiment 2, we manipulated the statistical distribution of revealed objects to create cases where the logically predicted outcome was statistically unlikely (by a ratio of 2:1). Remarkably, participants often misidentified the statistically likely object as the improbable one merely because logic compelled them to do so, despite their predictions being contradicted by previous statistical evidence. In other words, even when it would have benefited participants not to reason logically, they couldn’t help but do so. This work shows how methods from vision science can illuminate the mind's logical capacities. Our findings suggest the presence of core logical inferences that automatically facilitate visual processing and are hard to resist despite prevailing counterevidence.

Acknowledgements: NSF BCS #2021053 awarded to C.F.