BAYESIAN INTEGRATION MECHANISMS UNDERLIE NUMERICAL JUDGMENTS IN MONKEYS AND CROWS

Cognitive biases often lead people to misjudge sizes, amounts, or other magnitudes. However, it remains unclear how these biases evolved and which neural architectures support them. Nonhuman animals like primates and corvids share with humans a nonsymbolic number sense. To find out whether biases in magnitude estimation are evolutionarily widespread, we investigated numerical estimation in two distantly related but numerically proficient vertebrates: macaques and carrion crows. They were trained on a delayed match-to-numerosity task with dot arrays. An error-trial analysis revealed that both species exhibit the same key biases commonly observed in humans during numerical estimation tasks: scalar variability (increasing response variability with numerosity), regression to the mean (overestimation of small and underestimation of large numerosities), and a sequential effect (dynamic biasing toward previous trial numerosities). These biases have mostly been studied separately and attributed either to a static influence of overall stimulus statistics or to dynamic influences of recent trials. We demonstrate that a Bayesian model incorporating dynamic priors based on multiple previously encountered numerosities can account for all observed behavioral patterns. Our findings suggest that numerical judgments in both species are shaped by uncertainty and the integration of recent experience. More broadly, Bayesian-like mechanisms may explain cognitive biases in magnitude estimation across distantly related animals, independent of a neocortex.