Grant: Discovery Grants Program - Individual

Comparison making is among the most basic operation we can expect from a cognitive system. Comparisons are performed on physical (e. g., weight or size) but also on symbolic (e. g., letters or words) attributes. How comparisons are performed is still mysterious, knowing that attributes are not directly accessed but perceived by senses which can only detect frequencies amidst massive amount of noise (contaminating signals). Among the unresolved questions addressed herein is whether the absence of a valid attribute is perceived, and if irrelevant, contaminating, attributes are also contributing to the detection of a target stimulus. Further, current models of low-level cognition, collectively called sampling models (e.g., DDM, Ratcliff, 1978; LBA, Brown & Heathcote, 2008) fit data well by assuming that evidence are collected towards response alternatives. However, they do not specify how a missing piece of information can turn into a positive source of evidence. Three classic paradigms are relevant. First, the Same-Different task (Bamber, 1969) in which two stimuli are presented in close succession and where the participant must indicate if the second (the probe) is entirely identical to the first. Second, the Redundant-attribute detection task (Miller, 1982) in which the participant must indicate if the probe contains at least one attribute of the target. Third, the visual search task (Shiffrin & Schneider, 1977) where the participant must indicate if the test display contains at least one target object among a pool of possible targets. All three paradigms have been studied for decades and yet, a complete description of the processes to accomplish them is missing. This proposal builds on previous research funded by the NSERC which highlighted that contaminant attributes are contributing to a decision (Cousineau et al., 2023; S. Harding, Cousineau & Shiffrin, 2021). It also showed that base latency increases with the total number of attributes, but not the work accomplished, as variance is not increasing (also see B. Harding & Cousineau, 2022; Goulet & Cousineau, 2021). From the perspective of the comparison process, this increased latency would resemble a passive observation period. An explanation to these results could be to assume a preliminary phase estimating the entropy of the test display relative to expectancy. The more entropy is detected, the more resources are solicited to resolve the nature of the signals generating entropy. Five empirical studies and one model development study are described to explore this hypothesis and reject competing alternatives. An entropy-based theory of simple decision would provide a formal framework for the deployment of attention. This would have tremendous implications for vigilance and security as well as for robotics and artificial vision.