Andrea Massironi¹, Giulia Spinelli¹, Carlotta Lega², Luca Ronconi³, Emanuela Bricolo¹; ¹Department of Psychology, University of Milano-Bicocca, Milan, Italy, ²Department of Brain and Behavioral Sciences, University of Pavia, Pavia, Italy, ³Department of Psychology and Cognitive Science, University of Trento, Rovereto, Italy

Within a highly entropic visual world, experience-dependent mechanisms like statistical learning allow to take advantage of environmental regularities, functionally driving attentional guidance and optimizing visual search. The way attention unravels into space has been linked to a “Mexican-Hat” profile, described as an inhibitory ring encircling the attentional focus to minimize the interference of distractors. While experience-dependent mechanisms like reward have been shown to interact with the inhibitory ring, the role of statistical learning is still underexplored. The current study addresses this gap through a psychophysical task wherein we mapped the attentional profile of 26 participants asking them to report the gap orientation of a “C” letter displayed either as a salient target (Baseline Condition) or as a non-salient probe, with the salient target - now a distractor - located at progressively increasing distances from it. In both conditions, we applied statistical learning, increasing the likelihood of the salient target being in the spatial positions immediately adjacent to the probe, namely at the distance engendering the surround inhibition. Notably, such manipulation highlighted the highly uncertain nature of the salient target, who served with the same likelihood as a real target (Baseline Condition) or as an interfering distractor (Probe Condition). We found that statistical learning cancelled the inhibitory ring, reshaping the “Mexican-Hat” profile into a non-linear gradient, bringing a performance gain in the Probe Condition and maintaining it at ceiling in the Baseline one. Crucially, our findings do not fit with either the expected target vs. distractor behavioural pattern predicted by statistical learning literature. Instead, they fit with a zooming out of the attentional focus, which adaptively reshaped the priority map to resolve uncertainty and optimize search performance across conditions. Overall, our findings bring the intriguing possibility that statistical learning effects are contingent upon the uncertainty inherent to the search context.