Ensemble Perception Without Awareness

Abstract

A central challenge in consciousness research is to dissociate perception from awareness using rigorous, bias-free methods. Traditional approaches often suffer from two key limitations: the criterion biases in reporting perceptual (un)awareness, and the criterion content fallacy, where awareness measures do not capture the specific information required to perform the perceptual task. These problems are compounded by the widespread use of single-object paradigms, which typically involve presenting isolated stimuli near threshold and suppressing them with masking techniques. These constraints result in a low signal-to-noise ratio that severely limits sensitivity to detect effects and offers low ecological validity. Here, we introduce a novel paradigm based on ensemble perception that overcomes these long-standing limitations. Experiment 1 used a bias-free two-interval forced-choice task in which observers had to discriminate the predominant category in ensembles comprising animate and inanimate items and also detect which interval contained the task-relevant features. Experiment 2 extended these findings with a single-interval design, addressing concerns about potential detection inefficiencies. Multiple control analyses across experiments demonstrate unconscious ensemble perception: observers reliably discriminated ensemble-level properties even when detection of the task-relevant features was at chance. Convolutional neural network simulations further showed that this dissociation can arise from weak ensemble-level evidence, supporting discrimination even when detection fails. Bayesian ideal observer modeling situated these findings within a formal computational account of unconscious perception. Together, our findings demonstrate that perceptual processing of complex visual summaries can occur without conscious awareness. The work offers a new framework for studying visual consciousness beyond the limits of traditional single-target paradigms.