Humans can meaningfully express their confidence about uncertain events. Normatively, these beliefs should correspond to Bayesian probabilities. However, it is unclear whether the normative theory provides an accurate description of the human sense of confidence, partly because the selfreport measures used in most studies hinder quantitative comparison with normative predictions. To measure confidence objectively, we developed a dual-decision task in which the correctness of a first decision determines the correct answer of a second decision, thus mimicking real life situations in which confidence guides future choices. Whilst participants were able to use confidence to improve performance, they fell short of the ideal Bayesian strategy. Instead, behavior was better explained by a model with a few discrete confidence levels. These findings question the descriptive validity of normative accounts, and suggest that confidence judgments might be based on point-estimates of the relevant variables, rather than on their full probability distributions.

The use of prior knowledge to guide perception is fundamental to human vision, especially under challenging viewing circumstances. Underpinning current theories of predictive coding, prior knowledge delivered to early sensory areas via cortical feedback connections can reshape perception of ambiguous stimuli, such as 'two-tone' images. Despite extensive interest and ongoing research into this process of perceptual reorganisation in the adult brain, it is not yet fully understood how or when the efficient use of prior knowledge for visual perception develops. Here we show for the first time that adult-like levels of perceptual reorganisation do not emerge until late childhood. We used a behavioural two-tone paradigm to isolate the effects of prior knowledge on visual perception in children aged 4 - 12 years and adults, and found a clear developmental progression in the perceptual benefit gained from informative cues. Whilst photo cueing reliably triggered perceptual reorganisation of two-tones for adults, 4- to 9-year-olds' performed significantly poorer immediately after cueing than within-subject benchmarks of recognition. Young childens' behaviour revealed perceptual biases towards local image features, as has been seen in image classification neural networks. We tested three such models (AlexNet, CorNet and NASNet) on two-tone classification, and while we found that network depth and recurrence may improve recognition, the best-performing network behaved similarly to young children. Our results reveal a prolonged development of prior-knowledge-guided vision throughout childhood, a process which may be central to other perceptual abilities that continue developing throughout childhood. This highlights the importance of effective reconciliation of signal and prediction for robust perception in both human and computational vision systems.