In addition to attention, two factors play a decisive role in learning: the child's active involvement, and rapidfeedback. According to the theory of the "Bayesian brain", which we have examined in previous years, the fundamental algorithm that enables the brain to adjust its representations of the external world consists of three stages:
- top-down prediction, based on the current internal model ;
- comparison of these predictions with inputs received from the outside world, generating error signals ;
- using these error signals to adjust the internal model.
This model of the Bayesian brain suggests that two ingredients are essential for learning: the generation of anticipation about the external world (active engagement), and feedback in the form of error signals (from the environment or the teacher).
Numerous experiments in animals and humans, in the laboratory and at school, demonstrate that a passive organism does not learn. Echoing these fundamental neuroscientific findings, pedagogical research shows that lectures, where children's minds can wander, are less effective than active pedagogies, which engage the child. According to the experiments of Henry Roediger and his collaborators, learning is optimal when periods of explicit teaching are alternated with periods of knowledge testing. Testing not only measures what has been learned, but is an integral part of pedagogy, enabling children to evaluate and correct themselves. The study of metacognition shows that we frequently overestimate our learning. Many students practice re-reading the lecture, but this strategy has little effect: it is the explicit testing of knowledge, coupled with rapid feedback on any errors, that constitutes the best strategy.
