Dynamic salience as a basis for behavioural meta-malleability

Abstract Salience, the degree to which cues capture attention, is not fixed but updated through experience. Learning which cues anticipate meaningful outcomes requires cue–outcome relationships to be perceptible to begin with, rendering salience a gate for behavioural change. Because salience is itself malleable, behaviour exhibits ‘meta-malleability’, whereby the modifier of behaviour (learning) can itself be modified. A cue's salience is determined partly by its physical properties and capacity to stand out, and partly by the subject's prior experience with it. Cues reliably followed by important outcomes capture more attention, as if they were physically intense, whereas poor predictors lose their attention-capturing potential even when physically intense. Concurrent cues compete for attention, and less salient cues are undermined by more salient counterparts. This competition occurs regardless of how each cue acquired its salience status. Under certain circumstances, organisms orient towards informative cues at the expense of biological needs; yet, they may sometimes orient towards cues with uncertain outcomes. Together, these findings demand an overarching framework. We synthesize behavioural, computational and neurobiological perspectives on salience dynamics. One pay-off of this framework is a principled account of why order matters across otherwise equivalent encounters with cues of varying salience, channelling behaviour down diverging trajectories.