Neural networks reveal emergent properties of collective learning in democratic but not despotic groups

Collective learning, the improvement of behaviours through experience collective actions, is an area animal learning that has received little attention. We investigated how individual during actions could produce improvements in performance, and decision-making processes, including leadership dynamics, impact upon learning. trained artificial neural networks, either solo or paired, at orientation task, based navigation animals. In pairs, we implemented two rules decision making: ‘democratic’ (weighted average propositions) ‘despotic’ (one individual's proposition, determined randomly with weighted probabilities each trial). Decision-making weightings were varied between but fixed for a given pair, asymmetric generating ‘leaders’ ‘followers’. found nearly all pairs improved their orientation, more slowly than learners. Within leaders learnt quickly followers (‘the passenger–driver effect’). democratic performance individuals to compensate partner error. This emergent process was not observed despotic making, which similarly Our model helps clarify links making context navigation, behaviour, generally.