Adaptive learning strategies in purely observational learning

Individual learning (IL) and observational learning are both important for humans to acquire information. Observational learning consists of action-only observational learning (AL) and action-outcome observational learning (AOL). Heterogeneous results have been found in previous research on comparing these three kinds of learning (IL, AL and AOL), as a result of different paradigms. The current study was to seperate and compare the learning processes of the three learning styles with an adapted the two-arm bandit paradigm, and notably to propose a new computing mechanism based on reinforcement learning (RL) rules for AL. We also focused on the effect of the skill of demonstrators to distinguish the applicable situation of our new model, in which demonstrator's action preference was regarded as the inferred outcome to drive the learning processes in AL condition. Results showed that: a. With more information, IL and AOL led to better learning performance than AL; b. In skilled demonstrator group, apparent action preference in AL can make up for the decline in learning performance and confidence. Importantly, the new computational model explaining AL won only when the demonstrator was skilled, indicating learners adapted their learning strategies in different situations.