Reasoning about quantities and concepts: studies in social learning

Abstract

We live and learn in a ‘society of mind’. This means that we form beliefs not just based on our own observations and prior expectations but also based on the communications from other people, such as our social network peers. Across seven experiments, I study how people combine their own private observations with other people’s communications to form and update beliefs about the environment. I will follow the tradition of rational analysis and benchmark human learning against optimal Bayesian inference at Marr’s computational level. To accommodate human resource constraints and cognitive biases, I will further contrast human learning with a variety of process level accounts. In Chapters 2–4, I examine how people reason about simple environmental quantities. I will focus on the effect of dependent information sources on the success of group and individual learning across a series of single-player and multi-player judgement tasks. Overall, the results from Chapters 2–4 highlight the nuances of real social network dynamics and provide insights into the conditions under which we can expect collective success versus failures such as the formation of inaccurate worldviews. In Chapter 5, I develop a more complex social learning task which goes beyond estimation of environmental quantities and focuses on inductive inference with symbolic concepts. Here, I investigate how people search compositional theory spaces to form and adapt their beliefs, and how symbolic belief adaptation interfaces with individual and social learning in a challenging active learning task. Results from Chapter 5 suggest that people might explore compositional theory spaces using local incremental search; and that it is difficult for people to use another person’s learning data to improve upon their hypothesis.