Work

Belief revision as a socially situated rational process

Year: 2025
Type: dissertation
Abstract: A long tradition of research across the behavioural sciences has characterized human belief revision relative to a standard of rationality, whether by arguing that people are generally optimal reasoners who believe what they should given the evidence they encounter, that people exhibit systematic deviations from the predictions of rational models due to limited cognitive resources or motivated reasoning, or some balance between these two. In this dissertation, I propose a unified formal computational model of belief revision that recasts not only the influences of motivation and resource constraints on the processes underlying belief revision, but also considers the influence of social structure—such as trust and source knowledge about informants and the formation of social conventions—that shape the process of learning from and about social partners. I first compare the predictions of this model to several classic and contemporary empirical findings from psychology and cognitive science, focusing on the phenomenon of belief polarization, which various theories have argued can emerge from normative, resource-limited, or motivational processes (Chapter 3). I then test the predictions of this model in a series of experiments, arguing that people are best captured by a resource-rational model of belief revision, even in situations where normative models predict that people's beliefs will diverge from one another (Chapter 4). Lastly, I show in Chapter 5 that simulations of rational agents attempting to maximize their own expected utility nevertheless result in sustained misrepresentation of many agents when social coordination is necessary, because these representations mutually influence one another's behaviour. I also show that humans also behave similarly to the models, providing another explanation for why people might harbour inaccurate beliefs about social others even when their reasoning is not constrained by limited cognitive resources or motivational biases. Together, these findings provide not only a richer understanding of human belief revision across varying contexts, but also provide support to a broader concept of rationality, one in which people consider not only the value of representing epistemic truth, but also the costs of mental computation, the instrumental value of holding a belief, and the need to achieve social equilibrium.
Language: English
Open Access status: green
OpenAlex ID: W7139533766
Cites: 360
Cited by:
Related to: 10