UC Berkeley

Learning is a complex process that requires integrating new evidence with existing beliefs. This task is relatively straightforward when the evidence is consistent with and thus supported by existing beliefs. In such instances the evidence serves to confirm and hence strengthen beliefs. It becomes less clear, however, when the evidence is inconsistent—that is, when the evidence directly conflicts with one’s prior knowledge. Will learners revise their beliefs to reflect the new evidence or will they instead maintain their existing beliefs, perhaps ignoring the conflicting data as anomalous? How might their responses differ as a function of development, and what might this reveal about the mechanisms that support learning (and relearning)?

The current dissertation addresses these questions by asking how children and adults respond to evidence that violates their beliefs. Chapter 1 begins by describing a surprising developmental pattern: younger children are sometimes better than older children and adults at using statistical evidence to revise their beliefs. I then provide a Bayesian explanation for these findings, noting that younger learners may outperform older ones because they are intrinsically more flexible in their search for alternative hypotheses. Chapter 2 then provides empirical evidence that young children revise their higher-order beliefs by attending to the statistical strength of the data and weighing it against their prior belief. When shown a single event that conflicts with their belief, 4- and 5-year-olds largely maintain their higher-order belief. However, when shown several events that each support the same alternative hypothesis, children reliably update their belief to account for the new data. Chapter 3 then replicates these findings with even younger children, demonstrating that 3-year-olds similarly update their beliefs when the counterevidence is strong enough to outweigh the prior belief. Notably, children’s ability to update their beliefs is independent of their executive ability to deliberately change their behaviors in response to new goals and demands. Chapter 4 then contrasts children’s epistemic flexibility with adults’ inflexibility, exposing a striking developmental decline in adults’ sensitivity to new data. In particular, these findings highlight the robustness of adults’ beliefs by demonstrating that adults continue to endorse a weakly supported belief even after they are primed or incentivized to reason more broadly and creatively. Finally, Chapter 5 discusses the implications of this work and suggests future directions.

Taken together, these experiments contribute to our emerging understanding of belief revision by demonstrating how, as knowledge is accumulated and as beliefs are strengthened, our sensitivity to new evidence decreases and our willingness to revise declines. This pattern, although at times counterproductive, may be adaptive as learners transition from a strategy of exploration to one of exploitation.