Are Core Beliefs About Agents Revisable? Evidence From Preschoolers.

• In Event: Learning From Violation of Expectations in the Domain of Agents

Abstract

The core knowledge view (Spelke & Kinzler, 2007) proposes that humans are endowed with a small number of core knowledge systems (e.g., object, agent, space, and number), each accompanied by a set of principles. These principles are early emerging and are foundational for later learning. Yet one of the hallmarks of human learning is that beliefs can be revised given new evidence. Previous work has shown that apparent violations of core principles about objects lead to exploration and enhanced learning in infants and preschoolers (Stahl & Feigenson, 2015, 2017). Toddlers can also revise their more complex beliefs about agents (e.g., ToM, subjectivity of preferences) given counterevidence (e.g., Amsterlaw & Wellman, 2006; Lucas et al., 2014; Ma & Xu, 2011). Are the earliest-emerging core beliefs about agents also subject to revision, like other later developing beliefs? The present study investigates whether 4- to 6-year-olds can revise 3 core beliefs about agents: agents’ intentional actions are directed to goals, agents choose efficient means to achieve their goals, and agents’ preferences can be inferred based on violations of random sampling (henceforth referred to as Goal, Efficiency, and Sampling principles).

In 3 experiments (N = 96), 4- to 6-year-olds observed events that supported (Belief Consistent condition) or violated (Belief Violation condition) each of the 3 principles, or they did not receive any new evidence about the principles (Baseline condition). Then, children made predictions about the outcomes of new events. They chose between outcomes consistent (Belief-Consistent response) and inconsistent (Belief-Violation response) with the principles (see Figure 1 for a schematic of the procedure for the Goal principle). Strength of evidence varied across experiments. For each principle, children received 4 pieces of evidence in Experiment 1 and 6 pieces of evidence in Experiments 2 and 3. We included 3 types of test trials: easy test trials (predictions about the same agent), hard test trials (predictions about a new agent), and harder test trials (predictions about an agent that is not a geometric shape).

Results showed that children can revise their beliefs about these principles given counterevidence: They were more likely to choose the BV response in the BV condition than in the Baseline and the BC conditions for all 3 principles (ps < .03). We also found some interesting differences across principles. Children had stronger prior beliefs about the Sampling principle: In the Baseline condition, they were less likely to choose the BV response for the Sampling principle than the other 2 principles (ps < .001). In addition, children were more likely to revise their beliefs for the Goal principle: In the BV condition, they were more likely to choose the BV response for the Goal principle than the other 2 principles (ps < .001). Children’s responses did not differ across test trial types, suggesting that they generalized their revised beliefs about these principles to other agents. These findings indicate that even our earliest-emerging beliefs about agents are subject to revision, given a small amount of counterevidence, and preschoolers are flexible and powerful learners.

Authors

• Rongzhi Liu, University of California - Berkeley
  Presenting Author

• Fei Xu, University of California - Berkeley