Counterfactual reasoning facilitates belief revision in preschoolers

• In Event: Belief revision in a social world: How other people shape children’s ability to revise beliefs

Abstract

A distinctive feature of human cognition is the capacity to flexibly adapt our beliefs to align with a changing environment. Despite this, learners sometimes ignore, reject, or qualify anomalous evidence to maintain their existing beliefs. We explore whether counterfactual reasoning may facilitate anomaly detection and belief revision in an implicit rule switching task. Specifically, we examine whether considering alternatives may help 5-year-olds overcome the tendency to fixate on a single focal hypothesis (or rule). Prior work has found that reasoning about counterfactual outcomes helps children to detect evidence that discriminates between two causal rules (Engle & Walker, 2018). Here, we prompt children to consider alternatives to an irrelevant, non-causal variable, rather than the variable of interest. We predict that counterfactual reasoning will invoke a general openness to alternatives, reducing the amount of counterevidence required for children to revise an existing belief.

A total of 139 5-year-olds (M=65.6 months, SD=3.73 months) were recruited for the study. All participants were introduced to a novel causal system, composed of a machine and two types of objects: 5 ‘wugs’ (i.e., blocks that caused a machine to light up) and 5 ‘not wugs’ (i.e., blocks that did not make a machine light up). Each of the 10 training blocks had a combination of external features (i.e., antenna, switch, button). Notably, only one of these external features (e.g., the antenna) was causally relevant —it determined whether the block was a ‘wug’ (Figure 1)— during the pre-switch phase. The remaining two features were either (1) causally relevant during the post-switch phase (e.g., the switch), or (2) irrelevant/non-causal throughout the task (e.g., the button). After sorting all blocks as ‘wugs’ or ‘not wugs,’ children were given a pretest to determine whether they learned the relevant feature. Only those who passed the pretest (n=52) were randomly assigned to control (n=24, M=65.3 months, SD=3.79) or counterfactual (n=24, M=65.9 months, SD=4.04) conditions. Children in the control condition were asked to report what happened to the machine for each pre-switch training block. Children in the counterfactual condition were asked to consider what would have happened to the machine if each of these blocks had been modified. Critically, all counterfactual prompts focused on the irrelevant, non-causal feature (e.g., the button). Lastly, the experimenter introduced a second set of blocks and asked children to predict which were ‘wugs,’ before placing them on the machine (Figure 2). Critically, the relevant feature changed (e.g., from antenna to switch) in this second set. The experimenter then continued to present “anomalous blocks” that violated the initial rule, but provided evidence for the new rule. We then recorded how many pieces of counterevidence were needed for children to revise their belief.

There was a significant difference between conditions, F(1,46)= 7.41 , p<.01, n2p=.139, with children in the counterfactual condition more likely to revise their beliefs following fewer anomalies than controls. Results indicate that prompting children to think about alternatives—even those that redirect attention to an irrelevant, non-causal variable—helps them detect anomalies and promotes belief revision.

Authors

• Jae Engle, University of California, San Diego
  Presenting Author

• Caren M. Walker, University of California - San Diego
  Non-Presenting Author