Poster #192 - The Effects of Information Quality and Informants' Knowledge on Evaluations of Under-Informative Pedagogy

• In Event: 3-159 - Poster Session 12
  In Poster Session: PS 12 Section - Social Cognition

Integrative Statement

Humans often rely on others to learn about the world. However, informants vary in quality – some may be wrong, ignorant, or even deceptive. Thus, the ability to detect and evaluate unhelpful informants is critical for accurate learning. How do human learners face this challenge, and how does this ability develop? Here, we propose that even young children have an intuitive theory of what it means to be a good informant; rather than relying on heuristics such as "more information = better", they evaluate informants based on an abstract understanding of how helpful individuals should select information given what they know. Specifically, we suggest that informant evaluations should fluidly integrate information about the amount of information shared, the value of that information, and the knowledge state of the informant. We formalize this idea as a Bayesian model of informant evaluation and test its predictions across three behavioral experiments with adults and preschool-aged children.

In Experiment 1, we showed adults (N=1024) a novel device with four functions; two of these functions were more interesting or useful than the other two (validated in pilot studies). Participants then saw scenarios of an informant discovering all or a subset of the functions on this device, and then teaching all or a subset of what he had discovered to a naive learner. We systematically varied the number and value of functions discovered and taught, such that participants saw one of 15 possible teaching scenarios. Participants then rated the informant's quality. Adults' ratings were well predicted by our model (R2=0.96), integrating information about the number of functions taught, their values, as well as what the informant knew (see Figure 1).

In Experiment 2, we developed a simplified paradigm to test these abilities in preschool-aged children (N=24, M(age)=60 months). We additionally validated this method in a group of adults (N=24). Participants watched videos of informants discovering and subsequently teaching learners about a toy with four functions (2 high-value, 2 low-value). The number and value of functions the teachers discovered and taught were varied such that participants saw a total of five teachers.

Children and adults exonerated the unintentional omissions of an informant who had incomplete knowledge, rating this teacher more favorably than one who knowingly omitted the same information (Children: p=.017; Adults: p=.051). Participants also provided graded ratings based on the degree of omission (Children: p=.019; Adults: p<.001; see Figure 2).

Experiment 3 clarified the effects of information value by presenting children (N=40; M(age)=59 months) with just one of two informants, who discovered all four functions, and taught either two high-value functions, or two low-value functions. Here, children in the high-value condition rated the informant more favorably than those in the low-value condition (p=.006).

In sum, young children and adults appear to be able to go beyond the observable aspects of informants' behaviors to make nuanced inferences about their helpfulness. This suggests that even before entering formal schooling, preschoolers have highly sophisticated cognitive models of what it means to be a good informant.

Authors

• Ilona Bass, Rutgers University - Newark
  Presenting Author

• Daniel Hawthorne, Datawallet
  Non-Presenting Author

• Noah Goodman, Stanford University
  Non-Presenting Author

• Elizabeth Bonawitz, Rutgers University
  Non-Presenting Author

• Hyowon Gweon, Stanford University
  Non-Presenting Author