Antonia Langenhoff
Talk title: Disagreement and the Development of Intellectual Humility

A healthy public discourse is fundamental to democracy. However, current public discourse seems to be characterized by polarization and the idea that arguments are about winning and little else. How can we respond to disagreements in more constructive ways? One answer to this question is: we need to remember how to be intellectually humble. In a nutshell, intellectual humility means understanding the limits of one’s own knowledge. For example, when engaging in argumentation with others, an intellectually humble agent should demonstrate a willingness to revise their belief when others have better evidence for an alternative view. In my research, I have begun to explore the development of some of the key features of intellectual humility. I will present preliminary results from two sets of studies. In the first set of studies, I investigate how children change their beliefs when confronted with a disagreeing peer. Previous work has shown that with age, children are increasingly able to use different epistemic cues when forming beliefs. But in situations of disagreement, agents have typically formed a prior belief and must respond to an alternative belief. I show that some ability to respond reasonably to disagreement is present from at least the preschool years. However, my results also indicate that 4- to 6-year-old children struggle to suspend judgment in situations of ambiguous evidence. In a second ongoing study, I ask whether experiencing disagreement can reduce children’s overconfidence and trigger information search. In the remainder of my talk, I will discuss an idea for a future study in which I am planning to look at an additional feature of intellectual humility. Specifically, I want to investigate the circumstances under which young children show a systematic preference for finding the truth over purely winning an argument.

Yuan Meng
Talk title: Naïve Utility Calculus underlies the reproduction of disparities in social and non-social domains

Abstract: Racial disparities are ubiquitous in our society. Take the US criminal justice system for instance: While African Americans and Latinos only make up 29% of the US population, they comprise 57% of the inmates in the country and 63% of those serving life in prison. When you see those numbers, what do you think? If you are well aware of the history of slavery, segregation, and systematic racism, you may take them as evidence of biases against Blacks and Latinos. However, if you are oblivious to such history and believe the police know something you don't and are doing a good job, then you may well think Blacks and Latinos are more prone to crime than other groups. When it's your turn to "be the police", you will also scrutinize these groups, thereby reproducing disparities that you see. We argue that this vicious circle may be a natural consequence of rational learners understanding the action of others through a Naïve Utility Calculus (NUC, Jara-Ettinger et al., 2016): If a knowledgeable agent knows the true crime rate of each group and maximizes the expected utility of costly checking, then one should infer that groups checked more often have higher crimes rates. In Study 1, we found that people inferred population hit rates of social and non-social groups based on an agent's sampling behavior (check rates) in a similar way as the NUC model. In Study 2, we looked that whether providing information on sample hit rates (i.e., out of those checked, how many committed crime, or more generally, possessed that target trait) changed people's inference. We found that people utilized both the agent's sampling behavior and sample hit rates to infer population hit rates. This was supported by the fact that a NUC model that takes into account both rates outperformed models that only learned from sample check rates or hit rates. At the end of this talk, I will take about how this framework might explain how people interpret true crime data in the wild and discuss future directions both on the modeling and the experimental fronts.