Do children who trust the internet and its contents recognize chatbot errors better?

Benjamin Vonderheide

Research Mentor: Lauren Girouard
Mentor Department: Psychology, LSA
Author(s): Benjamin Vonderheide, Lauren Girouard
Session: Session 6 (3:00 PM – 3:50 PM)
Presentation Type: Poster 119

Abstract

As artificial intelligence becomes increasingly widespread, there is curiosity and concern regarding whether or not children will be able to recognize when generative AI chatbots make errors. Specifically, do children who trust the internet and its contents recognize chatbot errors better? In 2025, Xu et. al found that children seemed aware of and responsive to the accuracy of information they received from chatbots. This study aims to determine how children’s prior confidence in the truthfulness of the internet’s contents affects that ability to discern when chatbots make mistakes. In the error-monitoring section, children are tasked with identifying three types of chatbot errors: typos, factual inaccuracies, and inconsistencies. Later in the study, they are given four answer choices (“big yes, little yes, little no, or big no”) and asked if they think the internet contains more correct information than incorrect information, what kinds of information the internet can provide, and if they know how to and/or are allowed to use the internet on their own at home. We will examine the relationship between children’s preexisting judgments on the overall accuracy of the internet and their ability to determine when a chatbot has made a mistake. This study is ongoing, but I expect that the findings will demonstrate an inverse correlation between trust in the internet and error monitoring. Alternatively, it is also possible that children with greater confidence in the internet have more prior experience with it, enabling them to more accurately recognize chatbot errors and producing a direct correlation between trust in the internet and error monitoring.