Extreme responses in entomophobia research: Could AI assist in overcoming the bias?

When asked to rate their level of aversion toward arthropods such as cockroaches and spiders, survey participants often choose the most extreme values on the scale. This tendency results in a highly skewed data distribution with limited variation, which can obscure critical details about true population variance, thus diminishing the utility of the data for testing research hypotheses. In a recent study, my colleagues and I investigated the relationship between feelings of disgust toward insects and the use of household insecticides through an online survey (n = 2500). Approximately 70% of respondents reported strong disgust (ratings of 6 or 7 on a scale of 1 to 7) toward six species of arthropods commonly found in homes. This significant skew toward the extreme end of the Likert scale may have contributed to the unexpectedly small effect sizes observed, which were at odds with the high levels of reported aversion and the prevalent use of insecticides.

A potential solution to extreme responding in entomophobia research could be the development of research-grade, Large Language Model (LLM)-based chatbots capable of conducting in-depth, open-ended online interviews. In this presentation, I will demonstrate how ChatGPT, though not specifically trained for this task, can be prompted to conduct a Computerized Adaptive Test (CAT)-style interview on entomophobia. This allows it to detect and accurately quantify nuances in extreme aversion. By overcoming several challenges associated with turning LLM-based chatbots into reliable research tools, researchers may soon achieve novel insights into the origins and consequences of extreme entomophobia.