Chatbot Linguistics: Using Artificial Intelligence in Suspicious Activity Reporting

• In Event: Conspiracies, Linguistics, Willingness to Engage in Violent Extremism, and How AI Can Help Prevent It

Abstract

Public reporting of suspicious activity is integral to crime prevention, but current reporting methods are often lacking. For example, traditional tip line reporting methods are burdensome on phone operators, but switching to web forms removes the ability for clarifying questions during a report. Research has recently explored the merits of chat-based technology (e.g., chatbots) to solve these issues for customer service, healthcare inquiries, and even emotional support services. Recently, this technology was brought into the suspicious activity reporting (SAR) space. In a recent study, we compared a prototype chatbot reporting system to a traditional online webform and found that the two performed similarly (Elson et al., 2024), establishing a baseline for the use of chatbots in SAR. The current project seeks to use this baseline to test improvements to chatbot using generative pre-trained (GPT) technology and past research in law enforcement and linguistics. We use an online survey-experiment in which over 1,200 US adult participants watch a staged suspicious activity video and report what they saw to one of four different versions of an online chatbot. Findings indicate that reporting accuracy, length, and trust in the system vary as a function of the linguistic characteristics of a chatbot.

Authors

• Callie Vitro, University of Nebraska Omaha

• Erin M. Kearns, University of Nebraska Omaha / NCITE

• Joel Elson, University of Nebraska Omaha / NCITE