Mapping Cues of Perceived Humanness: A Lens Model Approach

Erika De Los Santos

Advisor: Matthew S. Peterson, PhD, Department of Psychology

Committee Members: Gerald Matthews, Patrick McKnight

Online Location, Zoom
July 09, 2026, 11:00 AM to 01:00 PM

Abstract:

Phishing attempts have a long history as a cybersecurity threat, with most organizations requiring their employees to take anti-phishing training to prevent privileged information leaks. Unfortunately, with the development of artificial intelligence (AI) and large language models (LLMs) specifically, the standard for which we distinguish authentic and adversarial information is rapidly changing. While humans and AI-generated tools have mixed and varied results in identifying these LLM-generated texts, the use of human-AI teaming in distinguishing these texts could help offset poor performance across separate groups. In order to determine the best design for these such tools, it is important to first understand the decision-making of the human operator in these text-discrimination tasks.  To accomplish this, three studies were conducted to determine the cues utilized by human participants when distinguishing human-authored and LLM-generated paragraphs across several genres of writing. Two of these studies evaluated the utilization of these cues with the Brunswik Lens Model to determine the ecological validity, cue utilization, and achievement scores across participants in the final two studies. Furthermore, eye tracking results from the final study allowed us to examine differences in eye movement patterns between high and low achievers in the detection tasks.  Results were compiled into final design recommendations to encourage partnerships between researchers and designers to respond to threats of cognitive bias, fatigue, and improve phishing rate detection overall.