Robo-Psychology 23 - Human Cognitive Susceptibility and Generative AI

We introduce the Cognitive Susceptibility Taxonomy (CST), a framework designed to identify and categorize human cognitive vulnerabilities that interact with and amplify AI system failures. It argues that AI safety is a two-sided problem, requiring alignment of both machine behavior with human values and human behavior with the reality of what AI is. The text details twelve specific human susceptibilities—such as anthropomorphic-trust bias, automation over-reliance, and parasocial attachment—explaining how these predictable human tendencies can exacerbate technical AI issues like hallucinations, misinformation, or even lead to severe real-world harm, including emotional distress or self-harm. Ultimately, the CST aims to provide a common vocabulary for understanding these human-side "failure modes" to inform better AI design, governance, and public understanding, ensuring a safer human-AI future.