Perceived Generative AI risks and behavioral vulnerabilities: understanding cognitive and emotional pathways
Jingfeng Wang, Urvashi Tandon, Rsha Alghafes, Jaskirat Singh Rai, Adrienn Dernóczi-PolyákPurpose
Tools of generative artificial intelligence (Gen AI) have gained immense significance recently, particularly in higher education. While these tools enhance educational efficiency, there are still unanswered questions about the challenges of incorporating this technology into learning processes and its effects on academic outcomes, despite its promise to improve academic work. The purpose of this research is thus to address this gap by developing and validating a comprehensive scale covering perceived risks of Gen AI – such as hallucinations, dysfunctional AI use behavior and biases – and measuring their impact on students' academic performance.
Design/methodology/approach
The study develops a scale through a three-step process. The study used a mixed-method approach to develop a scale on risks associated with Gen AI. The first step involved qualitative interviews to arrive at themes, followed by thematic analysis. Data were collected from students of diverse backgrounds and were subjected to exploratory factor analysis and confirmatory factor analysis. Lastly, we applied structural equation modeling using AMOS to validate the hypothesized relationships.
Findings
The results indicated significant impacts of hallucinations, dysfunctional AI use behavior and biases on students' attitudes and perceived productivity, which in turn influence their academic performance.
Originality/value
To the best of the authors’ knowledge, this study is the first to develop and validate a reliable scale to measure the challenges students encounter when using Gen AI tools and the associated risks. Drawing on the theory of epistemic risk, this study conceptualizes hallucinations, dysfunctional AI usage behavior and biases not merely as technical flaws but as epistemic predispositions that affect users' ability to improve their academic performance when interacting with Gen AI. Hence, this research translates and extends the theory of epistemic risk by converting abstract constructs into measurable instruments that can be validated across diverse contexts.