Abstract
Artificial intelligence is rapidly changing the economics of academic misconduct. Generative AI reduces the effort required to produce high-quality outputs while simultaneously lowering confidence in detection tools. Traditional policy responses have focused on punishment and detection; however, when the probability of detection declines, deterrence alone becomes less effective. This paper proposes a behavioural economic model that extends Becker’s crime and punishment framework (Becker, 1968) by incorporating three additional mechanisms: assessment design, identity and moral costs, and the perceived value of learning.
In this model, students engage in misconduct when the short-term performance gain from shortcuts exceeds the combined expected cost of punishment, internal identity costs, social norm costs, and the perceived long-term value of learning. Generative AI increases the payoff to shortcuts and may reduce perceived detection risk, shifting student behaviour unless institutional design changes simultaneously address incentive structure and learning value.
To test this model, we propose a field experiment in undergraduate business courses that randomizes three intervention types: (1) assessment design (traditional product-focused assignments versus process-verified assessments with checkpoints and short oral reasoning verification), (2) detection certainty (standard policy statements versus credible audit), and (3) behavioural integrity interventions. Outcomes include both misconduct prevalence, and deep learning outcomes, measured through transfer tasks and delayed retention tests.
The goal is to identify scalable policy bundles that reduce misconduct while improving deep learning in AI-enabled learning environments. Results will inform institutional academic integrity policy and contribute to behavioural public policy approaches to education system design.