Generative AI: a quantitative study on emerging risks and impacts
Résumé
This study examines the multifaceted risks associated with Generative AI (GAI) and their impacts on societal, organizational, environmental, and individual levels. Employing binary logistic regression analysis on data from the OECD AI Incidents Monitor, analyzing 858 incidents, we explore the relationships between various dimensions of GAI risks and their potential impacts. Our study reveals critical insights: data governance issues have significant effects across all examined levels, with the most significant positive effect observed at the individual level (particularly regarding privacy and disinformation incidents). While broadly influencing various levels, content generation issues exert the most significant positive effects on individuals (specifically psychological well-being and disinformation problems), organizations (reputation risk), and society (social cohesion issue). Furthermore, social and environmental concerns show a heightened positive impact on individuals (particularly quality of life incidents), organizations issues, and society (economic stability and social cohesion problems). The study advocates for future research to develop a dynamic framework for responsible GAI risk management.