The effects of AI-based detection on misbehaviors : empirical evidence from AIPark

Abstract

Though it is well known that AI could facilitate production activities in private sectors, few studies have examined the role of AI in public administration. In this paper, I utilize data from AIPark, one of the lead companies in the industry of smart parking, and empirically investigate how AI-based detection affects misbehaviors exemplified by illegal parking, which is aimed at filling the gap between industrial practice and academic research in the field. Employing the staggered activation of monitoring cameras by traffic police, I construct DID and RDD models, respectively. Empirical results from DID model show that AI-based detection has effectively reduced the number of illegal parking by about 6% (or 4%, as implied by the RDD model). Both estimates remain stable under a series of robustness checks and collectively indicate the significant deterrence effect resulting from AI-based regulation. Further analyses show that (i) the AI-based detection has not only reduced illegal parking behaviors but also improved the efficiency of parking facilities; (ii) the treatment effect is stronger for drivers of less expensive cars; (iii) drivers of cars registered locally response more sensitively to the enhanced regulation; (iv) the treatment effect is weaker during peak hours; (iv) regular commuters seem to be less affected by the AI-based regulation.