AI in Taxation — Experimental evidence on citizen design preferences and perceptions of legitimacy
通过两项实验（N=1072）研究芬兰公民对税收中AI的设计偏好和合法性感知，发现公民对AI拥有决策权持怀疑态度，且私人公司参与或使用私人数据会进一步削弱感知合法性。
This study investigates citizen preferences for the design of artificial intelligence (AI) in public administration and citizen legitimacy perceptions of AI-based tax-decisions. Analysing citizens’ perceptions of AI is important because the literature remains inconclusive and negative perceptions may undermine AI’s performance benefits. Prior research suggests that these perceptions depend on contextual, individual-level, and design-related factors. Empirically, we investigate the use of AI in taxation in Finland, a country context with high institutional trust and established use of automated decision making in taxation. Using two separate experiments with a conjoint design and a vignette design drawing on a randomly recruited citizen sample (N=1072), we offer robust evidence on the relative importance of AI decision-making characteristics and investigate perceptions of individual-level AI tax decisions. The results indicate that citizens are sceptical of AI in taxation when AI systems are granted decision-making discretion. Perceptions are further undermined when private companies are involved in developing public sector AI or when private data are used. The results have implications for the design of public sector AI, highlighting which features should be considered to ensure its use is perceived as legitimate by citizens.