How Much AI in Courts is Too Much? Conjoint Experiment on Citizens Trust in AI in Decision Making

AI tools are increasingly entering justice systems around Europe, with a varying degree of official communication and regulatory oversight. Based on the official reports, there are more than hundreds of (official) AI tools already in use across judiciaries in Europe. Informally, judges and law clerks are also integrating AI into their daily work in the absence of clear guidelines, transparency or understanding of their potentials and limitations – the so-called ‘shadow AI use’. Whether formal or informal, the use of AI tools by judges and courts has raised the issue of trust. Signals of the lack of trust or even mistrust in the use of AI tools in justice have been emerging across Europe. Yet, recent findings show that citizens do not oppose AI tools per se, but are simply more likely to accept decisions supported by AI tools when they believe these tools enhance efficiency and quality of delivered services while at the same time do not undermine accountability and human oversight. The key question, therefore, is not whether citizens trust or distrust the use of AI technologies in courts, but rather how much reliance on AI tools is acceptable in decision making on courts for citizens to continue to perceive the decision-making process trustworthy? In other words, how much AI is too much in courts? This paper addresses this question by empirically examining citizens’ trust in the use of AI technology in courts through a conjoint experiment, with the aim of identifying the conditions under which citizens trust the use of AI technology in decision making.