Psychological features of dispute content and public acceptance of AI in legal adjudication: evidence for systematic variation beyond individual differences

Does the Case Matter More Than the User?

Why do some people embrace AI in the courtroom while others recoil? As legal systems globally begin to integrate artificial intelligence for case management and outcome prediction, the success of these tools hinges on social acceptance. Most research suggests that acceptance is driven by the individual—their personality, their age, or their general attitude toward technology.

However, a new study from Masahiro Fujita and Eiichiro Watamura suggests we have been looking at the wrong side of the equation. People do not just decide to trust AI based on who they are. They decide based on what kind of problem is being solved. For rule-based issues like contract disputes, people are more open to AI. For emotional issues like family disputes, they strongly prefer humans.

Beyond the individual user

The researchers set out to investigate whether the psychological characteristics of the dispute itself—the "content"—systematically shape whether a person prefers an algorithm or a human adjudicator. Specifically, they wanted to know if legal disputes possess inherent dimensions that dictate the appropriate decision-making authority.

The study asks if citizens categorize legal problems into different mental buckets before they evaluate the technology. If a dispute is seen as a matter of applying rigid rules, an algorithm might feel like a natural fit. If a dispute is seen as a matter of navigating human suffering, an algorithm might feel fundamentally inappropriate.

The cracks in technology acceptance models

Until now, the field of technology acceptance has relied heavily on models that focus on the person. Researchers have looked at how traits like "openness to experience" or "conscientiousness" predict whether someone will use a new tool. While these individual differences are real, the authors argue they represent an incomplete picture.

By focusing almost exclusively on the user, the field has neglected the "situation." As shown in the conceptual framework in, the authors propose that the decision-making context—specifically the emotional weight and the typicality of a case—interacts with the user to produce a final judgment. The old model treated acceptance as a static trait of the person. This study treats it as a dynamic response to the task at hand.

Mapping the legal landscape

To test this, the authors conducted two studies with Japanese participants. In Study 1, they presented 1,384 people with 46 different legal "vignettes"—short, single-sentence descriptions of various disputes. These ranged from traffic accidents to child custody battles. They used exploratory factor analysis (a statistical method to uncover hidden patterns in data) to see how these cases grouped together in the minds of the public.

In Study 2, the researchers moved from observation to experimentation. They took an independent sample of 596 participants. They deliberately manipulated two features: emotional involvement (making a case feel personally painful versus fact-based) and prototypicality (making a case feel like a common, standard occurrence versus a rare, unique event). This allowed them to see if they could shift a person's preference for AI or humans by changing how the case was described.

A two-dimensional divide

The results revealed a remarkably stable two-dimensional structure in how people organize legal problems. The authors report that disputes fall into two distinct categories: "Institutional-Procedural" and "Interpersonal-Relational."

The first dimension, Institutional-Procedural, includes cases like patent infringement or regulatory violations. The study finds that AI acceptance is comparatively higher in this domain. This is because these cases are perceived as being governed by formal, objective rules. The second dimension, Interpersonal-Relational, involves family conflicts, violent crimes, and child welfare. In these scenarios, the authors report a strong, consensus-driven preference for human judgment. This stems from the perceived need for empathy and relational understanding.

The experimental results in Study 2 added a layer of complexity. The authors found a significant three-way interaction between emotional involvement, gender, and prototypicality, as illustrated in . For instance, under "prototypical" conditions (cases that feel common and standard), women showed a much stronger preference for human judges when a case was framed with high emotional involvement. Interestingly, AI-specific expectations—how much a person expects an AI to actually be able to do—emerged as the strongest predictor of acceptance. This factor's effect size ($\eta^2 = 0.252$) far outweighed the predictive power of personality traits.

Implementation and communication

If these findings hold across different cultures, they suggest that deployment strategies should be tailored to dispute characteristics. Instead of a uniform approach, the authors discuss several practical implications for implementing legal AI.

One suggestion is a phased implementation. Developers might prioritize "low-hanging fruit" in institutional and procedural domains. These are areas where public acceptance is higher and more stable. Conversely, in interpersonal disputes involving family or child welfare, the study suggests that maintaining human oversight may be essential for legitimacy.

Second, communication is key. Since expectations about AI capabilities are the strongest drivers of acceptance, generic campaigns are unlikely to work. Instead, institutions might benefit from addressing specific capabilities and limitations of AI systems. This could help calibrate public expectations more effectively than broad awareness campaigns.

The paper acknowledges several limitations. These include the use of brief vignettes, which may not capture the messy reality of a real courtroom. The study was also conducted in Japan. A vital next step for researchers would be to use "process-tracing" methods. These include eye-tracking or reaction-time studies to see if these mental categorizations happen automatically or through deliberate reasoning.

Figures from the paper