Skip to content
ContestabilityA research
publication

Research note / 2026-11-19

When explaining the AI lowers the wish to appeal, has anyone been given grounds?

Short article · November 19, 2026

When explaining the AI lowers the wish to appeal, has anyone been given grounds?

Kerry Thornhill

What was reported

On 10 November 2026 University College Dublin’s Smurfit School and Trinity College Dublin issued a press release about a study by Christopher Greene, Brian Barry and Marius Claudy, published online on 23 September in the Cambridge Forum on AI: Law and Governance. In an online experiment, 859 UK adults imagined losing a small civil dispute in a court that uses an AI tool to help judges, with the judge keeping the final decision. Some were told nothing about the tool. Others got an explanation of how it works, or of how it reached its recommendation.

Both explanations raised trust. Cognitive trust went from 2.82 to 4.22 on a seven-point scale, and affective trust from 2.72 to 3.60. Intention to appeal fell from 5.26 with no explanation to 4.46 and 4.33. The authors report that respect was the main route from explanation to trust, and that the sense of having a voice contributed very little.

What I make of it

The appeal result was an exploratory analysis. The authors offer two readings: people who received an explanation may have judged the process fairer and accepted the result, or people with none may have expected an unexplained decision to be easier to overturn. They also suggest that part of the effect may come from greater confidence that the decision was correct. Their conclusion is that the focus should move from whether people understand how the AI works to whether they feel the process is respectful and fair.

That is a defensible finding about legitimacy. It is a different question from the one I work on. Where an explanation is the only thing a person receives, the inclination to appeal could fall as much for a mistaken decision as for a correct one. From Notice to Contestable Grounds starts from the observation that notice, explanation and an appeal procedure can still leave a person unable to investigate a determination.

A proposed illustration

This is a hypothetical, not a result from the study. Two litigants lose small claims in a court that uses an AI tool. Both read the same general account of how it works, and both feel respected. In one case the recommendation leaned on a precedent that does not fit the facts. A general account of how the tool works would not point her to that, yet it may lower her intention to appeal as much as it lowers the other litigant’s. The court’s own procedural rules would govern whether either could then investigate.

What the paper does not say

The study measured trust and intentions in imagined cases where everyone lost. I did not find a measure of whether participants could identify an error, and none of the cases had one. The authors note that they tested explanations of the AI tool, not how the judge used it. The vignettes are in a supplementary appendix I have not read.

If an explanation reassures the wronged and the rightly decided equally, what is it for?