AI: making decisions contestable, not merely explaining them
An AI-assisted decision can be understandable while still leaving the person affected with no meaningful opportunity to have an objection examined. The managerial question is precise: how can a review process be organised so that it can actually change the decision, rather than merely receive complaints?
What the literature establishes
E. Allan Lind, Ruth Kanfer and P. Christopher Earley (1990, Journal of Personality and Social Psychology) show experimentally that having an opportunity to express one's views contributes to judgments of procedural justice, even when it does not provide control over the outcome. Being heard therefore matters for reasons beyond securing a favourable result. However, this study examines neither AI nor the effectiveness of an appeals procedure.
Explanation and contestability: two distinct dimensions
Jason A. Colquitt (2001, Journal of Applied Psychology) provides validation evidence for distinguishing distributive, procedural, interpersonal and informational justice. In the measure examined, the opportunity to appeal an outcome belongs to the procedural dimension; the quality of explanations belongs to another dimension. For managers, this suggests that a clear justification should not be treated as a substitute for an accessible review.
The premature conclusion
One might infer that adding a challenge form is enough to make an AI-assisted decision fairer. These studies do not demonstrate that a mechanism which improves perceived fairness actually corrects more errors. The opportunity to speak and the ability to secure a revision of a decision therefore need to be examined separately. (our executive and employee training programmes)
What the evidence cannot promise
These articles examine fairness judgments and their measurement, not appeals against decisions made with generative AI. They establish neither an optimal response time nor a universal review model. In practice, a low number of challenges remains ambiguous: it may reflect satisfactory decisions, but also an unfamiliar or inaccessible procedure.
A practical check in Berne
Within SHR's “Managing in the Age of Artificial Intelligence” programme, one exercise is to select a category of AI-assisted decisions in Berne: routing requests in the federal administration, administrative responses in public health, handling telecoms complaints or compliance decisions in precision manufacturing. During a pilot period defined in advance, each challenge received would be tracked through to a reasoned response, with an identified reviewer authorised to change the decision, without replacing applicable appeal procedures. The proposed measure is the proportion of challenges receiving this documented review within the announced timeframe, accompanied by the observed response time and the number of decisions changed. An accessibility test using a fictitious case would complement this monitoring: a procedure with no recorded challenges is not proof that it is usable. To go further: explore the Managing in the Age of Artificial Intelligence training in Bern in the canton of Bern, or browse our executive and employee training programmes in Switzerland.
In pictures: Managing in the Age of Artificial Intelligence in Bern
- managing in the age of artificial intelligence
- Bern
- research
- training bern