Credit-Scoring Paper Argues AI Decisions Need Legal Justification

A technical explanation can identify influential variables, describe a model's logic, or show why one applicant received a particular score. The paper's central argument is that this still may not establish whether the resulting decision is lawful. A legally meaningful justification must connect the automated outcome to the rules governing the credit relationship and give the affected person a basis for contesting it.

Gorski therefore rejects a narrow reading of a right to explanation and favors a broader interpretation that includes justification. The paper uses credit scoring as its concrete example and places the issue in the context of European legal protections. It is a doctrinal and interdisciplinary analysis, not a report of a newly trained credit model, a borrower experiment, or measured gains from a deployed system.

For data and governance teams, the argument points to a gap between model-interpretability tooling and decision accountability. Feature-attribution charts, reason codes, and local explanations may help describe model behavior, but they do not by themselves show that a lending rule is legally permitted, consistently applied, or open to an effective remedy.

A stronger implementation would preserve both layers: technical evidence about the model and a decision record tying the outcome to an approved legal and policy basis. That framing does not settle how regulators or courts will interpret any specific case, but it offers a useful design test for high-stakes AI systems: can the organization defend the decision, not merely reconstruct the prediction?