Rechtfertigung durch lernende Systeme

Erklärbarkeit als Element einer neuen Begründungsarchitektur im Recht

Book cover: Rechtfertigung durch lernende Systeme

Extent: 280 pages

Publisher: Wiley-VCH

Series: Grundlagen der Rechtswissenschaft

Language: German

Paperback (Forthcoming)

(February 2027)

ISBN: 9783162010056

6.1 x 9.13 inches

Price: $128.00

When machines are involved in decision-making, the state must be able to explain why. Arian Henning shows how AI-assisted decisions can be justified in accordance with the rule of law and demonstrates which explanations actually work in practice.


Artificial intelligence is increasingly being used in the exercise of public authority. However, the complex decision-making logic of learning systems can hardly be reconstructed in individual cases. This poses a direct challenge to a state governed by the rule of law, which is founded on the principle of justification, since the traditional instruments of legal reasoning risk reaching their functional limits. The European legislature has responded to this challenge: the AI Act, adopted in 2024, requires deployers of such systems for the first time to provide meaningful explanations, but leaves open what these explanations must contain and how they relate to the rule-of-law obligation to provide reasons.
This is where Arian Henning’s study begins. Legal theory has traditionally bridged the gap between the making and presentation of legal decisions by focusing not on the actual determinants of a decision, but on the normative reconstruction of the reasons relevant to it. Could this approach also be applied to algorithmic decisions? Answering this question brings law to the interface with computer science. Its explainability methods are not readily compatible with the requirements of justification under the rule of law. The author therefore translates the central procedural functions of legal reasoning into the logic of algorithmic explanation models and develops a normative framework for a new architecture of legal reasoning.
A behavioral science experiment involving 4,250 participants tests this concept in the context of administrative proceedings. It reveals a finding that has so far received little attention: in algorithmic decision-making, counterfactual explanations lead to disproportionately large gains in perceived fairness and significantly reduce the fairness gap compared with human decisions.
The study thus combines a normative theory of explainability with empirical evidence showing which explanations enhance the fairness and acceptance of algorithmic decisions by public authorities. It was awarded the Science Prize of the German Foundation for Law and Computer Science and the Doctoral Thesis Award of Bucerius Law School.