DIN SPEC 91512:2024-11 - Draft

This DIN SPEC has been prepared by a DIN SPEC consortium (temporary committee) in the course of the PAS procedure. The preparation and adoption of this DIN SPEC has been carried out by the authors named in the foreword. DIN SPEC are published for comment for two months free of charge. This document is applicable to AI applications in the context of financial services. This document specifies the following: - a process for determining appropriate perceptions and metrics of fairness (of AI applications) in the financial sector; - a concrete understanding of fairness may vary from person to person or depending on the situation and therefore cannot be formulated in a universally valid manner; - to this end, a basic framework for application- and situation-specific measurements and justifications in the context of financial services in practice is proposed, including: - a justification for the use of concepts such as equal treatment, equal rights, and equality; - relevant aspects of good measurement and justification of all concepts used; - assessment of proportionality in terms of effort and results; - handling of measurement results (evaluation, explanation, communication, need for action); - acceptable framework conditions for unequal treatment; - collection of methods for operationalizing the identified concepts (for example, via fairness measures). Identification of the respective strengths, weaknesses, and limitations of the operationalization approaches; - exemplary illustration of the process using relevant use cases. Furthermore, in the context of its use cases, this document is limited to: - classification systems (exclusion of scorers and generative systems), - group fairness measures (exclusion of individual fairness and alternative fairness measures). This document does not describe processes for adapting AI-based systems to meet fairness requirements.