Project SHERPA guidelines

The SHERPA project provides a set of guidelines for both AI users and developers, oriented towards operationalising ethical processes. They contain a useful expression of underpinning values, and provide e.g. a project-cycle model to outline process. “This report contains ethical guidelines for the technological development of artificial intelligence (AI) and big data systems. It is a Deliverable of the SHERPA project, an EU Horizon 2020 project on the ethical and human rights implications of AI and big data. The guidelines differ from others in that they are directly related to design and development practices. They are intended to be actionable guidelines for systems and software development, rather than abstract principles that have no direct application in practice. We call such guidelines operational, meaning ready for use. Applying these guidelines in development practices would result in more ethical AI and big data products. In constructing Guidelines for the Ethical Development of AI and Big Data Systems: An Ethics by Design approach, we have incorporated input from a wide diversity of stakeholders, SHERPA partners, and insights from other guidelines. In a survey of potential guidelines we found over 70 matching documents, which were reduced to 25 suitable guidelines that we built on. After an introductory section, we devote Section 2 of this report (“High-Level Requirements”) to present and discuss the high-level requirements that form the point of departure for this report. Our requirements are directly based on the guidelines of the EU’s High-Level Expert Group on Artificial Intelligence (HLEG AI), with minor adaptations to improve coherence and fitness for operationalization. This results in the following seven requirements that mirror those of the HLEG AI: human agency, liberty, and dignity The human right to be valued and treated with respect because of one's personhood. ; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination, and fairness; individual, societal, and environmental wellbeing; and accountability. For each, we specify three to four sub-requirements that constitute a first step towards operationalization.” (Brey et al., 2020, p. 2)“The guidelines that we present in this report are operational in the sense that they are, in our view, ready to be used by ethics officers or managers, who have a responsibility for ensuring the implementation of ethical practices within their organizations. They are perhaps not directly usable by system developers. A further step that is required, but not contained in this report, is the training of developers in this new framework, and the assignment of different roles and responsibilities to them for ensuring that the ethical requirements are met. This may also require the development of training materials and operational guides for professionals with different roles in the development process. We intend to produce further implementation documents in the EU Horizon 2020 SIENNA project (www.sienna-project.eu).” (Brey et al., 2020, p. 3)