As the ethical questions associated to AI projects, many of which relate to Data Science approaches, are becoming more and more debated outside of the academic word, one of the largest AI association in the word, AAAI, has introduced this year a new conference explicitly to look at those aspect from a cross-disciplinary research perspective: The AIES conference on AI Ethics and Society.
As part of this conference, several members of the Data Ethics Ireland community presented a paper which describes ongoing efforts to develop a methodology for AI developers and Data Scientists to integrate ethical considerations in their design. This methodology is referred to as an “AI by Design” methodology as it extracts from a number of case studies (including two presented in the paper and extracted from the magnacartafordata.org website) a set of requirements and principles for such a methodology, including:
- That ethics should be considered all along the Data Science/AI development process, including design, operationalisation and deployment.
- That ethics should be considered within a constructive and reflexive process.
- That ethics should be considered through anticipation of possible scenarios, and not only through the direct and expected effects of the design and realisation of the research.
- The ethics has to be considered as a conversation between the technical and the social aspects.
The last point is especially important as ethics in Data Science and AI is too often considered as a purely technical problem, looked at by computer scientists without sufficient background to address the social implications of technical solutions, or as a social phenomenon by social scientists who might not have a sufficient grasp of the technology to offer appropriate solutions. This paper has been written by authors from several organisations in Ireland and the UK, including AI/Data Science researchers, communications specialists, a IPR/Data Protection practitionner and a surveillance studies expert.