Societal Responsibility of Artificial Intelligence. Группа авторов

des libertés)⁴ on the ethical issues of algorithms and AI, drafted in 2017 (see Appendix 2);

       – the report published by the French member of parliament Cédric Villani (2018) on AI;

       – the Declaration on Ethics and Data Protection in the Artificial Intelligence Sector, at the 40th International Conference of Data Protection and Privacy Commissioners (ICDPPC), Tuesday, October 23, 2018, in Brussels;

       – the seven guidelines⁵ developed by the European High Level Expert Group on AI, published on April 8, 2019 by the European Commission;

       – the five principles set out in the OECD Council Recommendation on the development, implementation and use of AI, adopted on May 22, 2019, by the Council to OECD Ministers⁶.

      What is the best practice of ethical frameworks, regulations, technical standards, and best practices that are environmentally sustainable and socially acceptable? It is clear that these shared frameworks do not guarantee success. Mistakes and illegal behavior continue to occur. But their availability requires a clear and precise idea of what needs to be done and how to evaluate competing solutions.

This diversity of approaches and initiatives on the subject reflects the major challenge of establishing a common framework for ethical governance of AI. This raises a delicate and decisive question: how should the ethical governance of AI be defined, or by which characteristics? What are the “measurable” values, translating notions of loyalty, responsibility, trust and thus ethics applied to algorithmic decisions when they are the consequence or the result of a prediction?

It is from this vision of universalization that we felt the need to write this book around the framework of AI applicable to all. As a result, we have developed a moral framework to support digital AI projects by observing a number of requirements, recommendations and rules, elaborated, verified and discussed at each stage of design, implementation and use. This allowed us to design ethical criteria, according to our determinants, both essential and universal, based on the principle of Ethics by Design⁷ or Human Rights by Design to move toward a totally innovative principle of Ethics by Evolution that we will develop throughout this book. The objective is to achieve AI that is safer, more secure and better adapted to our needs, both ethical and human, over time. This will help optimize our ability to monitor progress against criteria of sustainability and social cohesion. AI is, therefore, not an end in itself, but rather a means to increase individual and societal well-being.

      ETHICS BY DESIGN.–

      An approach that integrates ethical requirements and recommendations from the design of NICTs.

      ETHICS BY EVOLUTION.–

      It is an approach that incorporates recommendations and ethical rules, in an evolutionary manner over time, throughout the lifecycle of NICTs, i.e. until its implementation and evolutionary use.

This book is intended to categorize ethical issues related to the digital environment, both from the point of view of the user and the designer of digital solutions and/or services. It invites reflection (what questions businesses can ask themselves about digital ethics) and suggests avenues for action. It is an approach that aims to provide guidelines to bring out the values that we want to collectively put forward to help legislators to formulate laws that will build a framework for AI. This repository is not exhaustive. It is intended to be general, open to all contributions and evolving. It must be regularly updated to ensure its consistency and constant relevance as the digital environment and our technological knowledge evolves. It is intended as a reminder of the company’s regulatory duties, which precisely define what is permitted or prohibited, and the sanctions that apply. The company has an obligation to comply, and this does not concern the area of ethics. However, the means by which it complies can be the subject of ethical reflection.

      Finally, this book is addressed to all stakeholders involved in the development, deployment or use of AI, including organizations, companies, public services, researchers, individuals or other entities. This document should, therefore, be considered as the first building block of a discussion between these different actors toward an ethical, responsible, trustworthy AI aimed at protecting and serving in a beneficial way individuals and the common good for a better adoption at the global level.

      1 1 In France, crowdsourcing is defined according to the Commission générale de terminologie et de néologie (2014) as the “mode of completion of a project or a product calling for contributions from a large number of people, generally Internet users”. JORF, 0179(91), 12995.

      2 2 ISO 2382-28:1995 defines artificial intelligence as “the capability of a functional unit to perform functions that are generally associated with human intelligence, such as reasoning and learning”.

      3 3 IEEE P7000: Model Process for Addressing Ethical Concerns During System Design; IEEE P7001: Transparency of Autonomous Systems; IEEE P7002: Data Privacy Process; IEEE P7003: Algorithmic Bias Considerations; IETF Research into Human Rights Protocol Considerations draft.

      4 4 CNIL (2017). Comment permettre à l’homme de garder la main ? Les enjeux éthiques des algorithmes et de l’intelligence artificielle. Summary report of the public debate led by the CNIL in the context of the mission of ethical reflection entrusted by law for a digital Republic.

      5 5 These seven essential requirements include human factor and human control, technical robustness and security, privacy and data governance, transparency, diversity, non-discrimination and equity, societal and environmental well-being, and accountability.

      6 6 On May 22, 2019, through the OECD Council of Ministers, 42 countries (the 36 OECD countries and Argentina, Brazil, Colombia, Costa Rica, Peru, and Romania) adopted the principles set out in the OECD Recommendation on AI, making it the first intergovernmental agreement to stimulate innovation and build confidence in AI by promoting a responsible approach to trusted AI, while ensuring respect for human rights and democratic values.

      7 7 This consists of integrating ethical rules and requirements from the design and learning of these NICTs, prohibiting direct or indirect damage to the fundamental values protected by the conventions.