The Report in 10 Pages norms, then it is vital to act now to legal proceedings, during an rally round the scientific community, investigation undertaken by an public authorities, industry, business independent administrative authority owners and civil society organizations. or on request by the Defender of Our mission has endeavored to put Rights (Défenseur des Droits). forward some humble suggestions Implementing Ethics by Design that could lay the foundations for the ethical development of AI and Research staff, engineers and promote debate on this issue within business owners who contribute to society at large. designing, developing and marketing Opening the Black Box AI systems play a decisive role in tomorrow’s digital society, so it is vital A large proportion of ethical that they act responsibly and factor in considerations are raised by the lack the socio-economic effects of their of transparency of these technologies. actions. With this in mind, it is AI provides spectacular results for important to make them aware of the reasons that researchers sometimes ethical issues involved in the have difficulty to explain: this is known development of digital technologies as the black box phenomenon, where right from the start of their training. we can see input data and output data This aspect is lacking in today’s for algorithm-based systems, but we courses at engineering school and in do not really understand what exactly universities’ IT programmes, yet the happens in between. AI can extent and complexity of ethical reproduce bias and discrimination issues these future graduates will face and is becoming increasingly present continue to grow. in our social and economic Looking beyond engineer training, environments, so opening the black ethical considerations must be fully box is a key democratic issue. factored into the development of Explaining machine-learning artificial intelligence algorithms. A algorithms has become a very urgent discrimination impact assessment matter and is now actually a separate could be introduced, similar to the field of research, which must be privacy impact assessments already supported by public authorities. made compulsory by General Data Three areas in particular require an Protection Regulation for some data extra focus: obviously the production processing. The overarching aim here of more explicable models, but also is very simple: have AI developers the production of more intelligible consider the right questions at the user interfaces and an understanding right time. of the cognitive mechanisms used to More broadly speaking, the produce a satisfactory explanation. increasing use of AI in some sensitive Transparency is clearly key, but areas such as policing, banking, looking beyond this issue, it is also insurance, the courts and in Defense vital to facilitate audits of AI systems. (with the question of autonomous This could involve the creation of a weapons) raises a real society-wide group of certified public experts who debate and implies an analysis of the can conduct audits of algorithms and issue of human responsibility. We databases and carry out testing using must also consider the role of any methods required. These experts automation in human decisions: are could be called on in the event of there areas where human judgement,