GUIDELINES FOR ENGINEERS AND THEIR ORGANIZATIONS The guidelines below have organically emerged from discussions with engineers as well as from an overview of other ongoing efforts to address the issues of AI and ethics. These are not exclusively for individual engineers to follow, because ethical development of AI will not come about only as a result of individuals taking on particular types of ethical responsibility. There are plenty of guidelines for what constitutes ethical conduct for engineers and some of the guidelines below can be taken on board by individuals and organizations alike as additions to those that are already in existence in the Nordic countries. However, many of the guidelines are oriented towards organizational practices rather than individual responsibility, because efforts towards ethical practices need strong institutional backing to be effective and therefore organizational commitment is a requirement for addressing ethics in AI. We present these guidelines with an understanding that their implementation will require effort and commitment on the part of the individual engineers and of their organizations together. GUIDELINES OF ETHICAL CONDUCT FOR AI DEVELOPMENT AND IMPLEMENTATION 1. Create spaces for discussion of the issues around AI and ethics. These need to be facilitated and supported by both workplaces and civil society organizations. 2. Invest into and develop tools that enable ethical discussions, questions and decision making throughout the design process and not only at the beginning and the end. 3. Establish a set of internal standards and checklists tackling ethical issues in AI development such as ensuring meaningful human control. 4. Support and facilitate internal reporting of risk and violations, establishing rules for clear action in response. 5. Establish internal training programs for staff to deepen an understanding of ethics and to develop skills for ethical reflection, debate and recognition of biases. 6. Pay special attention to potential biases encoded in system development, training data and model performance, especially those that may affect the most vulnerable. 7. Develop ways for accepting organizational responsibility for potential harm, for example, by establishing ways to address the harm inflicted on others by AI systems that the organization has built. 8. Establish an internal ethical review process that democratizes company decision-making by involving more internal actors. 5
Download PDF fil