This essay was originally published as a section in the paper Blue Sky Ideas in Artificial Intelligence Education from the EAAI'17 New and Future AI Educator Program by Eaton et al.
Ethics, the moral principles that govern a person's or group's behavior, cannot be incorporated into a curriculum around artificial intelligence (AI) without a systematic revision of the surrounding context within which AI takes place. We must go beyond just talking about ethics in the classroom; we need to put ethics into practice. I offer three recommendations for doing so, drawn from how ethics are treated within engineering and the social sciences.
Firstly, the Association for the Advancement of AI (AAAI) should institute an association-wide code of ethics. This recommendation is inspired by ethics codes in engineering, which include concern for the public good as a constituent part. For instance, the code of ethics of the National Society of Professional Engineers (2007) contains seven fundamental canons, the first of which is: "Engineers, in the fulfillment of their professional duties, shall hold paramount the safety, health and welfare of the public." An association-wide code of ethics would formally recognize our impact in and the responsibility that we owe to our society.
Secondly, research funding applications that deal with AI should be required to assess risks to society. This recommendation is inspired by similar requirements by Institutional Review Boards (IRB) within the social sciences (e.g., U.S. Dept. of Health and Human Services 2009). Whenever researchers conduct studies that deal with human participants, they are asked by an IRB to assess sources of potential risk; AI research applications should do the same. Importantly, these risk assessments should consider threats beyond immediate physical harm; e.g., the development of new analytical tools for understanding large amounts of data may inadvertently make it easier to reconstruct personally identifiable information, which constitutes a threat to anonymity, and which may disadvantage vulnerable populations.
Thirdly, students in AI project-based courses should be required, as part of the class' deliverables, to submit documents that assess the impact to society (in the context of the proposed AAAI code of ethics, and which should include an IRB-like risk assessment). Ideally, AAAI would serve as a facilitator of this kind of assessment, by providing a library of case studies and expert testimonies that can guide students in examining the broader implications of their work.
Incorporating ethics into a curriculum is more than a one-shot affair. It requires a systematic revision of the surrounding context within which AI exists, in terms of how we talk about it (first recommendation), how we fund efforts in it (second recommendation), and how it is put into practice (third recommendation). By leveraging existing models on ethics from engineering and the social sciences, we will be better equipped to offer concrete recommendations to ensure that ethics aren't an afterthought, but are integral to the development of AI.
National Society of Professional Engineers. 2007. Code of Ethics for Engineers. Technical Report 1102.
U.S. Department of Health and Human Services. 2009. 45 CFR 46.111: Criteria for IRB approval of research. Technical report.