The implementation of ethics review processes is an important first step for anticipating and mitigating the potential harms of AI research. Its long-term success, however, requires a coordinated community effort, to support experimentation with different ethics review processes, to study their effect, and to provide opportunities for diverse voices from the community to share insights and foster norms.
As artificial intelligence (AI) and machine learning (ML) technologies continue to advance, awareness of the potential negative consequences on society of AI or ML research has grown. Anticipating and mitigating these consequences can only be accomplished with the help of the leading experts on this work: researchers themselves.
This is an interesting discussion about Articialicial Intelligence, the presence of significant ethical questions, the need for public engagement and a question about the need for research ethics review. We have written a post for the next edition of the Research Ethics Monthly. There are a few areas where ethical reflection is required, but there are no obvious ways in which to handle the reflection on these matters. This is NOT another job for HRECs or AECs, but we do need a mechanism and some form of ethical guidance.
Although these initiatives are commendable, they have yet to be widely adopted. They are being pursued largely without the benefit of community alignment. As researchers and practitioners from academia, industry and non-profit organizations in the field of AI and its governance, we believe that community coordination is needed to ensure that critical reflection is meaningfully integrated within AI research to mitigate its harmful downstream consequences. The pace of AI and ML research and its growing potential for misuse necessitates that this coordination happen today.
Writing in Nature Machine Intelligence, Prunkl et al.1 argue that the AI research community needs to encourage public deliberation on the merits and future of impact statements and other self-governance mechanisms in conference submissions. We agree. Here, we build on this suggestion, and provide three recommendations to enable this effective community coordination, as more ethics review approaches begin to emerge across conferences and journals. We believe that a coordinated community effort will require: (1) more research on the effects of ethics review processes; (2) more experimentation with such processes themselves; and (3) the creation of venues in which diverse voices both within and beyond the AI or ML community can share insights and foster norms. Although many of the challenges we address have been previously highlighted1,2,3,4,5,6, this Comment takes a wider view, calling for collaboration between different conferences and journals by contextualizing this conversation against more recent studies7,8,9,10,11 and developments.
Srikumar, M., Finlay, R., Abuhamad, G., Ashurst, C., Campbell, R., Campbell-Ratcliffe, E., Hongo, H., Jordan, S.R., Lindley, J., Ovadya A. & Pineau, J. (2022)
Advancing ethics review practices in AI research. Nature Machine Intelligence 4, 1061–1064. https://doi.org/10.1038/s42256-022-00585-2
Publisher (Open Access): https://www.nature.com/articles/s42256-022-00585-2
