Discussions of ethical issues in Artificial Intelligence have moved from the realm of science fiction and academic debate and into headlines and broad public discussion. Researchers, company leaders, and politicians have ideas about what "Ethical AI" should look like. However, too rarely do these ideas take into account the working realities of the engineers building these systems, to whom ethics work is often relegated.

I conduct qualitative studies of an open source community building a deepfake tool; engineers working across disjointed AI supply chains; practitioners who de velop and seek to raise ethical concerns about what they create; and teams discussing AI ethics in a game context. Based on this, my dissertation shows how organizational norms, incentives, and boundaries limit software creators' sense of responsibility and agency over the downstream impact of what they create, and examines the possibilities and shortfalls of play as a way to expand these limits. I find that limited conceptions of and authority in their roles, limited visibility into downstream uses, time pressures, licensing constructs, and other reasons often meanengineers do not feel responsible for–or power to do–ethics work. I conclude by discussing how future tech ethics research, practice, education, and policy can better consider the sense of responsibility and power of those asked to do ethics work in technology,

188 pages

Thesis Committee:
Laura Dabbish (Co-Chair)
James Herbsleb (Co-Chair)
Dawn Nafus (Intel Labs)
Jay Aronson

James D. Herbsleb, Head, Software and Societal Systems Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu