In this talk, I will argue that we should study AI ethics from the perspective of complex systems. In particular, ML systems are not islands. To understand and mitigate the risks and harms associated with ML systems, we need to examine the broader complex systems in which ML systems operate. By broader complex systems, I mean our social, economic, and political systems [1]. Thus, we must remove our optimization blinders. That is, we should not focus only on maximizing some notion of constrained expected utility. I will provide examples from the impact of COVID-19 interventions on amplifying racial disparities in the U.S. criminal justice system [2], the impact of misinformation on democracy [3][4], the complexities of interventions for information access equality [5], and time-permitting the use of algorithms for school admissions [6][7]. All references are available at http://eliassi.org.