This chapter examines questions concerning the ethical oversight of Machine Learning (ML), focusing on approaches that emphasize the principles of transparency, fairness, and explainability. I begin in Sect. 8.2 by describing the most dominant approach to ML ethics, which focuses on finding technical means of implementing abstract ethical principles. In Sect. 8.3, I probe the dominant cognitive conception for understanding ML. The primary contention here is that the cognitive framing is useful for developers in creating these programs and politically useful for the organizations these developers operate within, but that it obfuscates the relevant factors at play when the conception is exported into ethical or regulatory discussions. Sections 8.4 and 8.5 delve into two alternative conceptions of this class of programs, one from the sociologist Elena Esposito that attempts to frame algorithms as communicative agents, which allow various systems and individuals to interface with another through constructing what she calls a ‘virtual double contingency’, and another framing from political geographer Louise Amoore, which aims to center the discussion of algorithms and machine learning within the always socio-political context the operate within and upon which they are empowered to act. In Sect. 8.6 I apply these framings to a toy case and describe where they fall short in illuminating the ethical landscape. Finally, I conclude in Sect. 8.7 by pointing to areas I believe require more attention from philosophers if a coherent understanding of ML is to be reached.