OBJECTIVES: Machine learning (ML) has the potential to facilitate "continual learning" in medicine, in which an ML system continues to evolve in response to exposure to new data over time, even after being deployed in a clinical setting. In this article, we provide a tutorial on the range of ethical issues raised by the use of such "adaptive" ML systems in medicine that have, thus far, been neglected in the literature. TARGET AUDIENCE: The target audiences for this tutorial are the developers of ML AI systems, healthcare regulators, the broader medical informatics community, and practicing clinicians. SCOPE: Discussions of adaptive ML systems to date have overlooked the distinction between 2 sorts of variance that such systems may exhibit-diachronic evolution (change over time) and synchronic variation (difference between cotemporaneous instantiations of the algorithm at different sites)-and underestimated the significance of the latter. We highlight the challenges that diachronic evolution and synchronic variation present for the quality of patient care, informed consent, and equity, and discuss the complex ethical trade-offs involved in the design of such systems.

Most robots currently being sold or developed are either stylized with white material or have a metallic appearance. In this research we used the shooter bias paradigm and several questionnaires to investigate if people automatically identify robots as being racialized, such that we might say that some robots are 'White' while others are 'Asian', or 'Black'. To do so, we conducted an extended replication of the classic social psychological shooter bias paradigm using robot stimuli to explore whether effects known from human-human intergroup experiments would generalize to robots that were racialized as Black and White. Reaction-time based measures revealed that participants demonstrated 'shooter-bias' toward both Black people and robot racialized as Black. Participants were also willing to attribute a race to the robots depending on their racialization and demonstrated a high degree of inter-subject agreement when it came to these attributions.

Summary form only given. In February 2015, a video posted online by Boston Dynamics, showing a man kicking their quadruped robot “Spot” in order to demonstrate the robot's capacity to regain its balance, “went viral”. A number of media websites subsequently published stories asking whether it was wrong to kick a robot dog. Most pundits seemed to conclude that there was nothing wrong with kicking a robot as long as doing so didn't encourage the kicker to go on to kick living things. In this presentation I want to revisit this controversy and argue that, according to one plausible ethical framework at least, kicking a robot may be wrong regardless of its impact on the future behavior of the kicker and/or the sufferings of other sentient creatures. An “agent-based virtue ethics” evaluates actions according to what they reveal about the character of an agent. That is to say, kicking a robot dog might be wrong simply because it reveals the kicker to be cruel or vicious in their dispositions. However, what our behavior towards robots says about us will also depend upon the behavior of the robot and, in particular, its capacity to express emotions that in turn determine the reasonableness of our interpretation of its treatment. In designing robots, then, we are also shaping the opportunities that users have to demonstrate virtues and vices in their interactions with the robot. I offer some tentative suggestions as to why it might often be more plausible to attribute vices than virtues on the basis of a person's treatment of a robot. I conclude with some observations about the ways in which this complexifies the project of the ethical design of robots.