Human emotional expressions can communicate the emotional state of the expresser, but they can also communicate appeals to perceivers. For example, sadness expressions such as crying request perceivers to aid and support, and anger expressions such as shouting urge perceivers to back off. Some contemporary artificial intelligence (AI) systems can mimic human emotional expressions in a (more or less) realistic way, and they are progressively being integrated into our daily lives. How should we respond to them? Do we have reasons (...) to reply to the appeals made by AI emotional expressions? In this paper, we examine the conditions under which AI emotional expressions could give us prudential or even moral reasons to change our behavior. We argue that these conditions do not depend on whether the emotional expression is genuine or not, but rather on the presence of features some of which can be implemented in emotive AI given our current level of technological development. We extract recommendations and warnings for the development of emotive AI. (shrink)

Emotive artificial intelligences are physically or virtually embodied entities whose behavior is driven by artificial intelligence, and which use expressions usually associated with emotion to enhance communication. These entities are sometimes thought to be deceptive, insofar as their emotive expressions are not connected to genuine underlying emotions. In this paper, I argue that such entities are indeed deceptive, at least given a sufficiently broad construal of deception. But, while philosophers and other commentators have drawn attention to the deceptive threat of (...) emotive artificial intelligences, I argue that such entities also pose an overlooked skeptical threat. In short, the widespread existence of emotive signals disconnected from underlying emotions threatens to encourage skepticism of such signals more generally, including emotive signals used by human persons. Thus, while designing artificially intelligent entities to use emotive signals is thought to facilitate human-AI interaction, this practice runs the risk of compromising human-human interaction. (shrink)

In this article, I investigate whether algorithmic fairness is an inconsistent concept (the inconsistency thesis). Drawing on the work of Kevin Scharp, inconsistent concepts can apply and disapply at the same time (2.). It is shown that paradigmatic issues of algorithmic fairness fit this description (3.). Similarities and differences to received views (4.) and alternatives to the inconsistency thesis are considered (5.). Suggestions are articulated on how the inconsistency thesis might hold ground nevertheless, or at the very least denotes a (...) distinctive option in argumentative space whose status and implications merit further evaluation. (shrink)

We propose a conceptualization and implementation of AI ethics via the capability approach. We aim to show that conceptualizing AI ethics through the capability approach has two main advantages for AI ethics as a discipline. First, it helps clarify the ethical dimension of AI tools. Second, it provides guidance to implementing ethical considerations within the design of AI tools. We illustrate these advantages in the context of AI tools in medicine, by showing how ethics-based auditing of AI tools in medicine (...) can greatly benefit from our capability-based approach. (shrink)

The literature on the ethics of machine learning in healthcare contains a great deal of work on algorithmic fairness. But a focus on fairness has not been matched with sufficient attention to the relationship between machine learning and distributive justice in healthcare. A significant number of clinical prediction models have been developed which could be used to inform the allocation of scarce healthcare resources. As such, philosophical theories of distributive justice are relevant when considering the ethics of their design and (...) implementation. This paper considers the relationship between machine learning in healthcare and distributive justice with a focus on four aspects of algorithmic design and deployment: the choice of target variable, the model's socio-technical context, the choice of input variables, and the membership of the datasets that models are trained and validated on. Procedural recommendations for how these considerations should be accounted for in the design and implementation of such models follow. (shrink)

The application of AI in healthcare has dramatically changed the practice of medicine. In particular, AI has been implemented in a variety of roles that previously required human physicians. Due to AI's ability to outperform humans in these roles, the concern has been raised that AI will completely replace human physicians in the future. In this paper, it is argued that human physician's ability to embellish the truth is necessary to prevent injury or grief to patients, or to protect patients’ (...) feelings. Such embellishments are usually defined as “white lies,” which not only include intentionally providing false, incomplete, or misleading information to patients, but also withholding information. White lies are not uncommon in medical practice, especially in some particular cultures, such as Asia; and to some particular patients, such as children. Here, white lies are defended based on the position of weak paternalism. And although medical AI may surpass human physicians in diagnostic accuracy, efficiency, and treatment, they cannot intentionally lie to patients, which is necessary to prevent injury or grief or to protect patients’ feelings in some cases. From this perspective, it is further argued that AI cannot completely replace human physicians. (shrink)