This paper develops a largely neglected parallel between prudence and population ethics. Prudence is generally understood to be concerned with the balancing of well-being over time. How, precisely, well-being ought to be balanced over time, however, is a fervently debated question. I argue that developing a standard guiding such evaluations is exceedingly challenging. This is due to an often overlooked fact about prudence, namely that it shares a structural similarity with population ethics: In both contexts, we assess the comparative value (...) of populations of personstages/people, which may vary in number and level of well-being. Based on this analogy, I show that the development of an adequate theory of prudence runs into very similar impossibility results as obtain in population ethics. In particular, I prove that Arrhenius’ fifth impossibility theorem can be applied to prudence. I develop and compare four possible answers to this challenge. First, I discuss the possibility of accepting the very repugnant conclusion intrapersonally. Second, I present and further develop Donald Bruckner’s Minimax Regret approach, which gives up Transitivity. Third, I apply Jacob Nebel’s Lexical Threshold View to prudence and critically evaluate it. Lastly, I introduce what I call the Negative Lexicality View, which is based on Lexical Threshold View but overcomes some of its problems. (shrink)

Artificial intelligence (AI) is booming, and AI ethics is booming with it. Yet there is surprisingly little attention paid to what the discipline of AI ethics is and what it ought to be. This paper offers an ameliorative definition of AI ethics to fill this gap. We introduce and defend an original distinction between novel and applied research questions. A research question should count as AI ethics if and only if (i) it is novel or (ii) it is applied and (...) has gained new importance through the development of AI. We argue that a lack of such a definition contributes to six disciplinary problems: ethics washing and lobbying, limited applicability, dilution of the field, conceptual bloating, costs of AI ethics, and an internal dispute. Based on our definition, we construct a methodological framework for AI ethics and show how it helps address these problems. (shrink)

We are constantly making choices about how to invest our time and resources. From a moral perspective, we must ask which moral concerns are most deserving of our attention. Longtermism, as e.g. defined by Greaves and MacAskill, holds that our moral focus should be on the long-term future, and that current and medium-term moral problems are comparatively insignificant. This theory is centrally based on the assumption that the moral importance of individuals can be aggregated. Since the number of individuals of (...) future generations far exceeds the number of current individuals and those closer in time, future generations are to bemorally prioritised, according to longtermism. This paper explores the implications of rejecting the premise of moral aggregation of individuals and adopting a strongly non-aggregationist position instead. It is argued that, according to strong non-aggregationism, the magnitude of the probability with which our intervention actually make a difference, as well as whether we look at the available interventions from an ex ante or ex post perspective, are relevant factors in their moral assessment. Ultimately, the conclusion is reached that strong non-aggregationism does likely not support strongly longtermist conclusions. (shrink)