Laura Fearnley

The use of synthetic data for training and testing Machine Learning in AI systems is common to boost the size of training sets, compensate for sparse examples or introduce edge cases impossible to gather otherwise. This can be a particular issue for healthcare, for example where certain groups or presentations of conditions may be poorly represented. However, missing information may in itself be an indicator of the patient’s health. In this paper we examine the impact of refraining from using data imputation on the performance of a Type 2 Diabetes (T2D) co-morbidity predictor. The training data for our predictor has many missing values in patient records, particularly those from more deprived backgrounds. Our first concern is that by using data imputation to compensate for the missing values we are biasing the performance and disadvantaging patients with higher deprivation. Second, missing data may indicate that the patient was too unwell to attend a clinic, therefore, it in itself is a health predictor. Common practice in the ML community is to allow 10% of missing data to be compensated for using data synthesis, but can this be justified and what happens if we instead use incomplete data to be more representative? We performed a series of training runs with increasing amounts of targeted empty data values to assess the impact on predictor performance. We found that although there was some performance drop with 10% of missing training values this increased much more at greater percentages. We discuss the implications of our experiments as part of a safety and ethical justification for the predictor deployment and the choice of model, noting the complex trade-offs this may require.

This paper argues that the concept “safety” in AI has undergone concept creep, a phenomenon which describes the gradual semantic expansion of harm-related concepts. Originally observed in psychology, concept creep involves concepts broadening their meaning both vertically, to include less severe phenomena, and horizontally, to encompass qualitatively new phenomena. We argue that safety, particularly when applied to AI, has crept along both axes. Our analysis traces this creep by contrasting a baseline definition of safety, which is grounded in the discipline of safety science, with contemporary discourse on the safety of AI systems. We demonstrate that safety has crept horizontally to cover new phenomena, such as systemic injustices and existential risks, and it has crept vertically to include less severe phenomena, such as those related to mental wellbeing. The primary aim of this paper is to map the conceptual expansion of safety. We stop short of arguing whether this expansion constitutes progress or regress for the design and development of AI systems. However, we argue that the process of concept creep produces both beneficial and costly effects for society, policy, industry, and academic research communities. We suggest that some of the promising developments and the problematic trends recently witnessed within AI safety discourse can be understood, at least in part, as a consequence of concept creep.

Intuitively, agents can be causally responsible for outcomes through failures to act—forgetting to salt the soup makes me causally responsible for the soup’s bland taste. At the same time, it seems that not all our omissions make us causally responsible for outcomes—my failure to wear pink whilst making the soup doesn’t make me responsible for anything. It has proven difficult to find a systematic way to distinguish between those seemingly inert omissions and those that connect us to outcomes. A central aim of this paper is to identify the feature in virtue of which some omissions make us causally responsible and others do not.

This policy report arises from the research project Augmented Reality: Ethics, Perception, Metaphysics, conducted at the University of Glasgow’s Centre for the Study of Perceptual Experience between November 2021 and November 2023. It was funded by a grant from the Royal Society of Edinburgh. The project brought together experts in various academic fields, with partners from industry and regulatory bodies, to explore the nature of augmented and mixed reality technology, the theories underpinning them, and the ethical and legal questions prompted by new technology in this domain.

There has been an increasing interest into how to build Artificial Moral Agents (AMAs) that make moral decisions on the basis of causation rather than mere correction. One promising avenue for achieving this is to use a causal modelling approach. This paper explores an open and important problem with such an approach; namely, the problem of what makes a causal model an appropriate model. I explore why we need to establish criteria for what makes a model appropriate, and offer-up such criteria which appeals to normative considerations.

This paper explores the question of what makes an action morally worthy. I start with a popular theory of moral worth which roughly states that a right action is morally praiseworthy if and only if it is performed in response to the reasons which make the action right. While I think the account provides promising foundations for determining praiseworthiness, I argue that the view lacks the resources to adequately satisfy important desiderata associated with theories of moral worth. Firstly, the view does not adequately capture the degree to which an action has moral worth, and secondly, the view does not identify if right actions produced from overdetermined motives have moral worth. However, all is not lost; I also argue that the account can satisfy the desiderata when it attends to the agent’s counterfactual motives in addition to their actual motives. By considering counterfactual motives, we can measure the robustness of the actual praiseworthy motive, and attending to motivational robustness allows the new proposal to fully satisfy the two desiderata. At the end of this paper, I respond to some criticisms typically brought against a counterfactual view of moral worth.

This thesis is about morality, causation and the connection between the two. Whether there’s some causal relation between flicking the switch and turning on the light, between donating blood and saving a life, or between rain falling and puddles on the ground, is typically understood to be a mind-independent, objective, precise matter of fact. It’s no surprise given this perspective that for a long time philosophers didn’t believe something so ostensibly nebulous as morality could be a determiner of causal relations. However, recent contributions to the literature have begun to pushback against this platitude by using moral considerations, alongside a whole host of other normative notions, to determine what causal facts there are in the world. One aim of this thesis is to contribute to this recent research by arguing that an appeal to moral considerations furnishes a causal account with the resources to meet various desiderata associated with theories of causation. Insofar as this is the case, I argue that causal accounts which incorporate moral considerations are more successful than those which do not. Thus, I argue that facts about causality partly depend upon facts about morality. Causation’s effect on morality is much less controversial. Moral philosophers agree that the truth of some moral claims partly depend upon what causal relations there are in the world. This is most noticeably true in the domain of moral responsibility. The thought is that if we are to be morally responsible for some event that is not itself an aspect for our conduct, then there must be some metaphysical relation — some metaphysical glue — that links our conduct to the event in question. As I and most others see it, there is one such relation: causation. Despite this consensus, there has been surprisingly little exploration into what kinds of causal facts determine the truth of moral assessments. I suspect this lacuna has arisen because it is tempting for those working in ethics to remain neutral on questions regarding causation, lest she have to face up to dealing with the thorny issues such questions invite. However, sometimes it is only by getting involved in such issues that progress can be made. With this in mind, the second aim of the thesis is to advance this debate by clarifying and evaluating what kind of causal facts determine moral facts. In particular, I address what kind of causal facts determine whether one is morally responsible for an outcome, and whether and to what extent one is praiseworthy for that outcome. With regards to assessments of moral responsibility, I argue that whether one can be held morally responsible for the effects of an omission depends upon the causal stability of that omission. In regards to praiseworthiness, I argue that whether and to what extent one is praiseworthy for doing the right thing depends upon how causally robust one’s motivation was for acting. Thus, I argue that facts about morality are partly determined by facts about causality. In making these arguments, I defend the view that causal facts partly depend upon moral facts, and that moral facts partly depend upon causal facts. This lends support to a novel claim; namely, that morality and causation are interdependent.