Anantharaman Muralidharan

Most ethical guidelines on AI tout algorithmic transparency, the openness of an algorithm’s inner workings to human scrutiny, as an important desideratum in algorithmic deployment. Algorithmic transparency has been touted as important for valuable goals like procedural fairness, AI trustworthiness, contestability and planning around AI decision-making. This paper argues that these goals are better served by a distinct desideratum, algorithmic justifiability, the ability of an algorithm to provide understanding about why the algorithm’s decision is correct.

In a number of policy, institutional, activist and advocacy contexts, attributing bias to an algorithm does not just describe the algorithm but also imposes a particular, normatively laden conception of bias on others. Given the normative content of such bias attributions, this would involve making moral demands on others to rectify the algorithm, compensate the victims of such bias and/or not unselectively deploy the algorithm. It is also the case that moral demands, especially in the above‐mentioned contexts, are subject to a public justification requirement. As it turns out, the dominant accounts of bias in the literature presuppose some version of egalitarianism about justice and that any action that causally contributes to an unjust situation is itself wrong. Since these presuppositions are subject to reasonable disagreement, bias attributions in such situations are wrong because they violate the public justification requirement. In response, we develop a publicly justifiable conception of algorithmic bias.

Researchers and Institutional Review Board members often disagree about the permissibility of some parts or all of the former’s proposed research. At least sometimes, this disagreement is reasonable. This is often because there are competing values and no clear solution as to how to balance these values. Arguably, given that there are multiple reasonable ways in which these values might be balanced, IRBs are subject to an asymmetric public justification requirement. This requirement implies that IRBs should defer to researchers when the disagreement is reasonable. After briefly defending this claim, this paper examines its implications with respect to three questions that IRBs face. We argue that IRBs should not impose a strict limit on payments to participants, that they should take a more expansive view of what types of research are authorized by broad consent forms, and there should be no hard cap on risks participants might choose to undergo.

In vitro gametogenesis (IVG) is a technology that allows the creation of gametes from stem cells. Given that IVG makes possible the production of cross-sex gametes, IVG, if successful, can, among other uses, allow a single person to solo reproduce (i.e. have children without a sperm or egg donor). This would involve using IVG to produce one cross-sex gamete and fusing that gamete with a same-sex gamete which is either artificially or naturally derived from the same person. Because solo reproduction is a highly experimental, artificial and asexual method of reproduction, it might be comparable in many ways with reproductive cloning. One of the arguments that has been made against cloning and, by extension, solo reproduction is that it expresses and encourages vices like hubris and narcissism. In this paper, we argue that the vice argument is insufficient to justify a ban on solo reproduction. This is because the mere fact that an act is vicious does not give anyone any kind of enforceable claim on us to not act viciously.

On extant accounts of trust, it is not apt to trust someone who seems or actually is untrustworthy. It also does not seem apt, on such accounts, to morally demand trust and blame people for not trusting. Yet, there seem to be cases where it is plausible that trusting the untrustworthy might be apt or where we might aptly demand trust and blame people for not trusting. In this paper, we sketch out an account of trust according to which to trust someone involves exercising a moral power which gives trustees permission or additional reasons to do as trusted. On this account, there can be moral reasons to trust other than the trustee’s reliability. This allows our account to better account for these cases than existing accounts of trust.

Recent calls to include religious bioethics on the table in policy and other public‐facing contexts have been made on the grounds of respect. This paper argues that these same considerations of respect point to an obligation to exclude religious bioethics from public‐facing contexts. This is because public‐facing religious bioethics is typically bio‐restrictionist in orientation and thus involves making demands on others that people could reasonably disagree with. At the same time, respect for persons grounds a public justification requirement according to which it is wrong to make moral demands on others that are subject to reasonable disagreement. Proponents of inclusion of these views are thereby committed to excluding such religious bio‐restrictionist views from public‐facing contexts.

Permissivism is the claim that more than one doxastic attitude towards a proposition can at least sometimes be justified by the same total body of evidence. Impermissivism, the negation of Permissivism, is commonly taken to involve evidentialist assumptions. The only versions of Permissivism consistent with evidentialist assumptions are those which combine coarse-grained accounts of doxastic attitudes with fine-grained accounts of evidential support, as with Blake Roeber’s equipollent cases or vice-versa with fine-grained accounts of doxastic attitudes combined with coarse-grained accounts of evidential support as with Thomas Kelly’s Intuitive Credal Permissivism (ICP). This paper argues that such mismatches in granularity are not justified. I address various motivations for mismatched granularities and show in each case that the argument for the mismatch does not succeed. If these arguments succeed, Permissive views consistent with certain evidentialist assumptions are not tenable.

A significant and important ethical tension in resource allocation and public health ethics is between utility and equity. We explore this tension between utility and equity in the context of health AI through an examination of a diagnostic AI screening tool for diabetic retinopathy developed by a team of researchers at Duke-NUS in Singapore. While this tool was found to be effective, it was not equally effective across every ethnic group in Singapore, being less effective for the minority Malay population than for the Chinese majority. We discuss the problematic normative nature of bias in health AI and explore the ways in which bias can interact with various forms of social inequalities. From there, we examine the specifics of the diabetic retinopathy case and weigh up specific trade-offs between utility and equity. Ultimately, we conclude that it is ethically permissible to prioritise utility over equity where certain criteria hold. Given that any medical AI is more likely than not to have lingering bias due to bias in the training data that may reflect other social inequalities, we argue that it is permissible to implement an AI tool with residual bias where: (1) its introduction reduces the influence of biases (even if overall inequality is worsened), and/or (2) where the utility gained is significant enough and shared across groups (even if unevenly).

This paper argues that one problem that besets black-box AI is that it lacks algorithmic justifiability. We argue that the norm of shared decision making in medical care presupposes that treatment decisions ought to be justifiable to the patient. Medical decisions are justifiable to the patient only if they are compatible with the patient’s values and preferences and the patient is able to see that this is so. Patient-directed justifiability is threatened by black-box AIs because the lack of rationale provided for the decision makes it difficult for patients to ascertain whether there is adequate fit between the decision and the patient’s values. This paper argues that achieving algorithmic transparency does not help patients bridge the gap between their medical decisions and values. We introduce a hypothetical model we call Justifiable AI to illustrate this argument. Justifiable AI aims at modelling normative and evaluative considerations in an explicit way so as to provide a stepping stone for patient and physician to jointly decide on a course of treatment. If our argument succeeds, we should prefer these justifiable models over alternatives if the former are available and aim to develop said models if not.

Ageing populations are increasing across the world. Many countries are exploring new ways to provide care for the elderly in hospitals, community care and family household settings. Scientific progress in robotics, artificial intelligence integrated systems and increasingly sophisticated software engineering have contributed to innovative developments in care robots in the Asia Pacific regions, Europe and the US. Whilst the use of care robots is not widespread, research is already occurring to integrate wider use of care robots in the elderly population towards improving their quality of life. Developments in care robots for the elderly are valuable yet they give rise to special concerns as they affect peoples’ health and safety. This chapter identifies key ethical and legal issues arising from the use of care robots in the ageing population. Ethical aspects such as deception and trust from care robots are highlighted, which implicate governance concerns ranging from balancing innovations and ensuring user safety to determining the appropriate levels of regulatory oversight and support for developers and market strategies. An examination of these issues provides some guidance for law makers and policy developers in these areas of concern in establishing applicable regulatory direction, responses and governance framework.

On the standard version of political liberalism, the exercise of political power is legitimate only if it is justifiable to all reasonable persons. Correspondingly, reasonable disagreement about the moral doctrines underlying a law makes that law not justifiable to all reasonable persons. In this paper, I argue that political liberals are committed to understanding reasonable disagreement as being rational, rather than praiseworthy disagreement between morally reasonable person because other conceptions of reasonableness commit them to an incoherent triad of claims.

Ethics committees like Institutional Review Boards and Research Ethics Committees are typically empowered to approve or reject proposed studies, typically conditional on certain conditions or revisions being met. While some have argued this power should be primarily a function of applying clear, codified requirements, most institutions and legal regimes allow discretion for IRBs to ethically evaluate studies, such as to ensure a favourable risk-benefit ratio, fair subject selection, adequate informed consent, and so forth. As a result, ethics committees typically make moral demands on researchers: require them to act in a way the committee considers ethically right or appropriate. This paper argues that moral demands are legitimate only if publicly justifiable; and as a result, committee decisions are subject to a public justification requirement. Ethics committees can permissibly request for more information, changes to the research protocol or that the research is delayed or even stopped only if these demands are publicly justifiable. This latter claim is in turn justified on the basis that moral demands to φ are permissible only if we are in a position to know that the addressee ought to φ and that we are in a position to know a proposition only if it is publicly justifiable. This argument suggests that ethics committees must consciously and explicitly appeal to public reasons in their decision-making. In cases where public reasons cannot be offered, committees would not be permitted to reject a given study or make approval conditional on an amendment.

The development of some COVID-19 vaccines by private companies like Moderna and Sanofi-GSK has been substantially funded by various governments. While the Sanofi CEO has previously suggested that countries that fund this development ought to be given some priority, this suggestion has not been taken seriously in the literature. Considerations of nationalism, sustainability, need, and equitability have been more extensively discussed with respect to whether and how much a country is entitled to advance purchase orders of the vaccine under conditions of absolute scarcity. Yet, little attention has been paid to whether prior investment into developing a vaccine entitles a country to some priority with respect to these orders. Moreover, while not a majority view, some survey results show that a significant minority of the populace does endorse some view like this. This article argues that the minority have a point: recognizing funder countries some priority is justified by the weak Lockean claim (WLC). According to the WLC, the fact that someone has contributed to the development of something gives them some entitlement to the resultant product. This article will defend the WLC, and address objections to the argument, including those pertaining to questions of historical injustice and medical need. This argument does not imply an unconstrained entitlement. Rather, contribution to development is one morally relevant factor that must be tempered by and weighed against potentially more substantial claims to priority based on need, equity, and other considerations.

Recently, self-fulfilling cases, that is, ones in which an agent's believing a proposition guarantees its truth, have been offered as counterexamples to uniqueness. According to uniqueness, at most one doxastic attitude is epistemically rational given the evidence. I argue that self-fulfilling cases are not counterexamples to uniqueness because belief-formation is not governed by epistemic rationality in such cases. Specifically, this is because epistemic rationality is not just about forming true beliefs, but about tracking mind-independent truths. In support of the latter claim, I offer three arguments, namely that self-fulfilling and non-self-fulfilling cases differ in their phenomenology, in the norms that guide belief formation, and in the way they relate to the evidence.

The supposed fact of reasonable disagreement plays a crucial role in standard accounts of political liberalism. The standard account posits reasonable disagreement as arising primarily from the supposed fact that, in at least some circumstances, people can permissibly respond in different ways to the same evidence. That is to say, the standard account presupposes permissivism: It is possible that more than one doxastic attitude towards a proposition is rationally permissible, given a body of evidence. In this thesis, I shall show that the standard account is mistaken in presupposing permissivism by arguing for Uniqueness: Given a total body of evidence at most one doxastic attitude is rationally permissible. I shall argue for this by first rehearsing White’s two objections to permissivism and showing that, at the very least, they show that an intrapersonal version of Uniqueness is true. That is to say, at most one doxastic attitude is justified for a given agent given a total body of evidence. I shall then present different accounts of permissivism consistent with intrapersonal uniqueness and show that none of these accounts are successful. If my argument is successful, the standard account of political liberalism may have to be revised.

Demands for improved access to innovative therapies have prompted a discourse that claims patients have rights to access treatments that may be of benefit, even if evidence that demonstrates safety and efficacy is lacking. This rights-based discourse is grounded in accounts of autonomy and assertions claiming that the state ought to not interfere with the free choices of patients and clinical decision-making. In this essay, we scrutinise these arguments to defend the ethical and legal permissibility of interference in contexts where the uncertainty of benefit and potential for harm creates vulnerabilities that undermine patient capacity for self-determination. In support of this argument, we draw on two theoretical approaches to explore the limits of autonomy in innovative contexts and analyse the legal bases of the rights-based discourse. We then apply this analysis to the case example of stem cell transplantation as an innovative treatment for multiple sclerosis.

The Uniqueness Thesis (U), according to Richard Feldman and Roger White, says that for a given set of evidence E and a proposition P, only one doxastic attitude about P is rational given E. Luis Rosa has recently provided two counterexamples against U which are supposed to show that even if there is a sense in which choosing between two doxastic attitudes is arbitrary, both options are equally and maximally rational. Both counterexamples work by exploiting the idea that ‘ought implies can’ and trying to spell out situations in which some inferences are beyond the capabilities of some reasoners. I argue that on a descriptive account of doxastic rationality, questions of whether ‘epistemic ought implies can’ can be bracketed and that at least one of the inferential moves that Rosa describes in his cases is irrational. I further argue that a descriptive account of doxastic rationality is the appropriate notion of rationality that is to be considered when evaluating U. If my argument for a descriptive account of rationality is successful, then we have reason to revise our use of the term rationality to fit this descriptive understanding.