Seth Lazar

LLMs are among the most advanced tools ever devised for understanding and generating natural language. Democratic deliberation and decision-making involve, at several distinct stages, the production and comprehension of language. So it is natural to ask whether our best linguistic tools might prove instrumental to one of our most important linguistic tasks involving language. Researchers and practitioners have recently asked whether LLMs can support democratic deliberation by leveraging abilities to summarise content, to aggregate opinions over summarised content, and to represent voters by predicting their preferences over unseen choices. In this paper, we assess whether using LLMs to perform these and related functions really advances the democratic values behind these experiments. We suggest that the record is mixed. In the presence of background inequality of power and resources, as well as deep moral and political disagreement, we should not use LLMs to automate non-instrumentally valuable components of the democratic process, nor should we be tempted to supplant fair and transparent decision-making procedures that are practically necessary to reconcile competing interests and values. However, while LLMs should be kept well clear of formal democratic decision-making processes, we think they can instead strengthen the informal public sphere—the arena that mediates between democratic governments and the polities that they serve, in which political communities seek information, form civic publics, and hold their leaders to account.

If we had perfect information, then we could say, for any given objectively permissible act, what makes it objectively permissible. But when we have imperfect information, when we must decide under risk and uncertainty, what then makes an act _subjectively_ permissible or impermissible? There are two salient possibilities. The first is the “verdicts” approach. It grounds judgments of subjective permissibility in probabilistically discounted judgments of objective permissibility. The principle “minimize expected objective wrongness” takes this approach. The second is the “reasons” approach. It grounds subjective permissibility in probabilistically discounted objective reasons. “Maximize expected utility” is one example. Chapter 10 considers whether the verdicts approach or the reasons approach to grounding judgments of subjective permissibility is better suited for _deontological_ decision-making with imperfect information. Perhaps surprisingly, the reasons approach comes out on top.

Suppose that you are trying to pursue a morally worthy goal, but cannot do so without incurring some moral costs. At the outset, you believed that achieving your goal was worth no more than a given moral cost. And suppose that, time having passed, you have wrought only harm and injustice, without advancing your cause. You can now reflect on whether to continue. Your goal is within reach. What's more, you believe you can achieve it by incurring—from this point forward—no more cost than it warranted at the outset. If you now succeed, the total cost will exceed the upper bound marked at the beginning. But the additional cost from this point is below that upper bound. And the good you will achieve is undiminished. How do the moral costs you have already inflicted bear upon your decision now?

As Large Language Models (LLMs) are increasingly deployed as Artificial Moral Advisors and autonomous agents making ethical decisions, evaluating their moral competence has become critical. However, existing evaluations may inadequately assess the moral reasoning capabilities needed for real-world deployment, focusing primarily on whether models can match human judgments on carefully curated ethical scenarios. We surveyed 69 papers evaluating LLM ethical competence (2020-2025) and developed a taxonomy categorising evaluations across datasets, behaviors, and metrics. Our comprehensive analysis maps the methodological landscape of this rapidly growing field and reveals several critical limitations. Most significantly, the vast majority of studies rely on pre-packaged scenarios that highlight morally relevant features, failing to test models' ability to identify ethical considerations in noisy, realistic contexts-what we term "moral sensitivity''. Additionally, evaluations overemphasize verdict accuracy rather than assessing moral reasoning quality and steerability, with few studies testing whether models can be appropriately guided toward different ethical frameworks. Most studies rely on "ground truth'' comparisons despite philosophical arguments that reasonable moral pluralism precludes definitive moral ground truth. We argue for a significant methodological shift: moving from curated scenarios to unfiltered information streams, from verdict accuracy to reasoning quality and steerability, and from ground truth metrics to assessments of reasonableness and consistency. This reorientation is essential for developing AI systems that can navigate moral complexity in real-world deployment scenarios.

Our information and communication environment has fallen short of the ideals that networked global communication might have served. Existing recommender systems very likely contribute to this shortfall. In this paper, which draws on the normative tools of philosophy of computing, informed by empirical and technical insights from computer science, we make the moral case for an alternative approach. We argue that existing recommenders incentivise mass surveillance, concentrate power, fall prey to narrow behaviourism, and compromise user agency. Rather than just trying to avoid algorithms entirely, or to incrementally alter the current paradigm, researchers and engineers should explore an alternative approach: the use of large language models (LLMs) and LLM-based agents to source and curate content that matches users’ preferences and values, expressed in natural language. The use of Language Model Agents for recommendation poses its own challenges, including those related to candidate generation, computational efficiency, preference modelling, and prompt injection. Nonetheless, if implemented successfully, these systems could: guide us through the digital public sphere without relying on mass surveillance; shift power away from platforms towards users; optimise for what matters instead of just for behavioural proxies; and scaffold our agency instead of undermining it.

Recent years have seen a resurgence of interest, among both philosophers, legal scholars, and military experts, on the ethics of war. Due in part due to post 9/11 events, this resurgence is also due to a growing theoretical sophistication among scholars in this area. Recently there has been very influential work published on the justificaton of killing in self-defense and war, and the topic of the ethics of war is now more important than ever as a discrete field. The 28 commissioned chapters in this Handbook will present a comprehensive overview of the field as well as make significant and novel contributions, and collectively they will set the terms of the debate for the next decade. Lazar and Frowe will invite the leading scholars in the field to write on topics that are new to them, making the volume a compilation of fresh ideas rather than a rehash of earlier work. The volume will be divided into five sections: Method, History, Resort, Conduct, and Aftermath. The contributors will be a mix of junior and senior figures, and will include well known scholars like Michael Walzer, Jeff McMahan, and David Rodin.

Recommender systems are artificial intelligence technologies, deployed by online platforms, that model our individual preferences and direct our attention to content we’re likely to engage with. As the digital world has become increasingly saturated with information, we’ve become ever more reliant on these tools to efficiently allocate our attention. And our reliance on algorithmic recommendation may, in turn, reshape us as moral agents. While recommender systems could in principle enhance our moral agency by enabling us to cut through the information saturation of the internet and focus on things that matter, as they’re currently designed and implemented they’re apt to interfere with our ability to attend appropriately to morally relevant factors. In order to analyze the distinctive moral problems algorithmic recommendation poses, we develop a framework for the ethics of attention and an account of judicious attention allocation as a moral skill. We then discuss empirical evidence suggesting that attentional moral skill can be thwarted and undermined in various ways by algorithmic recommendation and related affordances of online platforms, as well as economic and technical considerations that support this concern. Finally, we consider how emerging technologies might overcome the problems we identify.

The responsibility account of permissible killing in war states that only those responsible for unjustified threats may be intentionally killed in war. In recent papers, Jeff McMahan and Bradley Strawser have defended the responsibility account against an objection that it leads either toward pacifism, according to which force is nearly always unjustified, or towards total war in which combatants need not even respect noncombatant immunity, depending on how much responsibility is required for liability to be killed. This chapter rebuts their counterarguments with the challenge that defensive forces are unable to discern to what extent individuals are responsible for the threats they pose, and hence whether they are liable to defensive force.

Optimism about our ability to enhance societal decision‐making by leaning on Machine Learning (ML) for cheap, accurate predictions has palled in recent years, as these ‘cheap’ predictions have come at significant social cost, contributing to systematic harms suffered by already disadvantaged populations. But what precisely goes wrong when ML goes wrong? We argue that, as well as more obvious concerns about the downstream effects of ML‐based decision‐making, there can be moral grounds for the criticism of these predictions themselves. We introduce and defend a theory of predictive justice, according to which differential model performance for systematically disadvantaged groups can be grounds for moral criticism of the model, independently of its downstream effects. As well as helping resolve some urgent disputes around algorithmic fairness, this theory points the way to a novel dimension of epistemic ethics, related to the recently discussed category of doxastic wrong.

Military recruitment campaigns emphasize adventure, skills and camaraderie but rarely mention the moral complexities of armed conflict. Enlisting in state armed forces poses the risk of being complicit in unjust wars and associated war crimes. For prospective recruits concerned with morality, the decision is challenging. The probability of wrongdoing alone does not settle the matter; many lawful activities increase risks of future wrongdoing. The permissibility of enlisting depends on weighing expectations of doing good versus wrong. Armed forces provide security and humanitarian aid, so members often do much good. To assess individual decisions, consider whether the institution itself is justified. “Minimal justification” means armed forces cause less wrong than having none. “Full justification” means they cause less wrong than feasible alternatives. If justified, the participation needed for functioning should be permitted; volunteers take risks knowingly. If minimally justified, either participation is permitted as above, or impermissible "dirty hands" roles are required, and either conscription distributes burden fairly or volunteers show "moral courage". The real question is institutional justification, not individual permissibility. To argue enlisting is impermissible implies arguing for disbanding the military. If state armed forces are at least minimally justified, enlisting is likely morally permitted or even praiseworthy. Prospective recruits should consider institutional justification and expectations of contributing to good and wrong before deciding the morality of enlisting. For morally conscientious individuals, complex realities of armed conflict pose difficult questions with no easy answers.

This paper examines three approaches to the relationship between our moral reasons to bear costs for others’ sake before and beyond the call of duty. Symmetry holds that you are required to optimise your beneficial sacrifices even when they are genuinely supererogatory. If you are required to bear a cost C for the sake of a benefit B, when they are the only costs and benefits at stake, you are also conditionally required to bear an additional cost C, for the sake of an additional benefit B, when enough other costs and benefits are at stake that both of your alternatives are presumptively supererogatory. Disconnection rejects the requirement to optimise when your options are presumptively supererogatory and maintains that you have an entirely free hand to choose as you will among them. Asymmetry holds that when acting beyond the call of duty you are entitled to a measure of additional freedom compared to when you are not taking on supererogatory costs—you can prioritise your own well-being and reasons to a greater degree—but places constraints on the options that you may permissibly choose. We defend a version of Asymmetry and explore its implications for recent debates on charitable giving.

Increasingly secret, complex, and inscrutable computational systems are being used to intensify existing power relations and to create new ones; in particular, they are being used to govern. To be all-things-considered morally permissible new, or newly intense, power relations must meet standards of procedural legitimacy and proper authority. This is necessary for them to protect and realise democratic values of individual liberty, relational equality, and collective self-determination. For governing power in particular to be legitimate and have proper authority, it must meet a publicity requirement: reasonably competent members of the governed community must be able to determine that they are being governed legitimately and with proper authority. The publicity requirement can be satisfied only if the powerful can explain their decision-making to members of their political community. At least some duties of explanation are therefore democratic duties. This chapter first sets out the foregoing argument, then applies it to opaque computational systems, and clarifies precisely what kinds of explanations are necessary to fulfil these democratic values.

Normative ethical theories owe us an account of how to evaluate decisions under risk and uncertainty. Deontologists seem at a disadvantage here: our best decision theories seem tailor-made for consequentialism. For example, decision theory enjoins us to always perform our best option; deontology is more permissive. In this paper, we discuss and defend the idea that, when some pro-tanto wrongful act is all-things considered permissible, because it is a ‘lesser evil’, it is often merely permissible, by the lights of deontology. We show that this raises new problems for deontological decision theory, and we show that to resolve them, we need to take a more innovative approach to morally evaluating decision-making under risk and uncertainty.

Automated Influence is the use of Artificial Intelligence to collect, integrate, and analyse people’s data in order to deliver targeted interventions that shape their behaviour. We consider three central objections against Automated Influence, focusing on privacy, exploitation, and manipulation, showing in each case how a structural version of that objection has more purchase than its interactional counterpart. By rejecting the interactional focus of “AI Ethics” in favour of a more structural, political philosophy of AI, we show that the real problem with Automated Influence is the crisis of legitimacy that it precipitates.

We need a moral theory for decision-making with imperfect information; that is, decision-making under what decision theorists call both risk, when we can assign probabilities, and uncertainty, when we cannot. And yet contemporary philosophy has, for the most part, offered a division of labor: ethicists work out what we morally ought to do if we knew all the facts; decision theorists, working quite separately, focus on what we rationally ought to do, given our doubt. In this symposium, we start to bridge that divide—to bring ethicists and decision theorists together, to work out what we morally ought to do when we don’t know all the facts.

Wars are large-scale conflicts between organized groups of belligerents, which involve suffering, devastation, and brutality unlike almost anything else in human experience. Whatever one’s other beliefs about morality, all should agree that the horrors of war are all but unconscionable, and that warfare can be justified only if we have some compel- ling account of what is worth fighting for, which can justify contributing, as individu- als and as groups, to this calamitous endeavour. Although this question should obviously be central to both philosophical and politi- cal discussion about war, it is at the forefront of neither. In recent years, philosophical discussion of warfare has bloomed, but the debate has focused on whom we may kill, on the assumption that our aims are justified.1 Political debate, meanwhile, is more concerned with matters of prudence, international law, and public justification, than with reassessing what is worth fighting for. For wars of intervention to halt or prevent massive humanitarian crises, this gap is not so troubling. When warfare is the only means to prevent the mass killing or enslavement of the innocent, the purposes of military force are clear enough (though undoubtedly many other problems remain). The problem is more pressing, how- ever, for the justification of national defence.3 Although common-sense morality and international law view national defence as the paradigm case of justified warfare, grounding this consensus is surprisingly difficult.4 We typically believe that any state is justified in using lethal force to protect its territory against any form of uninvited military incursion by any other state. And yet we lack a good argument to explain why this should be so. In this chapter, I explain why one familiar and otherwise plausible approach to the justification of killing in war cannot adequately ground common-sense views of permissible national defence.5 Reductionists believe that justified warfare reduces to an aggregation of acts that are justified under ordinary principles of interpersonal morality.6 The standard form of reductionism focuses on the principles governing killing in ordinary life, specifically those that justify intentional killing in self- and other-defence, and unintended but foreseen (for short, collateral) killing as a lesser evil. Justified warfare, on this view, is no more than the coextension of multiple acts justified under these two principles. Reductionism is the default philosophical approach to thinking through the ethics of killing in war. It makes perfect sense to ask what principles govern permissible kill- ing in general, before applying them to the particular context of war. If it cannot deliver a plausible set of conclusions about when national defence is permitted, then we must either revise our beliefs about which conclusions count as plausible, or else face the significant challenge of developing a different theoretical model for justifying war- fare—an exceptionalist model, which views war as an exception to the regular moral landscape, to which principles apply which apply to nothing else but war.7 We must show, in other words, that there is something worth fighting for in wars of national defence, which is not engaged when we use force in any other context. The chapter proceeds as follows. Section 2.2 sets out the argument against reduc- tionism.8 Section 2.3 considers and rebuts one common response to the argument, which has often been thought sufficient grounds to disregard its conclusion. Section 2.4 then asks whether a modified reductionism would survive unscathed by the argu- ment. Finally, section 2.5 sets out some desiderata on a plausible exceptionalist alterna- tive. Section 2.6 concludes.

McMahan’s book develops each of these themes: rejection of the moral equality of soldiers, introduction and defense of his criterion of liability to lethal attack, and resistance to its unsavory implications for noncombatant immunity. The contributions to this symposium focus on the first two themes. John Gardner and Franc¸ois Tanguay-Renaud make a plea for culpability, testing McMahan’s endorsement of a thinner standard of responsibility for liability, while David Rodin’s paper explores the implications of McMahan’s novel analysis of proportionality in self defense, asking whether the proportionality metaphor can be extended even further than McMahan considers. Cheyney Ryan and Yitzhak Benbaji each argue that the intuitions underlying the moral equality of soldiers have firmer foundations than McMahan allows—Ryan aiming to develop the democratic justification for military obedience; Benbaji seeking to defend a more holistic contractarian justification of the conventional principles governing conduct in war. McMahan then offers a detailed response to his critics.

If we were required to sacrifice our own interests whenever doing so was best overall, or prohibited from doing so unless it was optimal, then we would be mere sites for the realisation of value. Our interests, not ourselves, would wholly determine what we ought to do. We are not mere sites for the realisation of value — instead we, ourselves, matter unconditionally. So we have options to act suboptimally. These options have limits, grounded in the very same considerations. Though not merely such sites, you and I are also sites for the realisation of value, and our interests (and ourselves) must therefore sometimes determine what others ought to do, in particular requiring them to bear reasonable costs for our sake. Likewise, just as my moral status grounds a requirement that others show me appropriate respect, so must I do to myself.

People often think that their special relationships with family, friends, comrades and compatriots, can ground moral reasons. Among these reasons, they understand some to be duties – pro tanto requirements that have genuine weight when they conflict with other considerations. In this paper I ask: what is the underlying moral structure of associative duties? I first consider and reject the orthodox Teleological Welfarist account, which first observes that special relationships are fundamental for human well-being, then claims that we cannot have these relationships, if we do not recognise associative duties, before concluding that we should therefore recognise associative duties. I then introduce a nonteleological alternative, grounded in the Appropriate Response approach to ethical theory.

I try to show that agent responsibility is an inadequate basis for the attribution of liability, by discrediting the Risk Argument and showing how the Responsibility Argument in fact collapses into the Risk Argument. I have concentrated on undermining these as philosophical theories of self-defense, although I at times note that our theory of self-defense should not be predicated on assumptions that are inapplicable to the context of war. The potential combatant, I conclude, should not look to the agency view to ease his conscience about the killing he will do in war. If he is to fight without violating rights, then he will need a different theory of self-defense; if there are no more defensible theories, then he must either accept pacifism—which I think is a live possibility—or seek to identify stronger countervailing reasons, which can override the great force of his victims’ rights to life.

Killing civilians is worse than killing soldiers. If any moral principle commands near universal assent, this one does. Few moral principles have been more widely and more viscerally affirmed. And yet, in recent years it has faced a rising tide of dissent. Political and military leaders seeking to slip the constraints of the laws of war have cavilled and qualified. Their complaints have been unwittingly aided by philosophers who, rebuilding just war theory from its foundations, have concluded that this principle is at best a useful fiction. Sparing Civilians aims to turn this tide, and to vindicate international law, and the ruptured consensus. In doing so, Seth Lazar develops new insights into the morality of harm, relevant to everyone interested in normative and applied ethics.