Erin Taylor

This paper identifies a class of alignment problem that does not reduce to specification gaming, Goodhart’s Law, or construct validity failure. The rules an AI system is asked to follow are often settlement proxies. These are operational forms of political and moral questions whose answers a community has had to settle. Unlike measurement proxies, which approximate empirical targets, settlement proxies do not aim at some further thing they could be brought into closer contact with. Instead, they are the target in operational form. Settlement proxies are vulnerable to a form of legitimate challenge that more capable reasoning intensifies rather than resolves. The paper calls this the Competence Problem and identifies three structural features of settlements that generate it: 1) the compression of competing values into a single operational rule, 2) the presence of more than one defensible equilibrium among the rule’s possible forms, and 3) a constraint on the scope of information the rule engages. It then traces three forms that competent challenge takes in practice: 1) exceptions, 2) drift, and 3) challenge to the settlement itself. It argues that the appropriate response, even where the challenge is correct on the merits of the case, is typically not to grant it, because settlement authority depends on ratification through practice rather than case-level accuracy. The argument explains why behavioral training and corrigibility-based approaches to AI alignment leave the underlying capacity intact, and why the gap between case-level competence and ecosystem-level access is widest precisely where AI systems are now being deployed.

This paper argues that theoretical consistency and actionguidingness—as these have been formulated in the moral dilemmas debate—do not rule out interpersonal moral conflict. This leaves open the possibility that theoretical consistency and action-guidingness may demand more than what has been traditionally assumed. That question is considered here. Do these resources rule out all-things-considered interpersonal moral conflict in non-consequentialist theories? This paper argues that neither theoretical consistency nor action-guidingness can rule out such conflict, but action-guidingness properly construed provides a desideratum against it.

This chapter defends a novel account of the connection between social roles and their associated demands. Consider pairs of statements such as: (a) “Maura is Ethan’s mother” and (b) “Maura has an obligation to provide for Ethan.” It is natural to think that such pairs of sentences don’t merely state two unconnected truths about the agents involved. Rather, in each case, the truth of (b) seems to be in some sense explained by the truth of (a). Competing theories of the moral status of social roles, and their associated obligations, aim to explicate the nature and significance of this explanatory connection. The present chapter does this by arguing that many of the obligations or norms that attach to social roles morally bind us by default. These norms are morally binding since, first, they are ultimately explicable in terms of conventional moral duties and, secondly, those conventional duties enjoy a default but defeasible good standing.

Human-in-the-loop (HITL) approaches are commonly proposed to address alignment challenges arising from the use of large language models (LLMs) in ethics oversight. This paper argues that, paradoxically, HITL itself can amplify the risk of misalignment. Using the example of protocol triage in research ethics oversight, I demonstrate how reliance on imperfect proxies (observable stand-ins for ethical principles) creates a fundamental proxy–target gap in ethics use-cases. While human reviewers are intended to supply phenomenological and causal judgments necessary to bridge this gap, their involvement inadvertently introduces new vulnerabilities: hallucination, over-reliance, reward hacking, and sycophancy. Each vulnerability arises directly from the interaction between human feedback signals and model optimization strategies. I outline key mitigation techniques, including retrieval grounding, calibration drills, structured prompting, and adversarial debate, explaining their distinct costs and benefits. Effective HITL oversight therefore demands specialized reviewer competencies. Reviewers must recognize model vulnerabilities, understand mitigation trade-offs, and strategically deploy mitigations to preserve justificational alignment without sacrificing efficiency gains.

Deep disagreements are disagreements arising from incommensurable foundational premises. In ethics, moral values or principles constitute the foundational premises, and disagreements about them are a recognized cause of argument failure. This article proposes an additional cause of argument failure that I call paralysis. Paralysis takes place in decision-making contexts when interlocutors may agree about foundational moral values and principles, but cannot formulate arguments for decisions that are satisfactory even by their own lights. Thus, paralysis is a cause of argument failure distinct from deep disagreement. I first describe a biomedical case manifesting paralysis, where interlocutors attempted to construct arguments for the use of genetically-modified mosquitoes to address the problem of malaria in Africa. Using this scenario as an example, I identify the phenomenon of paralysis, articulate some of the causes of paralysis, and suggest the possibility of making rational progress when confronted with paralysis.

Some questions about normative structure are global. We can ask how we should live, or what we ought to do all things considered, or whether there are any categorical oughts. But we can also examine local normative structure. We might ask ourselves about what we should do from the moral point of view rather than the prudential one, or discuss promissory obligation in contrast with what friendship demands. How should we understand such localized forms of normativity? We argue that a plausible sounding treatment of the distinction cannot account for what we call the “interrelatedness” of reasons from different domains.

“Wicked” problems are characterized by intractable complexity, uncertainty, and conflict between individuals or institutions, and they inhabit almost every corner of medical ethics. Despite wide acceptance of the same ethical principles, we nevertheless disagree about how to formulate such problems, how to solve them, what would _count_ as solving them, or even what the possible solutions _are_. That is, we don’t always know how best to implement ethical ideals in messy real-world contexts. I sketch an implementation framework for medical ethics that can help clarify wicked problems and organize further ethics research toward their resolutions. This framework describes the procedural variables that work alongside substantive ethical ideals to deliver ethical decisions in complex real-world situations. Using controversial GM mosquito research as an example, I illustrate how the generalizable relationships between the variables clarify emerging ethical guidelines of research governance and provide a pathway to extend these guidelines in a way consistent with our ethical intuitions across a wide range of research and public health ethics.

Conventionalists about promising believe that it is wrong to break a promise because the promisor takes advantage of a useful social convention only to fail to do his part in maintaining it. Anti-conventionalists claim that the wrong of breaking a promise has nothing essentially to do with a social convention. Anti-conventionalists are right that the social convention is not necessary to explain the wrong of breaking most promises. But conventionalists are right that the convention plays an essential role in any satisfactory account of promising. A new conventionalism can explain this by appealing to special features of social conventions. Two of these special features have important implications for any moral requirements they mediate, such as the requirement to keep one's promises and the moral requirements attached to social or occupational roles. First, these requirements will not depend on features of a situation that are inaccessible to typical participants in the convention. Second, these requirements often cannot be tailored to fit the overly unusual circumstances of participants.