Lauritz Munch

This paper argues that the idea that privacy concerns information admits two importantly different interpretations. On the aboutness view, a person's privacy is a function of others' access to information about that person. On the informational view, privacy is a function of access to states of affairs that have the capacity to be informative with regard to a person. We argue that these views are distinct and that the informational view is preferable. Our central argument turns on cases of informative absence: cases where the absence of a person from a database or record carries information about them despite containing nothing about them in the sense required by the aboutness view. We show that the informational view can explain when and why such absences matter for privacy in a principled way, while the aboutness view is systematically blind to this entire class of cases. Undergirding our argument is the adoption of Kelp and Simion's (2025) knowledge-based account of information, which we argue is superior to probabilistic alternatives both independently and for the specific purposes of privacy theory.

We argue that there are cases in which it is morally permissible to replace medical practitioners with machine learning algorithms. Our argument appeals to the uncontroversial view that it is sometimes morally permissible to make medical decisions based on cost-effectiveness considerations. For example, it is generally morally permissible to prescribe a treatment that is as effective as its alternatives but much cheaper. If this is so, we argue, then similar cost-effectiveness considerations can explain why it is sometimes morally permissible to replace human practitioners with algorithms. To reject our argument, one needs to show that when it comes to algorithmic medical decision-making, there are special considerations that would always override cost-effectiveness considerations. We go through a range of candidate considerations and argue that none of them is weighty enough to tip the balance in favor of human practitioners.

Many believe that relationships can make a constitutive difference to the moral status of paternalistic treatment. For example, it is often assumed that it’s easier to justify paternalizing a spouse than a stranger. But although this thought is widespread, there exists no detailed account of how relationships could mitigate paternalistic complaints. The aim of this paper is to develop an account of this phenomenon, drawing on the work of Margaret Gilbert and the notion of joint commitments. According to the resulting view, close relations can constitutively mitigate paternalistic complaints by rendering paternalistic interference consistent with the will of the paternalized agent.

Many believe that relationships can make a constitutive difference to the moral status of paternalistic treatment. For example, it is often assumed that it's easier to justify paternalizing a spouse than a stranger. But although this thought is widespread, there exists no detailed account of how relationships could mitigate paternalistic complaints. The aim of this paper is to develop an account of this phenomenon, drawing on the work of Margaret Gilbert and the notion of joint commitments. According to the resulting view, close relations can constitutively mitigate paternalistic complaints by rendering paternalistic interference consistent with the will of the paternalized agent.

Accuracy plays an important role in the deployment of machine learning algorithms. But accuracy is not the only epistemic property that matters. For instance, it is well-known that algorithms may perform accurately during their training phase but experience a significant drop in performance when deployed in real-world conditions. To address this gap, people have turned to the concept of algorithmic robustness. Roughly, robustness refers to an algorithm’s ability to maintain its performance across a range of real-world and hypothetical conditions. In this paper, we develop a rigorous account of algorithmic robustness grounded in Robert Nozick’s counterfactual sensitivity and adherence conditions for knowledge. By bridging insights from epistemology and machine learning, we offer a novel conceptualization of robustness that captures key instances of algorithmic brittleness while advancing discussions on reliable AI deployment. We also show how a sensitivity-based account of robustness provides notable advantages over related approaches to algorithmic brittleness, including causal and safety-based ones.

People working on the ethics of discrimination have struggled with accounting for a kind of moral wrongdoing that is thought to be present in all instances of wrongful discrimination. So far, any moral wrong claimed to be characteristic of wrongful discrimination in this way has failed to generalize to all cases of wrongful discrimination, moving many to abandon the idea that wrongful discrimination is normatively distinct at all. Motivated by this situation, we propose to abandon the assumption that the common moral wrong of wrongful discrimination must be a first-order moral wrong. Instead, we argue that the common wrong of discrimination is a normative mistake in a more general sense, namely the flaw of being biased. Drawing on Thomas Kelly’s (2022) recent work on bias, we show that this account provides a unified and plausible answer to the question of what makes wrongful discrimination characteristically morally wrong, while being able to accommodate the plurality of moral considerations at play in wrongful discrimination.

The concept of privacy is both significant and contested, with ongoing philosophical debate about whether it is best understood in terms of non-access or as involving some form of control. This paper advances the discussion by employing experimental philosophy to examine folk intuitions about privacy. Our findings show that these intuitions favor a control-based concept of privacy. Additionally, we show that the type of information at stake influences privacy judgments, indicating that privacy concerns are sensitive to not only access and control but also to informational content. Our results enable us to make progress on the question of how to understand the concept of privacy and calls for a new research agenda that pays closer attention to different informational distinctions.

Accuracy plays an important role in the deployment of machine learning algorithms. But accuracy is not the only epistemic property that matters. For instance, it is well-known that algorithms may perform accurately during their training phase but experience a significant drop in performance when deployed in real-world conditions. To address this gap, people have turned to the concept of algorithmic robustness. Roughly, robustness refers to an algorithm’s ability to maintain its performance across a range of real-world and hypothetical conditions. In this paper, we develop a rigorous account of algorithmic robustness grounded in Robert Nozick’s counterfactual sensitivity and adherence conditions for knowledge. By bridging insights from epistemology and machine learning, we offer a novel conceptualization of robustness that captures key instances of algorithmic brittleness while advancing discussions on reliable AI deployment. We also show how a sensitivity-based account of robustness provides notable advantages over related approaches to algorithmic brittleness, including causal and safety-based ones.

Since the early days of the Explainable AI movement, post-hoc explanations have been praised for their potential to improve user understanding, promote trust, and reduce patient safety risks in black box medical AI systems. Recently, however, critics have argued that the benefits of post-hoc explanations are greatly exaggerated since they merely approximate, rather than replicate, the actual reasoning processes that black box systems take to arrive at their outputs. In this article, we aim to defend the value of post-hoc explanations against this recent critique. We argue that even if post-hoc explanations do not replicate the exact reasoning processes of black box systems, they can still improve users’ functional understanding of black box systems, increase the accuracy of clinician-AI teams, and assist clinicians in justifying their AI-informed decisions. While post-hoc explanations are not a “silver bullet” solution to the black box problem in medical AI, we conclude that they remain a useful strategy for addressing the black box problem in medical AI.

Many believe that we ought to treat people as individuals and that this form of treatment is in some sense incompatible with treating people as members of groups. Yet, the relation between these two kinds of treatments is elusive. In this paper, we develop a novel account of the normative requirement to treat people as individuals. According to this account, treating people as individuals requires treating people as agents in the appropriate capacity. We call this the Agency Attunement Account. This view has the surprising implication that it's not only possible to fail in treating people as individuals but also possible to fail in treating people as members of groups, on the presumption that people's agency is sometimes nested in, and exercised via, agential groups.

It has often been argued that we face a trade-off between accuracy and opacity in deep learning models. The idea is that we can only harness the accuracy of deep learning models by simultaneously accepting that the grounds for the models’ decision-making are epistemically opaque to us. In this paper, we ask the following question: what are the prospects of making deep learning models transparent without compromising on their accuracy? We argue that the answer to this question depends on which kind of opacity we have in mind. If we focus on the standard notion of opacity, which tracks the internal complexities of deep learning models, we argue that existing explainable AI (XAI) techniques show us that the prospects look relatively good. But, as it has recently been argued in the literature, there is another notion of opacity that concerns factors external to the model. We argue that there are at least two types of external opacity—link opacity and structure opacity—and that existing XAI techniques can to some extent help us reduce the former but not the latter.

By now a consensus has emerged that people, when subjected to high-stakes decisions through automated decision systems, have a moral right to have these decisions explained to them. However, furnishing such explanations can be costly. So the right to an explanation creates what we call the cost problem: providing subjects of automated decisions with appropriate explanations of the grounds of these decisions can be costly for the companies and organisations that use these automated decision systems. In this paper, we explore whether large language models could prove significant in overcoming the cost problem. We provide an initial case for believing that they can but only with serious ethical costs.

Often, when we share information about ourselves, we contribute to people learning personal things about others. This may happen because what we share about ourselves can be used to infer personal information about others. Such dependencies have become known as privacy dependencies in the literature. It is sometimes claimed that the scope of the right to privacy should be expanded in light of such dependencies. For example, some have argued that inferring information about others can violate their right to privacy. Others have argued that sharing personal information about yourself that license such inferences can by itself violate the right to privacy. In this paper, we argue that the latter view should be rejected.

According to a popular view, privacy is a function of people not knowing or rationally believing some fact about you. But intuitively it seems possible for a perpetrator to violate your right to privacy without learning any facts about you. For example, it seems plausible to say that the US National Security Agency’s PRISM program violated, or could have violated, the privacy rights of the people whose information was collected, despite the fact that the NSA, for the most part, merely collected information without examining it, and thus, for the most part, did not acquire knowledge about specific people. This pair of judgements creates the challenge of aligning one of the most popular accounts of what privacy is with a suitable account of the right to privacy. Call this the alignment challenge. In this paper, I criticize two recent attempts to answer the alignment challenge: one based on risk, proposed by Björn Lundgren, and one based on modal robustness, proposed by Carissa Véliz. I then offer my preferred response, which I call the inquiry account. According to this account, the right to privacy protects against certain forms of inquiry. This account explains why there is a gap between what privacy is a what the right to privacy protects people against. Moreover, it explains why it is possible to violate someone’s right to privacy without obtaining knowledge.

Many believe that we ought to treat people as individuals and that this form of treatment is in some sense incompatible with treating people as members of groups. Yet, the relation between these two kinds of treatments is elusive. In this paper, we develop a novel account of the normative requirement to treat people as individuals. According to this account, treating people as individuals requires treating people as agents in the appropriate capacity. We call this the Agency Attunement Account. This view has the surprising implication that it's not only possible to fail in treating people as individuals but also possible to fail in treating people as members of groups, on the presumption that people's agency is sometimes nested in, and exercised via, agential groups.

Many seem to think that AI-induced responsibility gaps are morally bad and therefore ought to be avoided. We argue, by contrast, that there is at least a pro tanto reason to welcome responsibility gaps. The central reason is that it can be bad for people to be responsible for wrongdoing. This, we argue, gives us one reason to prefer automated decision-making over human decision-making, especially in contexts where the risks of wrongdoing are high. While we are not the first to suggest that responsibility gaps should sometimes be welcomed, our argument is novel. Others have argued that responsibility gaps should sometimes be welcomed because they can reduce or eliminate the psychological burdens caused by tragic moral choice-situations. By contrast, our argument explains why responsibility gaps should sometimes be welcomed even in the absence of tragic moral choice-situations, and even in the absence of psychological burdens.

If I decide to disclose information about myself, this act may undermine other people’s ability to conceal information about them. Such dependencies are called privacy dependencies in the literature. Some say that privacy dependencies generate moral duties to avoid sharing information about oneself. If true, we argue, then it is sometimes justified for others to impose harm on the person sharing information to prevent them from doing so. In this paper, we first show how such conclusions arise. Next, we show that the existence of such a dependency between the moral significance you are inclined to attribute to privacy dependencies and judgments about permissible self-defense puts pressure on at least some ways of spelling out the idea that privacy dependencies ought to constrain our data-sharing conduct.

In this paper, we defend what we call the ‘Hybrid View’ of privacy. According to this view, an individual has privacy if, and only if, no one else forms an epistemically warranted belief about the individual’s personal matters, nor perceives them. We contrast the Hybrid View with what seems to be the most common view of what it means to access someone’s personal matters, namely the Belief-Based View. We offer a range of examples that demonstrate why the Hybrid View is more plausible than the Belief-Based View. Finally, we show how the Hybrid View generates a more plausible fit between the concept of privacy, and the concept of a (morally objectionable) violation of privacy.

The stakes associated with an algorithmic decision are often said to play a role in determining whether the decision engenders a right to an explanation. More specifically, “high stakes” decisions are often said to engender such a right to explanation whereas “low stakes” or “non-high” stakes decisions do not. While the overall gist of these ideas is clear enough, the details are lacking. In this paper, we aim to provide these details through a detailed investigation of what we will call the “Simple Stakes Thesis”. The Simple Stakes Thesis, as it will turn out, is too simple. For even if the stakes associated with a specific one-off decision are low—and hence does not engender a right to an explanation—such decisions may nevertheless form part of a high stakes pattern or aggregate of decisions. In such cases, we argue, even a low stakes decision may engender a right to explanation. Not only does this show that the right to explanation is more demanding than so far recognized. It also shows that the stakes thesis is significantly harder to apply in practice.

Most studies on the ethics of privacy focus on what others ought to do to accommodate our interest in privacy. I focus on a related but distinct question that has attracted less attention in the literature: When, if ever, does morality require us to safeguard our own privacy? While we often have prudential reasons for safeguarding our privacy, we are also, at least sometimes, morally required to do so. I argue that we, sometimes, ought to safeguard our privacy for the sake of the possible wrongdoer, in order to prevent the possible wrongdoer from committing a morally defiling act that renders them worse off. I illustrate the variety of cases in which this duty binds us and contrast my view to some other claims in the literature which also purport to show that we sometimes have a duty to safeguard our own privacy.