Uwe Peters

For synergistic interactions between humans and artificial intelligence (AI) systems, AI outputs often need to be explainable to people. Explainable AI (XAI) systems are commonly tested in human user studies. However, whether XAI researchers consider potential cultural differences in human explanatory needs remains unexplored. We highlight psychological research that found significant differences in human explanations between many people from Western, commonly individualist countries and people from non-Western, often collectivist countries. We argue that XAI research currently overlooks these variations and that many popular XAI designs implicitly and problematically assume that Western explanatory needs are shared cross-culturally. Additionally, we systematically reviewed over 200 XAI user studies and found that most studies did not consider relevant cultural variations, sampled only Western populations, but drew conclusions about human-XAI interactions more generally. We also analyzed over 30 literature reviews of XAI studies. Most reviews did not mention cultural differences in explanatory needs or flag overly broad cross-cultural extrapolations of XAI user study results. Combined, our analyses provide evidence of a cultural bias toward Western populations in XAI research, highlighting an important knowledge gap regarding how culturally diverse users may respond to widely used XAI systems that future work can and should address.

It has been argued that our currently most satisfactory social epistemology of science can’t account for science that is based on artificial intelligence (AI) because this social epistemology requires trust between scientists that can take full responsibility for the research tools they use, and scientists can’t take full responsibility for the AI tools they use since these systems are epistemically opaque. I think this argument overlooks that much AI-based science can be done without opaque models, and that agents can take full responsibility for the systems they use even if these systems are opaque. Requiring that an agent fully understand how a system works is an untenably strong condition for that agent to take full responsibility for the system and risks absolving AI developers from responsibility for their products. AI-based science need not create trust-related social epistemological problems if we keep in mind that what makes both individual scientists and their use of AI systems trustworthy isn’t full transparency of the internal processing but their adherence to social and institutional norms that ensure that scientific claims can be trusted.

Drawing on empirical findings, a number of philosophers have recently argued that people who use English as a foreign language may face a linguistic bias in academia in that they or their contributions may be perceived more negatively than warranted because of their English. I take a critical look at this argument. I first distinguish different phenomena that may be conceptualized as linguistic bias but that should be kept separate to avoid overgeneralizations. I then examine a range of empirical studies that philosophers have cited to argue that people who use English as a foreign language are subject to linguistic bias in academia. I contend that many of these studies do not sufficiently support key claims that philosophers have made about linguistic bias, are challenged by counterevidence, and lack generalizability. I end by introducing methodological recommendations that may help philosophers develop more convincing empirically informed arguments regarding linguistic bias.

Scientists may sometimes generalize from their samples to broader populations when they have not yet sufficiently supported this generalization. Do such hasty generalizations also occur in experimental philosophy? To check, we analyzed 171 experimental philosophy studies published between 2017 and 2023. We found that most studies tested only Western populations but generalized beyond them without justification. There was also no evidence that studies with broader conclusions had larger, more diverse samples, but they nonetheless had higher citation impact. Our analyses reveal important methodological limitations of many experimental philosophy studies and suggest that philosophical training may not protect against hasty generalizations.

Many ethical frameworks require artificial intelligence (AI) systems to be explainable. Explainable AI (XAI) models are frequently tested for their adequacy in user studies. Since different people may have different explanatory needs, it is important that participant samples in user studies are large enough to represent the target population to enable generalizations. However, it is unclear to what extent XAI researchers reflect on and justify their sample sizes or avoid broad generalizations across people. We analyzed XAI user studies (N = 220) published between 2012 and 2022. Most studies did not offer rationales for their sample sizes. Moreover, most papers generalized their conclusions beyond their target population, and there was no evidence that broader conclusions in quantitative studies were correlated with larger samples. These methodological problems can impede evaluations of whether XAI systems implement the explainability called for in ethical frameworks. We outline principles for more inclusive XAI user studies.

The English language now dominates scientific communications. Yet, many scientists have English as their second language. Their English proficiency may therefore often be more limited than that of a ‘native speaker’, and their scientific contributions (e.g. manuscripts) in English may frequently contain linguistic features that disrupt the fluency of a reader’s, or listener’s information processing even when the contributions are understandable. Scientific gatekeepers (e.g. journal reviewers) sometimes cite these features to justify negative decisions on manuscripts. Such justifications may rest on the prima facie plausible assumption that linguistic characteristics that hinder fast and easy understandability of scientific contributions are epistemically undesirable in science. I shall raise some doubts about this assumption by drawing on empirical research on processing fluency and its biasing effects. I also argue that directing scientists with English as a foreign language toward approaching ‘native-level’ English can have the negative consequence of reducing their potential to make scientific belief formation more reliable. These points suggest that one seemingly compelling justification for linguistically discriminating against scientific contributions in ‘non-native’ English is questionable and that the common insistence by scientific gatekeepers on ‘native-like’ English may be epistemically harmful to science.

Many artificial intelligence (AI) systems currently used for decision-making are opaque, i.e., the internal factors that determine their decisions are not fully known to people due to the systems’ computational complexity. In response to this problem, several researchers have argued that human decision-making is equally opaque and since simplifying, reason-giving explanations (rather than exhaustive causal accounts) of a decision are typically viewed as sufficient in the human case, the same should hold for algorithmic decision-making. Here, I contend that this argument overlooks that human decision-making is sometimes significantly more transparent and trustworthy than algorithmic decision-making. This is because when people explain their decisions by giving reasons for them, this frequently prompts those giving the reasons to govern or regulate themselves so as to think and act in ways that confirm their reason reports. AI explanation systems lack this self-regulative feature. Overlooking it when comparing algorithmic and human decision-making can result in underestimations of the transparency of human decision-making and in the development of explainable AI that may mislead people by activating generally warranted beliefs about the regulative dimension of reason-giving.

Many scientists routinely generalize from study samples to larger populations. It is commonly assumed that this cognitive process of scientific induction is a voluntary inference in which researchers assess the generalizability of their data and then draw conclusions accordingly. We challenge this view and argue for a novel account. The account describes scientific induction as involving by default a generalization bias that operates automatically and frequently leads researchers to unintentionally generalize their findings without sufficient evidence. The result is unwarranted, overgeneralized conclusions. We support this account of scientific induction by integrating a range of disparate findings from across the cognitive sciences that have until now not been connected to research on the nature of scientific induction. The view that scientific induction involves by default a generalization bias calls for a revision of the current thinking about scientific induction and highlights an overlooked cause of the replication crisis in the sciences. Commonly proposed interventions to tackle scientific overgeneralizations that may feed into this crisis need to be supplemented with cognitive debiasing strategies against generalization bias to most effectively improve science.

Does algorithmic political bias contribute to an entrenchment and polarization of political positions? Franke argues that it may do so because the bias involves classifications of people as liberals, conservatives, etc., and individuals often conform to the ways in which they are classified. I provide a novel example of this phenomenon in human–computer interactions and introduce a social psychological mechanism that has been overlooked in this context but should be experimentally explored. Furthermore, while Franke proposes that algorithmic political classifications entrench political identities, I contend that they may often produce the opposite result. They can lead people to change in ways that disconfirm the classifications. Consequently and counterintuitively, algorithmic political bias can in fact decrease political entrenchment and polarization.

Some artificial intelligence systems can display algorithmic bias, i.e. they may produce outputs that unfairly discriminate against people based on their social identity. Much research on this topic focuses on algorithmic bias that disadvantages people based on their gender or racial identity. The related ethical problems are significant and well known. Algorithmic bias against other aspects of people’s social identity, for instance, their political orientation, remains largely unexplored. This paper argues that algorithmic bias against people’s political orientation can arise in some of the same ways in which algorithmic gender and racial biases emerge. However, it differs importantly from them because there are strong social norms against gender and racial biases. This does not hold to the same extent for political biases. Political biases can thus more powerfully influence people, which increases the chances that these biases become embedded in algorithms and makes algorithmic political biases harder to detect and eradicate than gender and racial biases even though they all can produce similar harm. Since some algorithms can now also easily identify people’s political orientations against their will, these problems are exacerbated. Algorithmic political bias thus raises substantial and distinctive risks that the AI community should be aware of and examine.

It has recently been argued that to tackle social injustice, implicit biases and unjust social structures should be targeted equally because they sustain and ontologically overlap with each other. Here I develop this thought further by relating it to the hypothesis of extended cognition. I argue that if we accept common conditions for extended cognition then people’s implicit biases are often partly realized by and so extended into unjust social structures. This supports the view that we should counteract psychological and social contributors to injustice equally. But it also has a significant downside. If unjust social structures are part of people’s minds then dismantling these structures becomes more difficult than it currently is, as this will then require us to overcome widely accepted ethical and legal barriers protecting people’s bodily and personal integrity. Thus, while there are good grounds to believe that people’s biases and unjust social structures ontologically overlap, there are also strong ethical reasons to reject this view. Metaphysical and ethical intuitions about implicit bias hence collide in an important way.

It is well known that on the Internet, computer algorithms track our website browsing, clicks, and search history to infer our preferences, interests, and goals. The nature of this algorithmic tracking remains unclear, however. Does it involve what many cognitive scientists and philosophers call ‘mindreading’, i.e., an epistemic capacity to attribute mental states to people to predict, explain, or influence their actions? Here I argue that it does. This is because humans are in a particular way embedded in the process of algorithmic tracking. Specifically, if we endorse common conditions for extended cognition, then human mindreading (by website operators and users) is often literally extended into, that is, partly realized by, not merely causally coupled to, computer systems performing algorithmic tracking. The view that human mindreading extends outside the body into computers in this way has significant ethical advantages. It points to new conceptual ways to reclaim our autonomy and privacy in the face of increasing risks of computational control and online manipulation. These benefits speak in favor of endorsing the notion of extended mindreading.

‘No-platforming’—the practice of denying someone the opportunity to express their opinion at certain venues because of the perceived abhorrent or misguided nature of their view—is a hot topic. Several philosophers have advanced epistemic reasons for using the policy in certain cases. Here we introduce epistemic considerations against no-platforming that are relevant for the reflection on the cases at issue. We then contend that three recent epistemic arguments in favor of no-platforming fail to factor these considerations in and, as a result, offer neither a conclusive justification nor strong epistemic support for no-platforming in any of the relevant cases. Moreover, we argue that, taken together, our epistemic considerations against no-platforming and the three arguments for the policy suggest that no-platforming poses an epistemic dilemma. While advocates and opponents of no-platforming alike have so far overlooked this dilemma, it should be addressed not only to prevent that actual no-platforming decisions create more epistemic harm than good, but also to put us into a better position to justify the policy when it is indeed warranted.

Demographic diversity might often be present in a group without group members noticing it. What are the epistemic effects if they do? Several philosophers and social scientists have recently argued that when individuals detect demographic diversity in their group, this can result in epistemic benefits even if that diversity doesn’t involve cognitive differences. Here I critically discuss research advocating this proposal, introduce a distinction between two types of detection of demographic diversity, and apply this distinction to the theorizing on diversity in science. Focusing on ‘invisible’ diversity, I argue that in one common kind of group in science, if group members have full insight into their group’s diversity, this is likely to create epistemic costs. These costs can be avoided and epistemic benefits gained if group members only partly detect their group’s diversity. There is thus an epistemic reason for context-dependent limitations on scientists’ insight into the diversity of their group.

When scientists or science reporters communicate research results to the public, this often involves ethical and epistemic risks. One such risk arises when scientific claims cause cognitive or behavioural changes in the audience that contribute to the self-fulfilment of these claims. I argue that the ethical and epistemic problems that such self-fulfilment effects may pose are much broader and more common than hitherto appreciated. Moreover, these problems are often due to a specific psychological phenomenon that has been neglected in the research on science communication: many people tend to conform to ‘descriptive norms’, that is, norms capturing (perceptions of) what others commonly do, think, or feel. Because of this tendency, science communication may frequently produce significant social harm. I contend that scientists have a responsibility to assess the risk of this potential harm and consider adopting strategies to mitigate it. I introduce one such strategy and argue that its implementation is independently well motivated by the fact that it helps improve scientific accuracy.

Can young children such as 3-year-olds represent the world objectively? Some prominent developmental psychologists—such as Perner and Tomasello—assume so. I argue that this view is susceptible to a prima facie powerful objection: To represent objectively, one must be able to represent not only features of the entities represented but also features of objectification itself, which 3-year-olds cannot do yet. Drawing on Burge's work on perceptual constancy, I provide a response to this objection and motivate a distinction between three different kinds of objectivity. This distinction helps advance current research on both objectivity and teleological action explanations in young children.

Recently, researchers and reporters have made a wide range of claims about the distribution, nature, and societal impact of political polarization. Here I offer reasons to believe that, even when they are correct and prima facie merely descriptive, many of these claims have the highly negative side effect of increasing political polarization. This is because of the interplay of two factors that have so far been neglected in the work on political polarization, namely that (1) people have a tendency to conform to descriptive norms (i.e., norms capturing (perceptions of) what others commonly do, think, or feel), and (2) claims about political polarization often convey such norms. Many of these claims thus incline people to behave, cognize, and be affectively disposed in ways that contribute to social division. But there is a silver lining. People’s tendency to conform to descriptive norms also provides the basis for developing new, experimentally testable strategies for counteracting political polarization. I outline three.

It has recently been suggested that politically motivated cognition leads progressive individuals to form beliefs that underestimate real differences between social groups and to process information selectively to support these beliefs and an egalitarian outlook. I contend that this tendency, which I shall call ‘egalitarian confirmation bias’, is often ‘Mandevillian’ in nature. That is, while it is epistemically problematic in one’s own cognition, it often has effects that significantly improve other people’s truth tracking, especially that of stigmatized individuals in academia. Due to its Mandevillian character, egalitarian confirmation bias isn’t only epistemically but also ethically beneficial, as it helps decrease social injustice. Moreover, since egalitarian confirmation bias has Mandevillian effects especially in academia, and since progressives are particularly likely to display the bias, there is an epistemic reason for maintaining the often-noted political majority of progressives in academia. That is, while many researchers hold that diversity in academia is epistemically beneficial because it helps reduce bias, I argue that precisely because political diversity would help reduce egalitarian confirmation bias, it would in fact in one important sense be epistemically costly.

Confirmation bias is one of the most widely discussed epistemically problematic cognitions, challenging reliable belief formation and the correction of inaccurate views. Given its problematic nature, it remains unclear why the bias evolved and is still with us today. To offer an explanation, several philosophers and scientists have argued that the bias is in fact adaptive. I critically discuss three recent proposals of this kind before developing a novel alternative, what I call the ‘reality-matching account’. According to the account, confirmation bias evolved because it helps us influence people and social structures so that they come to match our beliefs about them. This can result in significant developmental and epistemic benefits for us and other people, ensuring that over time we don’t become epistemically disconnected from social reality but can navigate it more easily. While that might not be the only evolved function of confirmation bias, it is an important one that has so far been neglected in the theorizing on the bias.

Social and medical scientists frequently produce empirical generalizations that involve concepts partly defined by value judgments. These generalizations, which have been called ‘mixed claims’, raise interesting questions. Does the presence of them in science imply that science is value-laden? Is the value-ladenness of mixed claims special compared to other kinds of value-ladenness of science? Do we lose epistemically if we reformulate these claims as conditional statements? And if we want to allow mixed claims in science, do we need a new account of how to reconcile values with objectivity? Alexandrova (2017, 2018) offers affirmative answers to these questions. In responding to Alexandrova’s arguments, this discussion note motivates negative ones and in doing so casts new light on mixed claims.

Members of the field of philosophy have, just as other people, political convictions or, as psychologists call them, ideologies. How are different ideologies distributed and perceived in the field? Using the familiar distinction between the political left and right, we surveyed an international sample of 794 subjects in philosophy. We found that survey participants clearly leaned left (75%), while right-leaning individuals (14%) and moderates (11%) were underrepresented. Moreover, and strikingly, across the political spectrum, from very left-leaning individuals and moderates to very right-leaning individuals, participants reported experiencing ideological hostility in the field, occasionally even from those from their own side of the political spectrum. Finally, while about half of the subjects believed that discrimination against left- or right-leaning individuals in the field is not justified, a significant minority displayed an explicit willingness to discriminate against colleagues with the opposite ideology. Our findings are both surprising and important, because a commitment to tolerance and equality is widespread in philosophy, and there is reason to think that ideological similarity, hostility, and discrimination undermine reliable belief formation in many areas of the discipline.

In empirically informed research on action explanation, philosophers and developmental psychologists have recently proposed a teleological account of the way in which we make sense of people’s intentional behavior. It holds that we typically don’t explain an agent’s action by appealing to her mental states but by referring to the objective, publically accessible facts of the world that count in favor of performing the action so as to achieve a certain goal. Advocates of the teleological account claim that this strategy is our main way of understanding people’s actions. I argue that common motivations mentioned to support the teleological account are insufficient to sustain its generalization from children to adults. Moreover, social psychological studies, combined with theoretical considerations, suggest that we do not explain actions mainly by invoking publically accessible, reason-giving facts alone but by ascribing mental states to the agent.

In the philosophy of science, it is a common proposal that values are illegitimate in science and should be counteracted whenever they drive inquiry to the confirmation of predetermined conclusions. Drawing on recent cognitive scientific research on human reasoning and confirmation bias, I argue that this view should be rejected. Advocates of it have overlooked that values that drive inquiry to the confirmation of predetermined conclusions can contribute to the reliability of scientific inquiry at the group level even when they negatively affect an individual’s cognition. This casts doubt on the proposal that such values should always be illegitimate in science. It also suggests that advocates of that proposal assume a narrow, individualistic account of science that threatens to undermine their own project of ensuring reliable belief formation in science.

Why do we engage in folk psychology, that is, why do we think about and ascribe propositional attitudes such as beliefs, desires, intentions etc. to people? On the standard view, folk psychology is primarily for mindreading, for detecting mental states and explaining and/or predicting people’s behaviour in terms of them. In contrast, McGeer (1996, 2007, 2015), and Zawidzki (2008, 2013) maintain that folk psychology is not primarily for mindreading but for mindshaping, that is, for moulding people’s behavior and minds (e.g., via the imposition of social norms) so that coordination becomes easier. Mindreading is derived from and only as effective as it is because of mindshaping, not vice versa. I critically assess McGeer’s, and Zawidzki’s proposal and contend that three common motivations for the mindshaping view do not provide sufficient support for their particular version of it. I argue furthermore that their proposal underestimates the role that epistemic processing plays for mindshaping. And I provide reasons for favouring an alternative according to which, in social cognition involving ascriptions of propositional attitudes, neither mindshaping nor mindreading is primary but both are complementary in that effective mindshaping depends as much on mindreading as effective mindreading depends on mindshaping.

It has been argued that implicit biases are operative in philosophy and lead to significant epistemic costs in the field. Philosophers working on this issue have focussed mainly on implicit gender and race biases. They have overlooked ideological bias, which targets political orientations. Psychologists have found ideological bias in their field and have argued that it has negative epistemic effects on scientific research. I relate this debate to the field of philosophy and argue that if, as some studies suggest, the same bias also exists in philosophy then it will lead to hitherto unrecognised epistemic hazards in the field. Furthermore, the bias is epistemically different from the more familiar biases in respects that are important for epistemology, ethics, and metaphilosophy.

This paper explores the nature of self-knowledge of beliefs by investigating the relationship between self-knowledge of beliefs and one's knowledge of other people's beliefs. It introduces and defends a new account of self-knowledge of beliefs according to which this type of knowledge is developmentally interconnected with and dependent on resources already used for acquiring knowledge of other people's beliefs, which is inferential in nature. But when these resources are applied to oneself, one attains and subsequently frequently uses a method for acquiring knowledge of beliefs that is non-inferential in nature. The paper argues that this account is preferable to some of the most common empirically motivated theories of self-knowledge of beliefs and explains the origin of the widely discussed phenomenon that our own beliefs are often transparent to us in that we can determine whether we believe that p simply by settling whether p is the case.

Recently, philosophers have appealed to empirical studies to argue that whenever we think that p, we automatically believe that p (Millikan 2004; Mandelbaum 2014; Levy and Mandelbaum 2014). Levy and Mandelbaum (2014) have gone further and claimed that the automaticity of believing has implications for the ethics of belief in that it creates epistemic obligations for those who know about their automatic belief acquisition. I use theoretical considerations and psychological findings to raise doubts about the empirical case for the view that we automatically believe what we think. Furthermore, I contend that even if we set these doubts aside, Levy and Mandelbaum’s argument to the effect that the automaticity of believing creates epistemic obligations is not fully convincing.

By drawing on empirical evidence, Matt King and Peter Carruthers have recently argued that there are no conscious propositional attitudes, such as decisions, and that this undermines moral responsibility. Neil Levy responds to King and Carruthers, and claims that their considerations needn’t worry theorists of moral responsibility. I argue that Levy’s response to King and Carruthers’ challenge to moral responsibility is unsatisfactory. After that, I propose what I take to be a preferable way of dealing with their challenge. I offer an account of moral responsibility that ties responsibility to consciously deciding to do X, as opposed to a conscious decision to do X. On this account, even if there are no conscious decisions, moral responsibility won’t be undermined.

Does evolutionary theory have the potential to undermine morality? In his book The Evolution of Morality, Richard Joyce (2006) argues for a positive answer. He contends that an evolutionary account of morality would undermine moral judgements and lend support to moral scepticism. I offer a critique of Joyce’s argument. As it turns out, his case can be read in two different ways. It could be construed as an argument to establish a general scepticism about the justification of moral judgements. Or it could be read as an argument that targets only a particular meta-ethical position, namely moral realism. My claim is that it fails on both interpretations. There is no reason to believe that evolutionary considerations undermine morality.