The Challenges of Hostile Epistemology

Like most political ideals, civility is hotly contested. According to some, civility is central to the democratic project, while others contend that it is superfluous and perhaps even counterproductive. However, contestation over the ideal of civility is burdened by the fact that, whatever civility amounts to, it is supposed to specify the norms of political contestation itself. It is perhaps no wonder, then, that the discourses surrounding civility are so messy. This talk clarifies the role a conception of civility plays in our idea of responsible democratic citizenship. Then, it raises a problem that we must confront in formulating a conception of civility. The tragic upshot is that although democracy calls for a functional conception of civility, it is likely that no such conception is available.

A natural way to understand conspiracy theorists is to see them as advancing alternative hypotheses to mainstream beliefs about socially important phenomena or events. On this framing, they resemble skeptics in epistemology, who argue that our beliefs are unjustified unless we can exclude competing skeptical hypotheses. Climate skeptics, for example, either deny anthropogenic climate change or suspend judgment pending investigation of alternative conspiracy hypotheses. The analogy suggests that conspiracy theorists should be taken seriously, where skeptics operate in the ‘space of reasons’. I argue that this analogy, while initially appealing epistemologically, is misleading.

Conspiracy theorists are heterogeneous in both their motivation and belief structure. Some are doubt-mongers, deploying skeptical hypotheses to advance ideological agendas, often grounded in a political view that reflects distrust of institutions. Others are conspiracy entrepreneurs, who promote conspiracy hypotheses for prudential reasons, like garnering ‘followers’ or personal profit. Still others hold conspiratorial beliefs as part of a broader worldview, where such hypotheses function as ‘hinges’ against which trust and evidence are evaluated. They harbor conspiratorial worldviews.

This reframing supports a differentiated response. Doubt-mongers and entrepreneurs should be exposed and debunked, as their motivations undermine their epistemic standing and can defeat the justification of those they mislead. These ‘skeptics’ should not be engaged on their own terms. By contrast, conspiracy beliefs rooted in a broader worldview are best not addressed through direct refutation, since they structure how evidence and trust are evaluated. Here, indirect strategies—revealing tensions within their broader belief system—are likely more effective. From a liberal democratic perspective, however, such cases may even fall within the bounds of reasonable pluralism, and thus should not be met with blanket intellectual or political hostility. The upshot is that some, but not all conspiracy theory has a place in liberal society.

This paper assesses science denial as a structural form of belief polarization in which political identity and ideology displace truth-tracking norms, generating epistemic obstructions that threaten democratic life. In contrast to individualist "epistemic deficit" accounts, we argue that science denial is a strategic and digitally manufactured phenomenon driven by politically motivated actors that systemically displace the norms, expertise, and institutions constituting scientific inquiry. We draw on Talisse's analysis of belief polarization as a democratic pathology, while arguing that science denial represents a distinct and deeper threat. Through "de-factualization," it erodes the shared epistemic preconditions required for democratic deliberation — undermining both civic friendship and civic solitude as proposed conditions for democratic renewal. Science denial is thus not merely an epistemized political stance, but a systematic assault on the infrastructure of inquiry itself. The paper distinguishes between epistemic and political dimensions of science denial. Epistemically, it does not constitute genuine dissent and must not be framed as a conflict between competing epistemic or rational ideals. Politically, however, a pluralist democracy cannot dismiss claims of symmetry, and engagement remains necessary both to reduce harmful polarization and protect the epistemic infrastructure democratic life requires. This diagnosis raises an urgent normative question: if science denial blocks the resources needed for democratic self-correction, can this autoimmune pathology prove to be lethal? We argue that the severity of epistemic obstruction warrants reconsidering whether epistemic paternalism or protectionism may be justified, and conclude by mapping the tension between these approaches and the democratic values they seek to protect.

Social media are increasingly, as is very well known, among the contributing causes of radical polarization in society. A recent case study is represented by a controversy in Italian football. News emerged publicly that some referees were under investigation for allegedly favouring one of the clubs. No details concerning the basis of the accusations were made public. Nevertheless, a media war started immediately, with bloggers, as well as mainstream journalists supporting rival teams, immediately condemning the mentioned club. However, after a few weeks, it appeared that no official representative of the club was under investigation. There were no public apologies either from bloggers or from mainstream journalists for their earlier categorical accusations, but the atmosphere among supporters became increasingly radicalized and hostile.

This case is representative of a wide range of situations. In order to contain such phenomena, several policies are needed. First, there is the question of institutional responsibility. In the specific case, part of the media speculation was made possible by unofficial and incomplete information regarding the investigation. This demonstrates the need to develop institutional strategies of public communication. For example, rigorous rules aimed at preventing information leaks must be established, bearing in mind that partial leaks distort interpretations of situations and facilitate manipulation. Furthermore, institutions must engage in timely public communication in order to anticipate and contain media manipulation. Second, professional ethical standards concerning journalists must be rigorously enforced. Third, the education of journalists should include training in epistemological skills for the verification of information and for drawing conclusions from it. Fourth, it is time to reorient school education from the preponderance of data provision toward the development of epistemological skills. The former will always remain necessary, but the balance should become more favourable to epistemological skills in a world where information is easily accessible, and where the real challenge lies in how to access, evaluate, and select information, as well as how to draw conclusions from it.

I argue that contemporary hostile epistemic environments are defined not only by the circulation of misinformation, but by the fragmentation of the causal intelligibility of public events. Drawing on Watzlawick’s concept of punctuation, I propose that digital platforms do not merely distribute information: they organise the interpretive sequences through which events become intelligible as provocation, reaction, threat, defence, victimhood, or abuse. The same facts can therefore be inserted into sharply different narrative orders, generating incompatible readings of political reality. I develop this claim in relation to algorithmic curation, networked visibility, and the formation of epistemically dense communities online. In such environments, conflict increasingly concerns the prior question of what is happening at all. What is at stake is not only whether particular claims are true or false, but whether public discourse still provides shared conditions for attributing agency, responsibility, and legitimacy. The broader consequence is the erosion of the epistemic common ground on which liberal-democratic contestation depends.

The publication in 2019 of Crises of Democracy by Adam Przeworski intensified an already ongoing and wide-ranging debate concerning the quality of democracy in Western political regimes. On the one hand, several scholars highlight specific factors that are said to justify this crisis, using them to explain autocratic backsliding, the spread of populist policies, and the erosion of the distinctly democratic features of these political regimes. On the other hand, other scholars argue that such crises are inherently democratic in nature and, in themselves, reveal the resilience of a democratic ideal grounded in citizen participation and representation. Our purpose is to demonstrate that this debate is truncated because it is based on an inadequate understanding of democracy in the twenty-first century. We argue that the specific characteristics of democratic regimes identified by Robert Dahl are no longer sufficient for understanding contemporary democratic systems. On the basis of a more appropriate conception of the democratic characteristics of current political regimes, we seek to show that there is no justification for an extreme or polarised version of this debate, and that the social and political epistemology underpinning it ultimately weakens the exercise of a properly informed citizenship.

Recent philosophical approaches to affective polarization have moved beyond purely identitarian and emotivist frameworks, emphasizing instead its multi-dimensional structure. In The Rise of Polarization (2025), Manuel Almagro identifies five core dimensions—identity, emotion, narrative, credence, and linguistic expression—and argues that political attitudes are shaped not only by ideological commitments but also by life-course contingencies, including socio-economic trajectories and biographical exposures that influence partisan attachments.

This paper builds on Almagro’s analysis while introducing performative contingency as an additional and analytically distinct dimension of affective polarization. Drawing on Greg Myers’ work (2004), we argue that political opinions should not be understood solely as relatively stable attitudes articulated through linguistic practices, but also as socially enacted within specific discursive contexts that shape whether, where, and how political views are expressed. While Almagro explicitly recognizes the context-sensitivity of political utterances, our contribution consists in making this performative dimension analytically explicit as a distinct aspect of affective polarization.

Integrating performative contingency refines the philosophical understanding of affective polarization by highlighting the synchronic contingencies of opinion expression alongside the diachronic contingencies of socialization emphasized in existing accounts. As discussed by Myers (2004), measuring polarization solely through survey responses or public statements risks conflating stable partisan commitments with context-driven performances. Likewise, interventions aimed at mitigating polarization must address not only belief revision and emotional regulation but also the institutional and discursive formats—media architectures, debate structures, and civic forums—that shape the public enactment of political opinion.

While performative contingency may not always be the primary driver of polarization, we argue that it constitutes a pervasive and analytically indispensable dimension of the phenomenon.

This paper examines how hostile epistemology manifests in U.S. higher education’s institutional approaches to international students. Drawing on Miranda Fricker’s concept of epistemic injustice, it investigates the connection between hostile epistemology and higher education’s epistemic approach to international students over the past 25 years, as well as whether this environment is changing how institutions epistemically engage with international students. Recent policy pressures create cascading epistemic harms in the lives of international faculty and students in U.S. institutions. Through document analysis of media discourse and qualitative investigation, the paper examines how institutional responses to political pressures can create chilling effects and restructure relationship-building and knowledge-seeking behaviors. It argues that this represents structural epistemic harm under political pressure, and concludes by sketching interventions that resist closure through cultural exchange and collaborative knowledge-building.

Recent philosophical approaches to affective polarization have moved beyond purely identitarian and emotivist frameworks, emphasizing instead its multi-dimensional structure. In The Rise of Polarization (2025), Manuel Almagro identifies five core dimensions—identity, emotion, narrative, credence, and linguistic expression—and argues that political attitudes are shaped not only by ideological commitments but also by life-course contingencies, including socio-economic trajectories and biographical exposures that influence partisan attachments.

This paper builds on Almagro’s analysis while introducing performative contingency as an additional and analytically distinct dimension of affective polarization. Drawing on Greg Myers’ work (2004), we argue that political opinions should not be understood solely as relatively stable attitudes articulated through linguistic practices, but also as socially enacted within specific discursive contexts that shape whether, where, and how political views are expressed. While Almagro explicitly recognizes the context-sensitivity of political utterances, our contribution consists in making this performative dimension analytically explicit as a distinct aspect of affective polarization.

Integrating performative contingency refines the philosophical understanding of affective polarization by highlighting the synchronic contingencies of opinion expression alongside the diachronic contingencies of socialization emphasized in existing accounts. As discussed by Myers (2004), measuring polarization solely through survey responses or public statements risks conflating stable partisan commitments with context-driven performances. Likewise, interventions aimed at mitigating polarization must address not only belief revision and emotional regulation but also the institutional and discursive formats—media architectures, debate structures, and civic forums—that shape the public enactment of political opinion.

While performative contingency may not always be the primary driver of polarization, we argue that it constitutes a pervasive and analytically indispensable dimension of the phenomenon.

The epistemology of public discourse has been experiencing a kind of renaissance, but this abundance frequently consists of parallel investigations that rarely explicitly discuss or argue their differences. The aim of this paper is to offer a structure for the main disagreements in the epistemology of public discourse. My proposal structures the domain as a conflict between monists and pluralists, maximalists and minimalists, internalists and externalists, and interface priority and political economy priority. Furthermore, I present these positions as connected, and the epistemology of public discourse as reducible to two sides: pluralism, minimalism, externalism, and the priority of political economy, and monism, maximalism, internalism, and the priority of interface. This structure provides investigators with an overview of the lines of discussion and opens the question of whether different combinations and compatibilities of the stated positions are possible.

This paper assesses science denial as a structural form of belief polarization in which political identity and ideology displace truth-tracking norms, generating epistemic obstructions that threaten democratic life. In contrast to individualist "epistemic deficit" accounts, we argue that science denial is a strategic and digitally manufactured phenomenon driven by politically motivated actors that systemically displace the norms, expertise, and institutions constituting scientific inquiry. We draw on Talisse's analysis of belief polarization as a democratic pathology, while arguing that science denial represents a distinct and deeper threat. Through "de-factualization," it erodes the shared epistemic preconditions required for democratic deliberation — undermining both civic friendship and civic solitude as proposed conditions for democratic renewal. Science denial is thus not merely an epistemized political stance, but a systematic assault on the infrastructure of inquiry itself. The paper distinguishes between epistemic and political dimensions of science denial. Epistemically, it does not constitute genuine dissent and must not be framed as a conflict between competing epistemic or rational ideals. Politically, however, a pluralist democracy cannot dismiss claims of symmetry, and engagement remains necessary both to reduce harmful polarization and protect the epistemic infrastructure democratic life requires. This diagnosis raises an urgent normative question: if science denial blocks the resources needed for democratic self-correction, can this autoimmune pathology prove to be lethal? We argue that the severity of epistemic obstruction warrants reconsidering whether epistemic paternalism or protectionism may be justified, and conclude by mapping the tension between these approaches and the democratic values they seek to protect.

The field of vice epistemology has focused on individual character flaws such as gullibility, laziness and bias. More recently, Nguyen has expanded the scope of inquiry to “hostile epistemology”—environmental conditions that exploit unavoidable cognitive weaknesses. However, hostile epistemology remains focused on orthodox vulnerabilities like heuristics and trust. Drawing on continental and critical approaches, I extend vice and hostile epistemology to a new category of epistemic vices and vulnerabilities that I call “critical epistemic vices”. I identify three kinds of such vices and try to recast them in more analytic form: ideological hegemony disguised as “common sense” explanations of the world, thought-ending clichés, and the tendency to moralize. I explain how structural conditions—intentionally constructed or not—shape individuals’ beliefs and belief systems in a way that is epistemically flawed and harmful, but not accounted for by existing explanations in analytic vice and hostile epistemology.

This paper examines the democratic significance of weak social ties in increasingly polarized societies. Drawing on Mark Granovetter's theory, it argues that weak ties foster exposure to diverse perspectives and facilitate empathy by connecting individuals beyond homogeneous social networks. Unlike strong ties, which often reinforce existing beliefs, weak ties mitigate the formation of ideological and affective enclaves. They provide a foundation for democratic citizenship by broadening horizons of understanding, thus supporting the capacities for reflection and mutual recognition. In this respect, weak ties complement Robert Talisse's concept of civic solitude by supplying the diverse viewpoints upon which meaningful democratic reflection depends. Rather than treating increased political deliberation as a sufficient remedy for polarization, the paper highlights the importance of social structures that enable citizens to encounter difference in less adversarial ways. Weak social ties are thus presented as a valuable democratic resource for sustaining civic engagement and resisting polarization.

The talk discusses reasons to (not) trust some source of purported knowledge. It considers under which conditions one should trust scientists, politicians, journalists and ordinary citizens. Further, it problematizes to which extent do we live in a “hostile epistemology”, and aims to propose practicable imperfect epistemological rules of thumb to orient in contemporary information environment.

Political knowledge is usually seen as essential to good democratic citizenship, yet research shows that knowledgeable citizens are not necessarily less polarized or biased; often the opposite is true. Meanwhile, efforts to diminish bias and promote political objectivity risk dampening the political engagement democracy depends on. Caught between the hammer of political bias and the hard place of political apathy, democratic citizens seem to face an epistemic conundrum. I argue this tension can be eased by cultivating what I call Mandevillian wisdom: a metacognitive virtue involving awareness of how political identity shapes belief formation and public reasoning. Mandevillian wisdom does not aim to eliminate bias. Rather, it recognizes that some bias can be productive, as recent work on collective Mandevillian intelligence shows. What matters is whether individual biases contribute to constructive democratic deliberation or spiral into structural pathologies. Mandevillian wisdom is a context-sensitive virtue that helps citizens deploy their political commitments in ways that sustain deliberative engagement rather than foreclosing it. This offers a more realistic ideal of democratic epistemic competence — one preserving the passions democracy requires while improving collective judgment.

Building on C. Thi Nguyen’s notion of hostile epistemology, I clarify its central features and further examine the inherent epistemic hostility of novel digital phenomena. Presupposing that epistemically hostile environments exploit the inevitable vulnerability of our attempts to make sense of the world with limited temporal, cognitive, and contextual resources, I illustrate how social networks, nascent forms of native advertising, and deliberate disinformation manipulate our pursuit of clarity, understanding, and belonging. I locate a further vulnerability in Thomas Reid’s influential discussion of testimony. Accepting Reid’s principle of credulity – or, in other words, the stance that people are naturally inclined to believe what others tell them – on pragmatic and epistemological grounds confronts us with the evident incongruity of epistemic agents prone to automatic trust and a hostile environment fabricated to disinform, persuade, or confuse them. I close the discussion by examining the implications of this misalliance for social and political polarization, democratic decision-making, civic participation, and personal responsibility.

The ethics training of Large Language Models (LLMs) raises critical questions about whose judgments and which rationales underpin the alignment of their normative behavior. Dominant RLHF-based approaches that utilize English-centric datasets such as the ETHICS dataset often reflect the preferences of homogeneous, predominantly Western annotators. As González Barman, Lohse, and de Regt (2025) argue, the cultural and social composition of feedback providers significantly influences model outputs, risking epistemic biases, value lock-in, and the marginalization of less-resourced linguistic communities. In short, AI ethics training should treat annotated cases as culturally situated judgments plus explicit reasons, not as isolated, neutral, and portable labels. This talk presents the initial steps of a reasons-based, multidimensional, and pluralistic ethics training framework developed for GaMS LLM, Slovenia’s open-source sovereign LLM. The project adapts the ETHICS dataset through culture-specific annotators, curating scenarios and enriching them with explicit rationales grounded in a pluralistic normative structure (moral principles such as harm prohibition, veracity, promissory fidelity, justice, and respectfulness). The initial results and comparison of such annotation with annotations done by the dominant, globally focused LLMs such as ChatGPT, Clause, and Gemini, reveal striking differences: LLMs show high uniformity, heavily anchoring on aspects like harm, disrespect, and justice, while human annotators display greater heterogeneity in rationale selection and contextual sensitivity, invoking a broader range of principles and implicit social norms. By emphasizing culturally situated moral grammars and explicit reasons-responsiveness, the approach advances moral pluralism and supports democratic accountability of ethics training for less-resourced language models.

Key words: AI ethics, AI alignment, ethics training of AI, Large Language Models (LLMs), moral competence, reasons-responsiveness, cultural diversity, pluralism, less-resourced languages.

The ethics training of Large Language Models (LLMs) raises critical questions about whose judgments and which rationales underpin the alignment of their normative behavior. Dominant RLHF-based approaches that utilize English-centric datasets such as the ETHICS dataset often reflect the preferences of homogeneous, predominantly Western annotators. As González Barman, Lohse, and de Regt (2025) argue, the cultural and social composition of feedback providers significantly influences model outputs, risking epistemic biases, value lock-in, and the marginalization of less-resourced linguistic communities. In short, AI ethics training should treat annotated cases as culturally situated judgments plus explicit reasons, not as isolated, neutral, and portable labels.

This talk presents the initial steps of a reasons-based, multidimensional, and pluralistic ethics training framework developed for GaMS LLM, Slovenia’s open-source sovereign LLM. The project adapts the ETHICS dataset through culture-specific annotators, curating scenarios and enriching them with explicit rationales grounded in a pluralistic normative structure, including moral principles such as harm prohibition, veracity, promissory fidelity, justice, and respectfulness.

The initial results and comparison of such annotation with annotations done by dominant, globally focused LLMs such as ChatGPT, Claude, and Gemini reveal striking differences: LLMs show high uniformity, heavily anchoring on aspects like harm, disrespect, and justice, while human annotators display greater heterogeneity in rationale selection and contextual sensitivity, invoking a broader range of principles and implicit social norms. By emphasizing culturally situated moral grammars and explicit reasons-responsiveness, the approach advances moral pluralism and supports democratic accountability of ethics training for less-resourced language models.

Keywords: AI ethics, AI alignment, ethics training of AI, Large Language Models, moral competence, reasons-responsiveness, cultural diversity, pluralism, less-resourced languages.

Disinformation posed a threat to epistemic trust long before the rise of generative artificial intelligence (Gen AI). However, the development of Gen AI brought several challenges to trust that were difficult to foresee. The aim of this talk is to outline an integrated account of these effects. Besides amplifying the speed and quantity of disinformation, Gen AI enables the democratisation of its generation and provides different tools for increasing credibility and persuasiveness. In addition, AI-generated disinformation poses issues for both human evaluators and AI-powered detectors. Pointing out several features of AI-generated disinformation and their detection, I argue that Gen AI does not merely amplify the dissemination of disinformation but also challenges the conditions under which epistemic trust is formed, distributed, and evaluated in contemporary digital society.