Bias

This interdisciplinary research explores the concept of Pure Reason not as a fixed cognitive state, but as a dynamic, self-reflective process that bridges philosophy, cognitive science, and artificial intelligence. Drawing from Kantian epistemology, cognitive psychology, neurobiology, and AGI architectures, the project proposes a model of rationality that includes the ability to identify and transcend its own limitations. This work argues that Pure Reason is not a natural capacity, but a practiced, meta-cognitive methodology — an emergent structure that only arises through (...) recursive analysis and epistemic honesty. (shrink)

According to an influential research program in cognitive science, philosophy, and linguistics, there is a deep, special connection between generics and pernicious aspects of social cognition such as stereotyping. Specifically, generics are thought to exacerbate our propensity to essentialize, lead us to overgeneralize based on scarce evidence, and lead to other epistemically dubious patterns of inference. Recently, however, several studies have put empirical and theoretical pressure on some of the main tenets of this research program. The goal of this paper (...) is to bring these results together in a comprehensive narrative and systematically evaluate their impact on the hypothesis that generics have a uniquely problematic effect on our social and cognitive capacities. (shrink)

1. EXPLICATING THE CONCEPT OF REFLECTION (under review) -/- To understand how ‘reflection’ is used, I consider ordinary, philosophical, and scientific discourse. I find that ‘reflection’ seems to refer to reasoning that is deliberate and conscious, but not necessarily self-conscious. Then I offer an empirical explication of reflection’s conscious and deliberate features. These explications not only help explain how reflection can be detected; they also distinguish reflection from nearby concepts such as ruminative and reformative reasoning. After this, I find that (...) reflection is not obviously limited to only practical or only theoretical reasoning. The chapter ends with reasons to prefer ‘unreflective’ and ‘reflective’ to dual process theorists’ labels about systems or types. -/- 2. BOUNDED REFLECTIVISM & EPISTEMIC IDENTITY (in Metaphilosophy) -/- Reflection features centrally in philosophy (e.g., Korsgaard, 1996; Rawls, 1971; Sosa, 1991) and psychology (e.g., Pennycook, 2018). Bounded reflectivism is an empirically adequate model of reflection that explains reflection’s capacity to either help or hinder reasoning—e.g., reflective equilibrium vs. reflective polarization (e.g., Kahan et al., 2017). One innovation of the model is epistemic identity: an identity that involves particular beliefs—e.g., religious and political identities. When we feel that our epistemic identity is threatened, we can reflectively defend its beliefs rather than update our beliefs according to the best arguments and evidence. The solution, I argue, is not to suppress epistemic identity but embrace it by appealing to shared, superordinate epistemic identities. -/- 3. ALL MEASURES ARE NOT CREATED EQUAL (under review) -/- A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost? “10 cents” comes to most people’s minds immediately. However, upon reflection, we can realize that “5 cents” is the correct answer. There are many reflection tests. Each has its limitations. Some reflection tests are mathematical—like the bat-and-ball problem (e.g., Frederick, 2005) and others are logical (e.g., Janis & Frick, 1943). So there are concerns that reflection tests track logical and mathematical competence rather than reflection per se. Also, some psychologists assume a priori that correct answers are reflective and lured answers are unreflective. New evidence raises additional concerns about the plausibility of these assumptions. I argue that think aloud protocols (Ericsson & Simon, 1998) and process dissociation (Jacoby, 1991) can assuage some of these concerns. -/- 4. GREAT MINDS DO NOT THINK ALIKE (in Review of Philosophy and Psychology) -/- Two large studies (N = 1299), one pre-registered, found that many correlations between reflection and philosophical beliefs among non-philosophers replicated among philosophers. For example, less reflective philosophers preferred theism to atheism and utilitarian rather than deontological responses to the trolley problem (Hannikainen & Cova, in prep.; Pennycook et al., 2016; Reynolds, Byrd, & Conway, forthcoming). However, philosophical judgments were sometimes better predicted by factors like education, gender, and personality than by reflection test performance. So although some relationships between reflection and philosophy were robust, there is more to the link between reflection and philosophy. Normative implications are also discussed—e.g., how we can infer the quality of philosophical views from their correlations with reflective or unreflective reasoning. -/- 5. WHAT WE CAN (AND CAN’T) INFER ABOUT IMPLICIT BIAS (in Synthese) -/- Contrary to some philosophers and psychologists, I argue that implicit bias is probably associative. However, I also argue that debiasing is not thereby entirely unconscious and involuntary. Indeed, strong evidence suggests that reflection can change implicitly biased behavior—e.g., conscious and deliberate counterconditioning. This has implications for the science and morality of implicit bias—e.g., evidence suggests that we can reform biases; so. intuitively, we can be responsible for biases. (shrink)

Was ist Wissen von morgen im Vergleich mit Wissen von heute? Die geschichtliche Erfahrung zeigt diesen Übergang als ein Prozess der Weiterentwicklung der Sprache, weil menschliches Wissen an die Sprache gekoppelt ist, und seine Inhalte zu begreifen nur in sprachlichen Formen möglich ist. Während alte Sprachen vergehen, entstehen neue Sprachen: in diesem Prozess der Erneuerung von sprachlichen Formen wird das Wissen gesammelt, geklärt, und weitergegeben, und das geschieht durch Auswahl der Worte, deren Gebrauchswert ständig geprüft wird.

Une des contributions phares de Luc Bégin à l’éthique publique est d’avoir bien documenté les obstacles auxquels les gardiens institutionnels québécois doivent porter une attention particulière. Bégin s’est notamment intéressé aux conditions dans lesquelles les ordres professionnels, comme l’Ordre des ingénieurs du Québec, sont moins susceptibles de protéger le public contre des pratiques répréhensibles (corruption, collusion, etc.). Je poursuis les réflexions de Bégin à partir d’un point de vue « épistémique ». Je montre d’abord comment les ordres professionnels ont tendance (...) à produire des connaissances sur l’état de la profession à partir des informations partagées par leurs propres membres. J’analyse comment cette pratique peut générer un biais du survivant, et obscurcir des problèmes centraux à une profession. Finalement, je propose des pistes de solution à ce problème pour les ordres professionnels. (shrink)

In this paper, we draw on Alfred Schutz’s theoretical framework to better understand how oppression is enacted through discriminatory acts. By closely examining the role of typifications and relevances in our experience of others, and by supplementing this analysis with contemporary social scientific resources, we argue that a Schutzian perspective on oppression yields important phenomenological insights. We do this in three key steps. Firstly, we contextualise Equality and the Meaning Structure of the Social World within Schutz’s broader body of work, (...) elucidating his central conditions of discrimination. Secondly, we highlight the limitations of Schutz’s account, in how it fails to capture more peripheral, subtle, and implicit oppressive practices. Finally, we introduce two underexplored insights derived from Schutz’s framework: the role of relevances in understanding the motivational underpinnings of implicit biases, and the connection between self-typification and stereotype threat. With this multidimensional approach, we hope to enhance our understanding of oppression whilst bridging gaps in Schutz’s original conceptualization. (shrink)

The method of intuition, the view that the best philosophical perspective will maintain as many of our intuitions as possible, is one of the pillars of analytic philosophy. Unfortunately, the reliance on intuition by analytic philosophers has created conditions such that the biases of those who do philosophy, predominantly those of hegemonic identities, are accepted as a basis for philosophical knowledge. This problem can be solved by rejecting intuition as a basis for philosophical knowledge and instead relying on a methodology (...) derived from the Principle of Sufficient Reason. (shrink)

This paper takes a stand on Hamblin semantics and its relation to the semantics-to-pragmatics interface. Biased questions, where the speaker finds one of the options more likely and expects the confirmation that p is true, raise a concern about the limits of Hamblin semantics. I argue that biased questions have modified Hamblin semantics, while unbiased questions have unconstrained Hamblin semantics. The optional bias feature explains compositionally. It is triggered by likelihood presuppositions ranging Hamblin sets and highlighting the preferred alternative(s). Biased (...) effects are exceptionally possible in alternative and wh-questions, but only in polar questions, they can be encoded lexically and grammatically. I argue that Hamblin semantics covers all core types of questions, while some of its applications for non-questions are problematic. (shrink)

While standard forms of discrimination are widely considered morally wrong, philosophers disagree about what makes them so. Two accounts have risen to prominence in this debate: One stressing how wrongful discrimination disrespects the discriminatee, the other how the harms involved make discrimination wrong. While these accounts are based on carefully constructed thought experiments, proponents of both sides see their positions as in line with and, in part, supported by the folk theory of the moral wrongness of discrimination. This article presents (...) a vignette-based experiment to test empirically what, in the eyes of “folks”, makes discrimination wrong. Interestingly, we find that, according to folks, both disrespect and harm make discrimination wrong. Our findings offer some support for a pluralistic account of the wrongness of discrimination over both monist respect-based and monist harm-based accounts. (shrink)

Artificial intelligence chatbots driven by large language models (LLMs) have the potential to increase public science literacy and support scientific research, as they can quickly summarize complex scientific information in accessible terms. However, when summarizing scientific texts, LLMs may omit details that limit the scope of research conclusions, leading to generalizations of results broader than warranted by the original study. We tested 10 prominent LLMs, including ChatGPT-4o, ChatGPT-4.5, DeepSeek, LLaMA 3.3 70B, and Claude 3.7 Sonnet, comparing 4900 LLM-generated summaries to (...) their original scientific texts. Even when explicitly prompted for accuracy, most LLMs produced broader generalizations of scientific results than those in the original texts, with DeepSeek, ChatGPT-4o, and LLaMA 3.3 70B overgeneralizing in 26–73% of cases. In a direct comparison of LLM-generated and human-authored science summaries, LLM summaries were nearly five times more likely to contain broad generalizations (OR = 4.85, 95% CI [3.06, 7.70], p < 0.001). Notably, newer models tended to perform worse in generalization accuracy than earlier ones. Our results indicate a strong bias in many widely used LLMs towards overgeneralizing scientific conclusions, posing a significant risk of large-scale misinterpretations of research findings. We highlight potential mitigation strategies, including lowering LLM temperature settings and benchmarking LLMs for generalization accuracy. (shrink)

Identity-protective reasoning---motivated reasoning driven by defending a social identity---is often dismissed as a paradigm of epistemic vice and a key driver of democratic dysfunction. Against this view, I argue that identity-protective reasoning can play a positive epistemic role, both individually and collectively. Collectively, it facilitates an effective division of cognitive labor by enabling groups to test divergent beliefs, serving as an epistemic insurance policy against the possibility that the total evidence is misleading. Individually, it can correct for the distortions that (...) arise from taking ideologically skewed evidence at face value. This is particularly significant for members of marginalized groups, who frequently encounter evidence that diminishes the value of their identities, beliefs, and practices. For them, identity-protective reasoning can counter dominant ideological ignorance and foster resistant standpoint development. While identity-protective reasoning is not without risks, its application from marginalized and counter-hegemonic positions carries epistemic benefits crucial in democracies threatened by elite capture. Against dominant views in contemporary political epistemology and psychology, identity-protective reasoning should be reconceived as a resource to be harnessed and not a problem to be eradicated. (shrink)

The principle of total evidence says that all relevant information should be considered when making an inference about a hypothesis. In this article, we argue that the criminalist’s paradox from the literature on the methodology of forensic science constitutes a counterexample against the principle of total evidence. The paradox arises, for example, when a forensic scientist uses the results from other forensic procedures to inform their own analysis. In such cases, their results can become more reliable, but at the same (...) time also dependent on those other forensic results and therefore less useful for decisions in court cases. Consequently, such influences are generally avoided in forensic science. We argue that structurally similar problems plague also other scientific disciplines, anticipate an objection to the counterexample, and propose two new versions of the principle of total evidence that do not fall prey to the criminalist’s paradox. (shrink)

Statistical fairness criteria are widely used for diagnosing and ameliorating algorithmic bias. However, these fairness criteria are controversial as their use raises several difficult questions. I argue that the major problems for statistical algorithmic fairness criteria stem from an incorrect understanding of their nature. These criteria are primarily used for two purposes: first, evaluating AI systems for bias, and second constraining machine learning optimization problems in order to ameliorate such bias. The first purpose typically involves treating each criterion as a (...) necessary condition for fairness. The second use involves treating criteria as sufficient conditions for fairness. Since the criteria are used for both roles, some researchers have treated them as both necessary and sufficient conditions, i.e., as definitions of algorithmic fairness. However, serious problems have been raised for the use of these fairness criteria. Under ordinary circumstances, it is impossible to satisfy multiple criteria at the same time. Moreover, there are counterexamples to both the sufficiency and necessity for fairness of each criterion. I argue that we should instead understand fairness criteria as merely providing evidence of fairness. In other words, satisfaction (or violation) of these criteria should be understood as potential evidence of fairness (or bias). Whether a criterion counts as evidence in a particular case will depend on stakeholders' background knowledge and the specific features of the system's task. This evidence account of fairness conditions provides guidance for recognizing both the appropriate uses and the limitations of fairness criteria. (shrink)

This article synthesizes some classic methodological keys about sample decisions in social research and attempts to highlight some dimensions that are not always problematized. First, we define the concepts of sample and universe, and ask general questions about what sampling decisions involve. Then, it is proposed to think the variety of sampling procedures by organizing them in two large families: case selection guided by experimental logic and case selection based on random sampling. Although the types of samples are not reduced (...) to these two, nor are they exclusive, they allow us to understand the rationales underlying different kinds of research. (shrink)

Over the last hundred years, competing and incompatible positions in relation to basic problems of knowledge and the use of the verb ‘to know’ have multiplied; and the prospect of a consensus solution emerging with respect to any of the problems has not seemed particularly good. We have a Gordian knot. Even so, I suggest that we also have a way to cut it. This will involve identifying why the cognitive mechanism that produces our intuitions of knowing evolved and was (...) maintained (by natural selection), i.e., identifying the ‘teleonomic function’ of that cognitive mechanism. Also, it will involve predicting, on the basis of this teleonomic function, the triggering-conditions of these natural knowledge intuitions. In this thesis, I develop a general theory of the origin, function and triggering-conditions of knowledge intuitions that will allow us to cut that knot. That theory follows basic biological theory (including that which pertains to natural altruism) and also signal detection theory. -/- My theory identifies a number of different circumstances under which the triggering-conditions of knowledge intuitions are different. Strikingly, these different circumstances (and their associated triggering-conditions) map onto the different competing and incompatible epistemological positions to which I referred. This suggests that these positions are all correct within the boundaries of one of the circumstances that my theory identifies; and that the Gordian knot is largely the result of epistemologists claiming universal applicability of a theory that in fact only applies under particular circumstances. We cut the knot by specifying the different circumstances under which each of the different epistemological positions will hold, and the reason we should expect it to hold in just these circumstances, in light of the teleonomic function of knowledge intuitions. (shrink)

Error Management Theory (EMT) suggests that cognitive adaptations evolve to minimize the cost of false negative and false positive errors in detections of consequential environmental conditions. These adaptations manifest as biases tailored to specific environmental conditions. This paper proposes that the same selection pressure fostered the evolution of a self-biasing ability, allowing us to minimize such costs based on experience and culturally transmitted information. The research indicates that this ability specifically applies to productions of belief or doubt about the existence (...) of an environmental condition that is not adequately perceptible to the senses, e.g., doubt that the man is being honest. A model of this self-biasing process, rooted in signal detection theory, is developed and its explanatory reach is demonstrated through many diverse examples from epistemological and social psychological literatures. These examples feature an evidentiary standard for belief that appears too high or low, or seems to have modulated. This study aims to correct and to significantly enhance EMT by introducing this evolved self-biasing ability. (shrink)

People often discriminate based on negative or positive stereotypes about others. Important examples of this are highlighted by the theory of ambivalent sexism. This theory distinguishes sexist stereotypes that are negative (hostile sexism) from those that are positive (benevolent sexism). While both forms of sexism are considered wrong towards women, hostile sexism seems intuitively worse than benevolent sexism. In this article, we ask whether the difference between discriminating based on positive vs. negative stereotypes in itself makes a morally relevant difference. (...) We suggest that it does not. By examining a number of prominent accounts of what makes discrimination wrong, we defend the Moral Irrelevance View according to which stereotype valence is irrelevant to the moral evaluation of discrimination, all else equal. (shrink)