Abstracts (Day 2)

Seminar Room 7: HPS, Physics, Metaphysics

Reichenbach’s probabilistic Realism - In what sense it can be compatible with contemporary pragmatic realism?

Hyung goo Kang

Hans Reichenbach(1891-1953), who was an eminent philosopher of Logical Empiricism, defended a weak version of scientific realism in his Experience and Prediction (1938). His version of scientific realism can be characterized as “probabilistic realism”, because he repeatedly emphasized the concept of probability in his epistemology and exhibited a clear commitment of its centrality in his scientific realism. But his realism can be also characterized as ‘pragmatic realism’, because he argued that the probability concept can make scientific propositions(sentences) “usable(available)” for our scientific practices. If Reichenbach’s probabilistic realism was a kind of pragmatic realism, then what is the relation between his version of realism and contemporarily reviving pragmatic realism such as Hasok Chang’s active realism(Realism of realistic people, 2022)? In this paper, I will compare Reichenbach’s realism with Chang’s realism and argue that they are on a continuous line of empiricist tradition.

Between Realism and Perspective: The Limited Force of Massimi’s Perspectival Realism

Kai Zhang

This paper critically examines Michela Massimi’s perspectival realism as a proposed middle ground in the debate between scientific realism and anti-realism. It argues that while Massimi’s account offers a compelling, practice-oriented account of how perspectival science can yield knowledge of a mind-independent world, it ultimately constitutes a weaker form of scientific realism rather than a fundamentally distinct position. First, I distill the core commitments of traditional scientific realism – metaphysical, semantic, and epistemological – to establish a baseline for comparison. Second, I provide a detailed exposition of Massimi’s perspectival realism. Third, I identify internal tensions within Massimi’s account – the key concept of “lawlike dependency” is critically underdeveloped, and her theory of truth struggles to bear its heavy epistemic load. Finally, I argue that Massimi’s perspectival realism shares metaphysical and epistemological stances, yet licensing fewer realist commitments without surpassing traditional realism in resolving its central challenges.

What’s True in a Realism about Structures?

Kent Win Ho

Structural realists claim that the history of theory change vindicates their brand of scientific realism because there are structural continuities despite ontological discontinuities about unobservable entities. However, the history of theory change has also shown a prevalence of structural losses, such that a pessimistic induction can be made on relations and not just objects. This has prompted some structuralists to advocate for structural empiricism and some structural realists to abandon the preservationist thesis. The root of these problems lie with a conception of truth borrowed from standard scientific realism: truth is defined in terms of correspondence between representations and the world. For structural realism to genuinely distinguish itself from standard accounts of scientific realism, I recommend a shift from a theory of structuralist truth as ‘truth in a structure’ to that of ‘truth of a structure’, linking the truth of theoretical structures with their roles in inferential contexts.

Quidditism and Laws of Nature

Ju Yeon Kim

Quidditism holds that fundamental properties possess primitive identities independent of their nomic roles. This commitment generates Ramseyan Humility: multiple qualitatively indiscernible worlds could be actual, and we cannot know which one is. While this epistemic cost is often accepted as unavoidable, its extent depends on one’s theory of laws. I introduce a framework distinguishing the Mosaic (M), the qualitative pattern of a world, from the Quiddity Mosaic (MQ), which adds facts about which properties realize that pattern. Under Humean Supervenience (HS), laws supervene on MQ, yielding exactly k indiscernible worlds. Under Anti-Humean Supervenience (AHS), laws are primitive and independent of MQ, expanding the ignorance space to at least k × m worlds. I argue that AHS-Quidditism incurs additional costs: it violates both Occam’s Razor and Hume’s Razor simultaneously, and generates “double humility”, the independent ignorance of both quiddities and laws. Quidditists should therefore prefer Humean Supervenience.

The parallel between pure mathematics and analytic metaphysics. What can we learn?

Fabio Ceravolo

Pure mathematics is applicable and therefore valuable. Mainstream analytic metaphysics is also applicable, in a similar sense. Therefore, withholding value from mainstream metaphysics requires special justification. Recent philosophers endorse versions of this argument but the first premise makes the applicability of mathematics sufficient for value and ‘applicable’ is polysemous. It is not clear that on every reading of this term being applicable guarantees value. So I recommend a different parallel between pure mathematics and metaphysics. Take the class of past events that we would agree to consider causal outcomes of mathematics ‘being used for scientific purposes’ (on reasonable interpretations of that phrase). Some such events are good in respects that critics of metaphysics do not deem trivial. Therefore, critics should not argue from the value of these consequences to blanket claims that applying metaphysics to science has only trivially good outcomes. Any justification for such claims must come from elsewhere.

Reduction and Reductive Explanation: From Statistical Mechanics to Statistical Physics

Nanxin Wei

Philosophical discussions of statistical mechanics often frame it as a reduction of thermodynamics via two approaches: Boltzmannian and Gibbsian statistical mechanics. Although complementary, the Boltzmannian approach is typically regarded as more fundamental due to its clearer ontological commitments, while the Gibbsian approach is favored by practicing physicists for its computational power. However, the orthodox Boltzmannian approach has limited scope for underpinning modern thermodynamics, which is itself essentially statistical, and alternative conceptions of reduction may align better with the Gibbsian approach and its indispensability in practice. More importantly, philosophical work has largely overlooked that statistical physics primarily aims not at metaphysical reduction but at providing reductive explanations of complex phenomena with many degrees of freedom. The Gibbsian framework supports this explanatory role through a mode of reasoning best described as “abduction to characterization”, enabling the construction of generative mechanisms under physical constraints that produce salient characterizations of the target phenomenon.

Why unboundedness is not a problem for quantum mechanics? In response to Carcassi, Calderón, and Aidala

Zhonghao Lu

In a recent paper, Carcassi, Calderón, and Aidala argue that conventional Hilbert spaces contain unphysical states of which the expectation values of physical observables are undefined, and suggest that we should use some dense subsets of Hilber spaces, the Schwartz spaces for instance, to represent physical states in quantum theories. I demonstrate that the existence of unbounded self-adjoint operators is the cause for their worry. I argue that we have reasons to treat all states in the whole Hilbert spaces as physically meaningful. By the spectral decomposition, physical observables represented by unbounded self-adjoint operators have a well-behaved probability distribution on any states in the Hilbert spaces, and their expectation values cannot be directly observed. Furthermore, replacing Hilbert spaces with the Schwarz spaces in quantum theories will bring many problems. More specifically, Schwarz spaces are not closed under some generic Hamiltonian evolutions, and excluding these Hamiltonians is ungrounded.

Object Boundaries, State Attribution, and Quantum Entanglement

Juhee Han

This paper examines the status of objecthood by following a conceptual trajectory from Max Black’s two-sphere example to Simon Saunders’ account of weak discernibility. Black’s case shows that numerical distinctness can depend on relational properties without appealing to intrinsic properties. By contrast, in cases of quantum entanglement, the unit of physical description shifts from the parts to the composite system as a whole. Saunders takes weak discernibility to be sufficient for objecthood, a move that leads to an ontological asymmetry between fermions and bosons. This paper argues that this asymmetry does not arise from any physical difference due to entanglement itself, but rather from a restrictive criterion of objecthood based on discernibility. Moreover, it shows that the more fundamental problem posed by quantum entanglement concerns the boundaries of state attribution rather than the numerical distinctness of objects, thereby motivating a reconsideration of system-centered approaches to quantum ontology.

Neurath’s Ballungen: A Postmodern and Feminist Reading

Sharon Yoon

Logical Empiricism prima facie appears to be opposed to Postmodernism or Feminist philosophy of science. However, Neurath’s anti-foundationalist and anti-reductionist position, unlike other Vienna Circle members including Carnap, can be revisited from Postmodernist and Feminist viewpoints. I will analyze Ballungen, the concept of imprecise verbal clusters that constitute everyday language, as the core of Neurath’s philosophy. Then, I will claim that Ballungen accords with Jacques Derrida’s concept of différance, where meaning emerges through relational differences rather than fixed reference. This postmodern reading of Ballungen as inherently non-foundational and relationally constituted extends to Donna Haraway’s notion of “Situated Knowledge” in feminist epistemology and philosophy of science. Yet unlike postmodernists/feminists who question the objectivity itself and view knowledge as subordinated to specific power structures, Neurath had the grand goal of starting from Ballungen to create an intersubjectively verifiable scientific language through which science would progress toward unified science.

Counter Indispensabilist Platonism: Counterfactual Explanations and the Indispensability Argument

Chen Zhibin

Although the Explanatory Indispensability Argument has been informed by discussions on scientific explanation, it has been viewed as reaching a stalemate. This paper argues that this ontological debate should give greater attention to epistemological dimensions of distinctively mathematical explanations, and how they potentially constrain the ontological debate. It examines the prominent counterfactual programme, and argues that each of its three components faces a stability problem for indispensabilist platonism. The non-vacuous counterpossibles semantics threatens the absoluteness and mind-independence of mathematical truth; the “mathematical twiddle” presupposes counterfactual variation of abstracta that strains platonist identity conditions; the fixed morphism linking mathematics to physics creates a new epistemological challenge. The upshot is that counterfactual models of distinctively mathematical explanations impose epistemological constraints that platonist ontologies struggle to satisfy.

Seminar Room 8: AI, Technology, Mind

The Misleading Question of AI Consciousness: Subjectivity, Evolution, and the Limits of Functional Criteria

Qiaoying Lu, Liqian Zhou

We argue that the popular question “Can AI be conscious?” is conceptually misleading. Current debates often rely on a “checklist approach,” inferring consciousness from the presence of certain cognitive functions in AI systems. I show that such function-based criteria cannot bridge the explanatory gap between objective performance and subjective experience, as highlighted in classic discussions by Nagel and Searle. Drawing on evolutionary theories of animal consciousness, we propose that consciousness is grounded not in cognitive complexity but in subjectivity: a first-person point of view possessed by self-maintaining organisms for whom things matter. Minimal biological cases indicate that subjectivity evolved gradually and is rooted in biological organization. By contrast, contemporary AI systems lack embodied, self-sustaining, valenced organization and thus provide no basis for ascribing consciousness. We hence should start with a prior question: “What kinds of systems could host subjectivity at all?”

How LLMs Challenge Enactive Agency: A Critical Examination of Embodied Cognition in the Age of AI

Xue Yu and Zihan Gao

The unexpected success of Large Language Models (LLMs) challenges the core enactivism thesis that cognition requires biological embodiment. This paper critically examines Tom Froese’s “AI Dilemma”, which suggests LLMs may possess a “latent embodiment” derived from training data, effectively functioning as non-biological sense-makers. This paper argue against this reconceptualization by revisiting the minimal conditions for genuine agency: self-individuation (autopoiesis), interactional asymmetry, and normativity. The analysis demonstrates that while LLMs exhibit sophisticated structural mimicry, they categorically fail to possess autonomous organizational closure or intrinsic concern. Consequently, this paper conclude that LLMs simulate linguistic competence without achieving genuine sense-making. To preserve the insights of the life-mind continuity thesis, enactivism theory must maintain a strict distinction between the statistical processing of language and the authentic, lived enactment of meaning inherent to biological life.

AI and the Limits of Language: A Wittgensteinian Therapeutic Examination of AI Subjectivity

Song Jing Li

The development of artificial intelligence is deeply intertwined with philosophy’s linguistic turn. Turing’s pioneering work first explored, from a language-philosophical perspective, the possibility of machines recognizing formal languages and reasoning, while envisioning robots with human-like intelligence. Subsequent AI research has primarily focused on machines’ linguistic capabilities, ultimately leading to today’s large language models. These models demonstrate remarkable linguistic proficiency, yet fundamental controversies persist regarding their status as genuine cognitive agents and their ethical accountability. From a language-philosophical viewpoint, whether AI can qualify as a cognitive subject depends not on its ability to simulate human speech, but on its capacity for embodied interaction within the practical “lifeworld” and participation in authentic language games. Only through a paradigm shift from disembodiment to embodiment can AI evolve from an instrumental existence to a form of “co-being” as subjective entity.

Understanding Without Assuming Mental States: A Behavior-Based Framework for Animal Death Cognition 

Qiushi Huang

Can animals understand death? Susana Monsó (2021, 2022, 2024) proposes animals possess a “minimal death concept.” However, concept-centered approaches face methodological difficulties: animals are “clever black boxes” (Halina, 2024), and attributing human-like mental states risks anthropomorphism. Moreover, such approaches prematurely commit to underlying mechanisms (concepts as mental structures) before establishing phenomenal-level causal regularities, violating proper explanatory ordering between etiological and constitutive explanations (Siegel & Craver, 2024).

I propose an alternative behavior-based framework drawing on Woodward’s interventionism (2003) and Bayesian probabilistic models. Rather than inferring hidden concepts, I argue behavior itself embodies understanding. By identifying counterfactual dependence relations between death-related environmental variables and systematic behavioral responses, we can explain animal death cognition without invoking unverifiable mental representations. This framework avoids anthropomorphism, applies across species with different neural structures, and provides scientifically rigorous explanation through testable predictions rather than unobservable psychological commitments.

Deontic Modality, Free Choice, and Linear Logic

Jixian Yu

This paper studies Barker’s resource-sensitive account of free choice (FC) inference under linear logic frameworks in [2]. We first present Barker’s account and some fragments of linear logic. Then we propose: Barker’s conditional variation of deontic modal sentence is inconsistent with the resource-sensitive use of ⊸ in general linear logic framework. Since the generality of δ eliminates the exact amount of resources required for the action, which makes the resource- sensitive use of ⊸ inconsistent with the conditional variation of deontic sentence. We then discuss possible solutions to the problem in contrast to Aloni’s BSML and propose that both Barker’s and Aloni’s accounts emphasize the constructive nature of deontic modality of FC phenomenon.

How Generative Music AI Performs Gender

Minyeong Bae

This study explores how generative music artificial intelligence (AI) performs gender through auditory and musical elements, focusing on the widely used music generation model ‘Suno’. It examines the system’s user interface (UI) that requires the selection of vocal gender, the gendered voices and associated musical features produced by the model, and the generated lyrics. By exploring user–AI interactions, the study analyzes how a gendered, fictional vocal subject is constructed through this process. Rather than treating generative music AI as a tool for mere imitation or representation, this study conceptualizes it as a digital–technical performative apparatus through which gender is enacted in sound and music, highlighting new forms of technologically mediated gender construction.

Embodying Chronic Shame: A Feminist Enactivist Approach

Valencia White, Anthony Chemero

This paper adopts a feminist, non-individualist philosophy of mind to argue that chronic shame arises from distorted social interactions and manifests in one’s embodiment. Examining its functions, we detail how chronic shame becomes an oppressive form of embodiment in women, drawing on feminist scholarship to highlight shaming social pressures targeting female bodies and behavior. We outline gender as an embodied practice sustained through interactions with others. Utilizing enactivist cognitive science, we demonstrate how identities are co-constructed through environmental interactions. Subsequently, we analyze the perpetuation of chronic shame as an embodied behavior upheld by interactions that preserve gendered power imbalances. Ultimately, chronic shame constitutes an interactional asymmetry, demanding women sacrifice their sense of self and affective self-regulation to foreground others’ embodied needs and expectations. Finally, we argue that embodied resistance, with a focus on self-regulation in social interactions, can be a tool in mitigating chronic shame.

Seeing What to Do: An Imperativist Account of Affordance Perception

Yukun Chen

I argue that affordance perception has directive rather than descriptive experiential content. It’s commonly assumed that perceiving affordances (re)presents the environment as affording an actions (e.g. “the stairs are climbable”). Against this, I defend an imperativism, on which affordances are experienced in an imperative format (“DO φ!”), presenting the environment as calling for action. For soliciting affordances (i.e. those exerting a felt “pull” toward action) imperativism is parsimonious and phenomenologically adequate. It avoids positing an unmotivated transition from descriptive contents to directive states and better reflects the overall structure of the solicitation. This account finds support in neurological evidence. The dorsal stream’s AIP-F5 network automatically translates perceived object properties into motor commands, with F5 neurons encoding action goals rather than objects. Medial premotor areas modulate these directives through inhibition, prioritization and planning, which is an architecture naturally understood as operating over directive contents that can be suppressed or weighted.

Caring-Counseling: Zhu Xi’s Learning Theory and Its Moral Implications for Alleviating Learned Helplessness 

Fengyuan Wang

Learned helplessness, a psychological condition arising from repeated exposure to uncontrollable adversities, manifests as motivational and cognitive impairments and is often implicated in academic misconduct. While conventional interventions such as attributional retraining and cognitive-behavioral therapy present limitations, the Neo-Confucian learning theory of Zhu Xi provides a culturally resonant alternative. His framework prioritizes: (1) “learning for the self” (wei ji zhi xue) to reconstruct agential subjectivity amid external pressures; (2) “dialogical engagement” as a pedagogical practice to articulate learning obstacles and cultivate collaborative support networks; and (3) disciplined methods of reading and self-cultivation. Crucially, Zhu Xi integrates moral reflection into learning process, endowing his approach with ethical dimensions that embody relational care and self-care. Consequently, his theoretical system serves as a resource for philosophical counseling, offering insights from the Chinese intellectual tradition to inform contemporary psychological strategies for addressing learned helplessness.

Epistemic Innocence and the Episodic-Semantic Distinction

Jethro Daryll Pugal

In Epistemic innocence and the production of false memory beliefs, Katherine Puddifoot and Lisa Bortolotti view three broad categorizations of false memory beliefs, and argue that they must be assessed as epistemically innocent, as opposed to the intuitive view that they are epistemically harmful. While the paper remains neutral on the type of memory, most of their examples are better described as both episodic and semantic memories, referring to Endel Tulving’s distinction. I propose to improve on this view by looking closer at Tulving’s distinction, and mapping them out to Kourken Michaelian’s metacognitive position where I treat semantic memory as the belief-endorsing mechanism, while episodic memory serves the function of information-production. I believe that this account offers a more complete view of the epistemic innocence of cognitive mechanisms, particularly in memory, and contributes to the overall epistemology of memory.

Seminar Room 9: General Philosophy of Science

Scientific Testimony Beyond the Lab: Simplification, Hype, and the Limits of Justification Reporting 

Axel Gelfert

This paper examines science communication as a philosophical problem arising from the circulation of scientific claims beyond specialist communities in contexts marked by selective distrust in science. It focuses on selective uptake, whereby lay audiences reject particular public scientific testimony despite broadly accepting scientific authority, and argues that traditional consensus reporting is often ineffective under these conditions. The paper assesses the recently proposed alternative of Justification Reporting, which demands that scientific justification be made explicit in public reporting. While philosophically well-motivated and aligned with core features of scientific practice, Justification Reporting faces practical limitations when adapted for non-specialist audiences. These limitations are explored through a case study of media coverage of experimental plant physiology research on stress induced airborne sound emissions, widely framed through anthropomorphic metaphors. The paper argues that such oversimplification undermines epistemic warrant, challenges assumptions about scientific testimony, and risks assimilating Justification Reporting to consensus reporting.

Epistemic trust injustice in public health and how to ameliorate it

Elena Popa

Trust plays a key role in public health. The success of guidelines and recommendations requires a trusting relation between scientists, policymakers, practitioners, and members of the public. This makes warranted distrust especially among groups that have suffered oppression, but also more broadly in contexts where the relevant institutions involved are not sufficiently trustworthy, a problem. The term ‘epistemic trust injustice’ refers to cases where the conditions that ground public trust in experts, or enable public access to scientific knowledge, are not met, especially for members of oppressed groups. This can bring about additional harm, as it can limit access to needed health services or relevant medical knowledge. In this paper, I will explore several amelioration strategies, starting from increased attention to population-level approaches in public health, which are more likely to target the needs of marginalized groups, structural competence among healthcare professionals, and participation from the relevant stakeholders.

Bias and the Base Rate: revisiting ‘Why most published research findings are false’

Kit Patrick

In his widely cited 2005 paper, Ioannidis argued that bias makes most published findings false. He attempted to establish this not on the basis of empirical studies but rather from a formal model of scientific research. The paper became a modern classic and remains influential in academia and beyond. However, we have been misreading Ioannidis’s formal model. For it actually shows that bias cannot explain most falsehoods. And, once we clearly understand the role bias plays in Ioannidis’s model, we can see that there are other sorts of bias it excludes that are more prevalent, and which counteract the second-rate effects of the bias he does consider. Moreover, empirical evidence gathered in the years since 2005 indicate that bias in Ioannidis’s sense isn’t prevalent enough for his argument to succeed. So, in fields of science where we find an over-abundance of falsehoods, we must look for alternative explanations.

Reliability-Preserving Pivots and Epistemically Justified Pivot Penalties

Anish Seal

A recent Nature article reports the “pivot penalty in research” (Hill et al., 2025): a general pattern whereby work researchers produce after professional (research) pivots tends to garner significantly less impact than pre-pivot work. This presentation abstracts from the extent to which actual cases of reduced impact have robust epistemological significance. Instead, it focuses on the normative modelling task of discerning when reduced impact, once understood as evidence that a researcher’s post-pivot work is less reliable than their pre-pivot work, has epistemic justification. In particular, I focus on pivots where methods considered reliable in pre-pivot domains are deployed in post-pivot domains with respect to which they are novel. I propose three jointly non-exhaustive, domain-general conditions tracking the extent to which such methodological adaptation respects standards of reliability in both the pre- and post-pivot domains. I conclude by noting some implications for the conservatism-versus-creativity debate in the social epistemology of science.

Reframing Simplicity: Defending Its Truth-Conduciveness with a Two-Step Model of IBE

Chenyang Lu

Many try to fit Inference to the Best Explanation (IBE) into Bayesianism by treating explanatory virtues like simplicity as properties raising an explanation’s probability. However, this approach cannot explain why auxiliary hypotheses would ever be needed, precisely what simplicity is often measured by. Nor can simplicity straightforwardly improve explanatory goodness: the conditions under which simpler explanations score higher on explanatory goodness would screen off any independent role for simplicity in producing that increase. Hence probability and explanatory goodness should be integrated systematically rather than treated in isolation. Instead of integrating them into a single function, I propose a two-step model of IBE: first, filter for “good enough” explanations using explanatory goodness; second, evaluate the remaining candidates by probability and form beliefs accordingly. This model is compatible with Bayesianism, preserves a monotonic link between simplicity and probability, and better matches IBE in practice. Accordingly, truth-conduciveness means benefiting explanations in either step.

Higher-Level Causal Explanations, Spurious Specificity, and Model-Based Proportionality

Jiaming Zhan

List and Menzies (2009, 2010) argue for the autonomy of higher-level causal explanations by appealing to proportionality—the requirement that a cause be at the right degree of specificity, neither too abstract nor too detailed. They argue that a proportional cause, identified through a counterfactual test, can exclude both more abstract and more detailed candidates, thereby securing its autonomy. However, their account faces the spurious specificity challenge—the problem that conjunctive candidates artificially satisfy proportionality and threaten to displace non-conjunctive causes that are frequently used in ordinary causal reasoning. To address this, I propose model-based proportionality. Within a well-formed causal model, structural rules and model-specific evaluative information filter out illegitimate candidates before applying proportionality. This approach not only resolves spurious specificity in a principled way but also preserves the autonomy and practical utility of ordinary higher-level explanations.

Mechanistic Explanations in History: A Case Study 

Yifan Li

This paper examines Ellen Meiksins Wood’s explanation of capitalism through the lens of the mechanistic approach to explanation. The paper argues that her explanation is recognizably mechanistic yet differs significantly from the paradigmatic mechanistic explanations in the literature. Her explanation is mechanistic in that its about how the system-level tendency came out from the interactions between the components of the system, i.e., landlords and farmers. It is distinctive in two ways:  first, besides the usual (de-)compositional structure of mechanistic explanations, it contains a genealogical element beside showing how a society became capitalist. Moreover, the explanatory strategy of the genealogical element involves an interesitng positive feedback loop structure. Second, it also aims at showing the compulsiveness of capilism by accounting for the robustness and causal power of the tendency by pointing out features of the agent-level interactions. This demostrates how mechanistic explanations can be adapted to satisfy diverse explanatory goals.

Evaluating understanding without its content: a process-based account 

Congjia Zhou

What is understanding? Candidates include knowledge, epistemic achievement, intelligibility, grasping, a cluster of abilities, etc. Abilities may include answering related questions, using theories properly, being able to give an explanation, etc. In this paper, I will give a critical review of the above accounts, in order to argue: (1) to understand is to grasp, and (2) understanding consists of the grasping process plus the result of grasping. That means to evaluate one’s understanding, one needs to not only check the result of grasping but also justify the process. The result of grasping concerns the process per se, not its content or object. To do that, I argue that understanding is not knowledge or anything that looks like a propositional attitude. Unlike those ability accounts and hybrid accounts, mine is mainly normative and can draw a clear line between explanation and understanding.

Epistemic Infrastructure and the Mechanization of Proof: Neuro-Symbolic AI, Formal Verification, and the Politics of Mathematical Knowledge

Fan Yang

Neuro-symbolic AI systems are reconfiguring what “proof” means in contemporary mathematical practice. This paper argues that formal verification should be treated not as a binary property of a proof text, but as an infrastructural achievement sustained by socio-technical dependencies (standards, libraries, maintenance, and governance). Building on bi-level epistemology, it distinguishes an infallibilist first-level justification delivered by proof-assistant kernels from a fallible second-level justification grounded in human-managed infrastructure and semantic interpretation. A three-layer L1–L3 framework (kernel verification; heuristic search; socio-technical embedding) clarifies why neuro-symbolic provers suppress step-level errors while concentrating risk at the translation interface between informal intent and formal specification. The paper concludes with normative proposals—provenance-first artifacts, decoupled evaluation of translation vs. reasoning, human-in-the-loop interaction, and recognition of library maintenance as epistemic labor—to sustain trustworthy “industrialized proof.”

Probabilities of counterfactuals and imaging 

Jingzhi Fang, Jiji Zhang

The truth or assertability of a conditional is often linked to its corresponding conditional probability, a principle known as Adams’ thesis. According to Stalnaker’s hypothesis, the probability of a counterfactual is equivalent to the conditional probability of the consequent given the antecedent. However, this equivalence leads to a triviality result shown by Lewis and is recovered when  conditioning is replaced by a more general probability revision operation known as “imaging”. The equivalence between probabilities of counterfactuals under Lewis’ semantics and the corresponding imaging probabilities does not hold in general. This work identifies the specific conditions under which this equivalence can be restored by refining Lewis’ semantics or specifying the imaging operation. Building on this connection, we propose a probabilistic semantics for counterfactuals, grounded in the updated probabilities generated by imaging.

Covid-19 in Retrospect: Reassessing the Role of Modeling in Policy-Making

Arnon Levy

The role of scientists in shaping policy is philosophically interesting and has concrete socio-political significance. But while policy-relevant science, as such, has received significant attention, there isn’t much discussion of the roles of different sorts of science and scientists vis-à-vis policy. In this paper I want to take a few initial steps on this score, focusing on models and modelers, especially during the Covid-19 crisis. I begin by reviewing some work on Covid-19 science-for-policy – by Birch, Winsberg and Harvard and Broadbent and Streicher. Against this background, I argue, first, that responsibility for both the process and outcomes of policy-making should rest chiefly with political leaders. Second, that political leaders must be better informed about the role and limitations of models. Finally, that a fundamental lesson of the pandemic is the need to base policy-making, perhaps in a crisis, on a range of modeling tools and approaches.

Seminar Room 4: Philosophy of the Physical Sciences: Quantum Foundations

Violating Statistical Independence: the problem of Fundamentality

Gabriele Cafiero, Luca Molinari, Jonte Hance

In this work we approach the issue of theories that introduce a violation of statistical independence as a means of circumventing Bell’s theorems, highlighting that some of them have the character of a more metaphysical than scientific proposition. After a review of such proposals, we present a distinction between two types of superdeterministic theories, naive (NSD) and metaphysical (MSD) ones, and show how NSD presents significant epistemic flaws. We  illustrate how NSD justifies itself through claims to fundamentality, thus connoting itself as a metaphysical proposal rather than a scientific one. We finally show that the most developed MSD model so far, Invariant Set Theory, implicitly proposes a confused form of priority monism. Our paper thus reinforces the thesis that theories should demonstrate rather than presuppose their fundamentality and that it is methodologically flawed for them to assume fundamentality with the sole purpose of defending against criticisms.

Phenomenology and Quantum Physics: Cutting the Chain of Correlations?

C. D. McCoy

Steven French’s recent book /A Phenomenological Approach to Quantum Mechanics/ (OUP 2023) elaborates what he characterizes as a phenomenological approach to understanding measurement in quantum mechanics, where “phenomenological” here refers not to the context of experiment, as it usually does in physics, but rather to the philosophical program of Edmund Husserl. Husserl’s transcendental phenomenology, developed roughly in the first half of the 20th century, aims to disclose universal, essential structures of consciousness through a presuppositionless inquiry into conscious experience. In this paper, I argue that French’s proposal is not based on a correct understanding of Husserl’s views or phenomenology more generally, and furthermore that the problem of measurement addressed by this view is not even unique to quantum mechanics but instead arises also in classical statistical theories, such as classical statistical mechanics.

Algebraic Quantum Gunk

Isaac Wilhelm

When arguing over the structure of physical reality, metaphysicians often appeal to Newtonian mechanics, non-relativistic quantum mechanics, and other theories fairly far removed from the quantum field theories studied in physics today. This paper shows what metaphysicians stand to gain from focusing on more up-to-date theories, like quantum field theories, instead. Specifically, I show that certain algebraic quantum field theories (AQFTs) support an interesting argument for mereological gunk. 

The paper proceeds as follows. To start, I summarize the AQFT framework which will be relevant here, and I explain the important role which projections play in that framework. Then I propose an analysis of mereological parthood among the physical structures which projections represent. Finally, by combining that analysis with well-known results that physically reasonable AQFTs generically involve type III von Neumann algebras (factors)—algebras with no nonzero minimal projections—I formulate an argument for mereological gunk.

Seminar Room 6: Formal Philosophy

The Disastrous Implications of the Argument from a Bad Lot

Seungbae Park

The argument from a bad lot (ABL) has three disastrous implications on the pessimistic induction (PI). The first one is that the conclusion of PI requires the disprivilege principle that most current theories are more likely to be picked out from bad lots than from good lots. The second one is that the inference from the premise to the conclusion of PI requires the privilege principle that the conclusion is more likely to be picked out from a good lot than from a bad lot. The third one is that ABL makes it more likely that PI is infinitely problematic.

Proof, Computer and Mathematical Agent
Min Cheol Seo

This paper critically examines Hamami and Morris’s plan-based account of proof and proof understanding, arguing that its core thesis—that a written proof is an isomorphic report of a rational plan—is fundamentally challenged by machine-assisted proofs exhibiting strategic autonomy, most notably, McCune’s automated proof of the Robbins Conjecture. Further, I demonstrate that the “as-if” stipulation introduced to address understanding of such cases destabilises the theory, divorcing the text from the prover’s planning rationale and reducing understanding to post hoc rationalisation. To address this tension, I propose a framework called, Stratified Intentionality: a four-layer architecture disaggregating proof construction into Theorem-level, Strategic, Tactical, and Operational commitments. This hierarchical model accommodates the distributed nature of modern mathematics, where high-level strategies may be human while operational layers are algorithmically generated. Understanding, I argue, consists in recognising Intentional Realisation, where lower-level operations fulfill higher-level strategic constraints.

When the Practice Doesn’t Settle Philosophy

Chanwoo Lee

The practice-oriented approach to philosophy of science heeds the scientific practice, but the practice is not always sufficient for deciding the relevant philosophical questions. To complement this approach, I suggest a further methodological norm that, given two options equally well-supported by the scientific practice, we should prefer an option that is more philosophically neutral. I call this the ‘Neutrality Norm’. As case studies, I consider two debates on set-theoretic practices and illustrate how the Neutrality Norm can inform them. Also, I provide two independent arguments for the Neutrality Norm, arguing that it aligns with the spirit of the practice-oriented approach and enjoys support from general epistemological considerations.

Seminar Room 7: Philosophy of the Cognitive Sciences: Predictive Processing and Active Inference

Integrating Phenomenological Insights into Predictive Processing: The Case of The Paradox of Pain

Junjie Huang

Pain poses a long-standing paradox: it is epistemologically accessible only through first-person experience, yet it seemingly demands an objective, physicalist explanation. This paper addresses this tension by integrating predictive processing theory with phenomenological analysis. It argues that pain is not generated by nociceptive input alone, but by the precision-weighted prioritization of bodily threat signals within a hierarchical predictive system. Phenomenological accounts constrain this model by specifying the experiential structures that any adequate explanation must capture, while predictive processing provides a generative mechanism for those structures. The paper proposes a non-reductionist interpretation of pain, according to which its unity lies in its functional role within a biologically and socially shaped predictive hierarchy. The so-called pain paradox is thus re-interpreted as a failure of alignment between explanatory perspectives, rather than an ontological dilemma.

Reclaiming Realism for Models of the Free Energy Principle: A Perspectivist Account

Yihan Jiang, Yuyou Wu

The free energy principle (FEP) has emerged as a unifying framework in biology and cognitive science, yet realist interpretations of the active inference models remain contested. A growing instrumentalist consensus holds that, because these models rely on heavy idealization and permit explanatory pluralism, they cannot be veridical descriptions of cognitive architecture. In this talk, we defend a realist interpretation by arguing that these critiques presuppose an outdated “God’s-eye” conception of scientific realism. Drawing on perspectival realism (Massimi 2022), we show that models need not offer literal, isomorphic depictions of their targets to be truth-apt; rather, idealization and pluralism are compatible with capturing mind-independent reality from specific scientific vantage points. We further supplement this account with the real pattern theory, arguing that FEP models are realist insofar as they track objective, large-scale patterns in neuronal dynamics at particular scales of resolution.

Naturalizing Sosa’s Virtue Epistemology: An Active Inference Perspective

Stephanie Chng

This paper draws on cognitive science— specifically, Karl Friston’s active inference framework as applied to cognition— to examine Ernest Sosa’s framework of virtue epistemology. I argue that Sosa’s framework is naturalistically implausible and lacks naturalistic utility; with both ultimately undermining his attempt to properly naturalize epistemology. The central problem of Sosa’s framework lies in his empirically tenuous, and normatively unhelpful, distinction between animal and reflective knowledge. First, it is inconsistent with predictive-processing theories about how the brain works, under which first- and second-order knowledge are governed by the same underlying inferential mechanism. Second, Sosa’s characterization of reflective knowledge as requiring coherence-based internalism introduces first-person perspective considerations that elude the scientific standards demanded by a fully-fledged naturalism. I argue that Sosa’s problem stems from an inverted methodological strategy, in which insights from cognitive science are treated as a subsidiary consideration, rather than an ex ante design constraint, on his epistemological framework.

Seminar Room 8: Philosophy of Technology and AI: Meaning, Automation, and Discovery

Materializing Meaningful Technology

Farooq Alvi

Until when will we rely on technology producers to balance the benefits of technology with its ethical considerations? This approach has not only failed to deliver meaningful change, it overlooks how technology’s impact is shaped by systemic adopters, organizations that adopt it internally and translate it into products. I argue that meaningful impact on technology’s uneven benefits and harms requires intervention where systemic adopters operationalize it by critically engaging with the assumptions shaping technology, rather than relying solely on producers, whose limited engagement with these assumptions leaves outcomes largely unchanged. I illustrate the approach with an organizational technology strategy framework for interrogating technology in context, using applied examples from leading contemporary cases. This reframes the challenge from accepting uneven outcomes and inaction to a practical orientation toward resolution.

Staying Human: Meaning in Life in the Age of Abundant Intelligence

Minji Jang, Ilia Tsetlin

Advances in artificial intelligence raise the redundancy threat: once superefficient AIs produce knowledge, art, and valued outcomes faster and better than humans can, human engagement may seem pointless. Drawing on Wolf’s (2010) hybrid account, on which meaning requires subjective fulfillment in objectively valuable activities, we distinguish a pursuit’s contributory value (non-redundant contribution to valuable outcomes) from participatory value (a person’s own engagement) and argue that AI may threaten the former without eliminating the latter. This motivates a revised hybrid account that insulates many pursuits from the threat: a person’s relationship to a project is not merely a source of subjective satisfaction but part of what makes the pursuit objectively valuable when it consists in the exercise of self-authorship. AI may further expand opportunities for meaning by redistributing what Rawls (1999) calls “the social bases of self-respect,” provided that it scaffolds rather than substitutes for self-authorship and institutions distribute access broadly.

Artificial Intelligence and the Problem of Unexplainable Discovery

Tarun Menon and Sumeet Agarwal

There has been considerable debate about the extent to which we can automate scientific explanation with AI. Often, humans cannot understand how an AI model arrives at its outputs, which seems inconsistent with the explanatory project of science. One response to this concern is that building AI models is part of the context of discovery, while explanation and understanding belong to the context of justification.

We argue for two reasons why the context of discovery cannot be separated from the goal of constructing scientific explanations: because explanation is typically not a property of just one model in isolation, but of a larger theoretical framework; and because human-like constraints on the space of models can be essential for counterfactual robustness.

We suggest that better AI alignment with human inductive biases may be necessary to align prediction and explanation, and consider how to design AI systems that work within these constraints.

Seminar Room 9: General Philosophy of Science: Values, Responsibility, and Trust

Epistemic Projection and Value-Free Science: A Failed Reconciliation

Eunhyeong Kim

This paper examines Wendy Parker’s Epistemic Projection Approach (EPA), an attempt to reconcile proponents and opponents of the value-free ideal in science. I argue that EPA’s proposed reconciliation does not succeed and explore broader implications of this failure. I first maintain that EPA cannot satisfy traditional supporters of the value-free ideal, which creates pressure to identify the moderate supporters Parker aims to satisfy. I then explore whether a pluralist approach that distinguishes multiple formulations of the value-free ideal can specify such moderate positions and thereby rescue EPA’s reconciliation project. I argue that this pluralist approach likewise fails by identifying the most plausible moderate positions within this framework and demonstrating that even these positions permit precisely the problematic value influences that the value-free ideal was designed to prevent. Finally, I suggest that EPA’s failure invites us to reconsider what counts as “core science” worthy of protection from values.

Trust in Experts, Values in Science, and Risk Preferences 

Tomasz Żuradzki

Debates about trust in science increasingly recognize that values play substantive roles within scientific inquiry, particularly in setting evidential thresholds for accepting hypotheses or issuing recommendations. This insight is central to the argument from inductive risk, according to which empirical evidence often underdetermines acceptance or rejection, requiring scientists to weigh the consequences of false positives and false negatives. Acknowledging this value-ladenness has prompted proposals emphasizing transparency, shared methodological standards, alignment between scientific and public values, or the democratic grounding of scientific decision-making.

This paper argues that these debates overlook a further and distinct dimension of scientific judgment: scientists’ risk preferences. I argue that such rationally permissible differences pose an additional challenge for accounts of public trust in science that focus exclusively on values, and that existing approaches must be revised to accommodate the role of risk preferences in expert judgment.