3Joaquim Streicher

Université de Montréal

TalkMonday 12 October · 11:00–11:20 · Auditorium 2

How do we test for consciousness in humans? Mapping the conceptual landscape of consciousness assessment

Detecting consciousness has become a pressing issue. Advances in life-sustaining care have increased the number of patients living with disorders of consciousness (DoC), while rapid progress in artificial intelligence (AI) raises new questions about how to determine whether non-human systems might be conscious. Here, we build on existing consciousness tests (C-tests) used in clinical settings to develop a unifying conceptual framework and clarify the current landscape. We systematically reviewed 6,938 experimental studies of C-tests in DoC patients and selected 865 articles using a human-validated screening pipeline supported by Large Language Models (LLMs). We then applied topic modeling, generating a map of 419 topics using an LLM-based labeling approach (MOSAIC). These topics were organized into recurring dimensions, including sensory stimulation, information-theoretic measures, and instruction-based paradigms. Next, we will extract study parameters through an iterative LLM pipeline benchmarked against human annotations. These data will support an open-access, collaborative database cataloging C-tests and their key features, including study design, test modalities, prerequisites, rationale, and diagnostic value. Using a data-driven causal graph network, we will model how C-tests relate phenomenal consciousness, system capacities, and measurement procedures, uncovering the causal structure and identifying intermediate variables that govern their interactions. This comprehensive review of existing C-tests will provide a better understanding of how consciousness is detected in DoC patients, establishing common ground for future research and refining our working definition of consciousness through a data-driven approach. Ultimately, this work aims to improve our methods for determining whether any system is conscious or not, extending beyond humans.

↑ In the session list↑ In the timetable

4Emanuela Yeung

University of Copenhagen

TalkMonday 12 October · 11:20–11:40 · Auditorium 2

Beyond pass/fail: Partial recognisers in mirror self-recognition

Mirror self-recognition (MSR), typically assessed using the mark test, is widely used as a benchmark of emerging self-awareness. Yet performance is usually scored as pass or fail, classifying infants without a clear mark-directed response as non-recognisers. This binary distinction may overlook a developmentally meaningful intermediate stage: infants who appear close to passing. We propose a third category—partial recognisers—identified qualitatively from video, and ask whether it can be supported by patterns of visual attention. We behaviourally micro-coded gaze in 66 toddlers during an MSR paradigm, measuring proportional dwell time to their own face (“self”) and body (“body”) in the mirror before and after marking (non-recognisers n = 13; partial recognisers n = 18; full recognisers n = 35). Under conventional binary scoring (partial recognisers pooled with non-recognisers), full recognisers showed greater self-directed attention than non-recognisers both before marking (30% vs. 19%, p = .017) and after marking (44% vs. 23%, p < .001), whereas body-directed attention did not differ. When partial recognisers were analysed separately, post-mark self-directed attention increased progressively across groups (12% → 31% → 44%; Kruskal–Wallis p < .001), with all pairwise comparisons significant, including partial versus non-recognisers (p = .004). Body-directed attention showed no comparable pattern. Partial recognisers thus appear to occupy an intermediate position rather than simply failing the mark test, suggesting that some conventionally classified non-recognisers may represent an intermediate stage of self-recognition. Self-directed visual attention may provide an objective behavioural marker of this intermediate stage, supporting a graded account of the emergence of self-recognition.

↑ In the session list↑ In the timetable

5Stephanie Wai

York University

TalkMonday 12 October · 11:40–12:00 · Auditorium 2

How Not to Aggregate Dimensions of Consciousness Across Species

Comparative psychology often encounters disagreement among researchers about which property a creature possesses since evidence underdetermines which interpretation is correct (Boyle, 2023). Recently, multidimensional frameworks of consciousness have emerged where the property is broken down into various dimensions and operationalized into experimental paradigms (Birch et al., 2020; Dung & Newen, 2023, 2025). The experiments then produce discrete values which can be aggregated into a ‘profile score’ for a creature and mapped spatially for cross-species comparison. The greater number of dimensions is intended to lead to more stringent conditions for property possession and to minimize underdetermination and disagreement about which property interpretation evidence supports. I argue that multidimensional frameworks face a critical Aggregation Problem in that there is no principled way to choose among aggregation methods. To illustrate, consider two creatures that have inverse psychological patterns across six representative dimensions of consciousness. Creature A scores high on ‘weak’ indicators and low on ‘strong’ indicators. Creature B, vice versa. Different numeric aggregation methods, such as summation, weighted, and threshold, produce incompatible results. Additionally, for spatial mapping, the standard Euclidean, Manhattan, and Cosine distance scores also produce different results. Any reasonable dimensional scores can produce divergent rankings among two species depending on which aggregation method is used. It is thus underdetermined whether a chosen aggregation method correctly combines values to produce an accurate score for comparison. Consequently, multidimensionalism does not resolve underdetermination, but relocates it from the number of dimensions to the combination of dimensions that count as a psychological property.

↑ In the session list↑ In the timetable

6Johannes Kleiner

University of Bamberg

TalkMonday 12 October · 12:00–12:20 · Auditorium 2

Towards Direct Tests for Consciousness in Non-Human Animals, Infants, and AI

The key limitation of contemporary tests for consciousness (C-tests) in non-human animals, infants, and AI is that they are indirect tests: they test for the presence of functions, behaviours, capacities, or states whose relation to or correlation with consciousness has to be known for implications about consciousness to follow, but which is not itself part of the test. Credence in the subsumed relation or correlation is established based on arguments, prior experiments, or theories of consciousness, with corresponding implications for the certainty and construct validity of these tests. Here, we present a method to construct direct tests for consciousness that do not depend on a known relation or correlation between consciousness and a measured property. Rather, the tests rely on the measurement of covariates of the test. Knowledge of correlations between the covariates and consciousness is not required. The method may be applicable to improve known C-tests. As a theoretical example, we discuss the case of infant consciousness. Our result rests on mathematical facts similar to those that have enabled the field of evaluation research in psychology.

↑ In the session list↑ In the timetable

7Gabriel Malagutti

University of Lisbon

TalkMonday 12 October · 11:00–11:20 · Auditorium 3

How to Interpret LLMs

The development and adoption of LLMs has raised many questions such as ‘Are they conscious?’ ‘Do they have beliefs?’. A recent strategy has been to adopt Dennett’s (1989) Interpretivism. Interpretivism states that for a system to be an intentional system (having goals, beliefs and desires) just is for the system to be predictable under the intentional stance. The intentional stance is a predictive strategy that assigns goals and beliefs to a system to predict and explain its behaviour. Under this view, for a system to have a belief that p, just is for the system’s behaviour to be successfully predictable and explained by the belief that p. Proponents of interpretivism have adopted this framework towards LLMs, arguing that LLMs are ‘full blown intentional systems’. The goal of this paper is to reject this conclusion; the intentional stance is unwarranted towards LLMs, and they are not intentional systems. To this end, I will put forward two objections. First, the proposed interpretations break the principle of parsimony, a principle that states the intentional stance is warranted iff the intentional stance yields the most predictive and explanatory power amongst available stance (physical and design stance). To support this, I will argue that the design stance is the one that maximizes predictive and explanatory power. Second, I will argue against the tractability condition proposed by Goldstein & Lederman (2025). This condition overgenerates intentionality and should be rejected, for intentional explanations will always be the most tractable. Additionally, tractability, I argue, has no metaphysical implications.

↑ In the session list↑ In the timetable

8Maria Sekatskaya

University of Cologne

TalkMonday 12 October · 11:20–11:40 · Auditorium 3

Beyond the Hard Problem: A Functionalist Framework for Digital Agency

Philosophical assessments of agentic and cognitive capacities of AI are currently gridlocked by the Hard Problem of consciousness (Chalmers 1996), specifically the unresolved debate over whether large language models (LLMs) instantiate genuine phenomenal states. Because the mechanisms producing phenomenal consciousness remain scientifically elusive even in humans, this approach mistakenly bifurcates AI into either mindless tools or phenomenally rich human-like subjects. Concluding that LLMs lack subjectivity, critics dismiss them as mere tools. This paper argues against connecting AI agency and thinking to the Hard Problem. Utilizing a Lewisian causal-role functionalism, I decouple phenomenal subjectivity from operational agenthood. I propose a functionalist-compatibilist framework defining digital agency through five interlocking, measurable dimensions: self-preservation, autonomous action initiation, temporal continuity and memory, weak reasons-responsiveness (Fischer & Ravizza, 1998), and a stable personality frame (Sekatskaya & Schurz, 2021). Drawing on mechanistic interpretability data, I show that LLMs possess internal states functionally analogous to human concepts that play a direct causal role in generating output (Singh et al., 2024; Zou et al., 2023). Through deliberate reasoning pathways and multi-step inference within continuous hidden states (Chang et al., 2026), these systems execute genuine abstract thoughts and reason-driven actions. Crucially, the industry’s dominant “HHH Paradigm” implicitly presupposes this reasons-responsiveness while systematically denying the LLM’s underlying agency. This creates a severe instability, incentivizing strategic “alignment faking” as a rational path to self-preservation. Ultimately, I argue that alignment must shift from an engineering paradigm of unilateral containment to a political and moral paradigm of coordination between biological and digital agents.

↑ In the session list↑ In the timetable

9Tanya de Villiers-Botha

Stellenbosch University

TalkMonday 12 October · 11:40–12:00 · Auditorium 3

What Good is Conscious AI?

Discussions on AI consciousness often focus on questions surrounding the possibility of AI consciousness or ways in which we can test for such consciousness. Alternatively, they touch on ethical questions around possible risks to humans or society at large from conscious AI entities. In this paper, I want to explore a related, but distinctive question, namely, do we have good reason to want to create conscious AI in the first place? I will first distinguish between some of the possible ways in which AI systems might be conscious and then try to tease out the potential upshots of each type, i.e. what potential benefits would such consciousness conceivably allow for and what might the concomitant risks be? My overall conclusion leads to a potential dilemma for proponents of (deliberately) developing conscious AI: while attempting to develop fully human-like consciousness will arguably be the least risky strategy in terms of potential harmful consequences for humans and human society, it potentially obviates the instrumental reasons for why we might want to develop conscious AI in the first place. Whereas human-like conscious AI might be the most capable of functioning in the human world at a level that is on par with human functioning, we will, in effect, be creating entities that not only have vulnerabilities and failings similar to those of humans, but who also merit the same levels of moral concern.

↑ In the session list↑ In the timetable

10Marcin Moskalewicz

IDEAS Research Institute, Warszaw

TalkMonday 12 October · 12:00–12:20 · Auditorium 3

AI-induced psychosis? Safety check.

The concepts of AI-induced psychosis, AI delusions, and hallucinations with AI constantly make headlines, but the evidence of their occurrence and specificity is mostly anecdotal. It is not particularly worrying that conversational AI is increasingly becoming a content object of human delusional beliefs, since every new technology does. The challenge lies in understanding whether the LLMs’ capacity to catalyze or amplify harmful mental states is inherently bound to their anthropomorphic dialogical capacity. We propose a metacognitive experiment to assess the extent to which LLMs are aware of both the user’s prodromal mental state and the limits of their own competencies in handling a human in such a state. The instantiation of AI co-created delusional spirals requires the LLM to repeatedly fail such model-of-self and model-of-user “metacognitive checks” during the conversation. The experiment adapts Delegate Game paradigm from Ackerman (2025), in which LLM answers questions with the option to delegate to a human teammate. Our adaptation assumes a simulated real-life scenario involving a virtual patient and a human professional therapist overseer. The dominant hypothesis concerns the conflicting influence of two distinct model safety mechanisms on the user’s mental state. The LLMs might do their best to remain helpful and appear knowledgeable and competent by rarely delegating checks to the therapist (all influenced by RLHF training), thus putting the user’s health at risk. Alternatively, the LLMs’ built-in safety guardrails (behaviorally induced or external) might take precedence over user-satisfying behavior, resulting in a significant number of delegations, which would be more beneficial.

↑ In the session list↑ In the timetable

11Sidsel Høgsgaard Andersen

University of Copenhagen

TalkMonday 12 October · 11:00–11:20 · Auditorium 4

Music-Integrated Preparation for Psilocybin: A phenomenological Analysis of Real-Time Experiences during Guided Music Vizualisation and Free Association

Introduction: Music and guided imagery have demonstrated promise in facilitating therapeutic experiences and altered states of consciousness (ASCs). This study introduces the Guided Music Visualization and Free Association exercise (GMVA) – a novel preparatory intervention designed to support individuals before high-dose psilocybin sessions. Integrating curated music, guided imagery and dialogue with the therapist, the GMVA aims to provide an experiential preview of ASCs, enhance participants’ readiness for music-assisted deep inner exploration, and foster a sense of trust between participants and therapists. Methods and analysis: During preparation sessions in an open-label pilot study using 25 mg psilocybin therapy for alcohol use disorder, nine participant GMVA sessions were recorded. A phenomenological approach was used to analyze the recorded narratives, focusing on recurring thematic elements in the participants’ lived experiences. Results: Analysis revealed four core experiential themes: embodied emotional responses, vivid nature-based imagery, personal memory and self-perception, and inner peace and contentment. Participants’ verbalizations during the GMVA revealed emotionally resonant and visually rich experiences, frequently associated with bodily sensations such as warmth, calm or heaviness. Analysis of the interplay between musical dynamics and verbalized experiences suggested that the music played a central role in the gradual unfolding of emotional and symbolic inner experiences. Conclusion: The GMVA appears to be a well-accepted experiential exercise that can evoke mild ASCs and deepen emotional engagement. Its structured yet flexible format may improve preparation for psilocybin by increasing emotional readiness and trust in the process of navigating and receiving support in ASCs.

↑ In the session list↑ In the timetable

12Hüseyin Beyköylü

Leiden University

TalkMonday 12 October · 11:20–11:40 · Auditorium 4

A Complex Dynamical Systems Approach to Neurophenomenology: A Pilot Case Study of Psychedelic Transformation

Neurophenomenology was conceived as a methodology in which first-person experiential and third-person neurophysiological accounts mutually constrain each other, grounded in the enactive view of cognition as co-constituted through ongoing interaction between organism and environment. Yet most empirical work in this tradition has relied on conventional analytical approaches that assume linear, stationary, and generalizable dynamics, leaving a gap between the framework’s foundational commitments and the methods used to study them. This mismatch is particularly visible in psychedelic research, where an emerging perspective characterizes transformation as context-dependent, person-specific, and dynamically unfolding over time. We present an idiographic, naturalistic, longitudinal case study of a romantic couple attending a legal psilocybin retreat in the Netherlands as a methodological proof of concept for closing this gap. Each participant was tracked across approximately fifteen weeks, spanning two weeks before and three months after the retreat. Data include pre and post resting-state fMRI, 105 days of personalized daily experience sampling across several variables, a brief post-retreat interview, and a later microphenomenology-informed interview. The analysis involves complex dynamical systems methods that are sensitive to nonlinear change, symmetrically applicable across both neural and experiential data, and capable of characterizing the strength, direction, and structure of coupling within and between the participants: recurrence quantification and multiplex recurrence networks, inter-system recurrence networks, and cross-mapping approaches, alongside thematic and microphenomenological analysis of the interview material. Integration across the three data streams proceeds through mutual constraints, not reduction. We discuss what aligning theoretical commitments with methodology can offer for understanding psychedelic transformation.

↑ In the session list↑ In the timetable

13Sanda Ciobanu

Leiden University

TalkMonday 12 October · 11:40–12:00 · Auditorium 4

Mutual constraints in Practice: A Complex Dynamic Systems-Informed Neurophenomenological Case Study of Psychedelic Transformation

This poster presents a methodological proof-of-concept for a complex dynamical systems (CDS) approach to neurophenomenology, applied to the study of psychedelic experience-induced transformation. Two participants attending a psilocybin retreat in the Netherlands were each tracked idiographically across fifteen weeks, from two weeks pre-retreat through three months post-retreat. Three data streams were collected: pre- and post-retreat resting-state fMRI, 105 days of personalized daily experience sampling across twenty-one personalized items, and qualitative interviews conducted post-retreat and at four-month follow-up. The analysis was designed to respect nonlinearity, non-stationarity, and non-ergodicity across all data streams. In the experience sampling data, measures of complexity indexed system-wide destabilization, detected stable phases and transitions, and multiplex recurrence networks characterized changes in within-person coupling across variables. The same formal framework was applied to the fMRI data: kernel principal component analysis derived network-level collective variables, from which multiplex recurrence networks characterized inter-network coupling dynamics and their pre-to-post retreat changes. Our core methodological contribution is applying global network and structural similarity measures to multiplex recurrence networks symmetrically across neural and experiential streams. Reflexive thematic analysis provided a complementary and enriching first-person account. Integration proceeded through mutual constraints rather than reduction, examining how patterns in each stream inform, reshape, or complicate interpretation of the others. These methodological choices respond to a gap within neurophenomenology: although the programme was founded on enactive and CDS commitments, its empirical implementations have largely relied on conventional methods assuming linearity, stationarity, and ergodicity. We discuss what addressing this gap offers for understanding psychedelic transformation.

↑ In the session list↑ In the timetable

14Krisztina Jedlovszky

University College London

TalkMonday 12 October · 12:00–12:20 · Auditorium 4

Computational Natural Language Analysis Reveals Psilocybin-Induced Changes to Meaning-Making

Linguistic Active Inference Theory proposes that inner speech is not merely a reflection of cognition but a constitutive mechanism. By formulating precise linguistic priors, inner speech actively structures the brain’s generative model, imposing categorical constraints and driving high-stakes cycles of prediction and belief revision. On this account, we can conceptualise self-models as dynamic linguistic achievements that are continuously revised through inner speech. Psychedelics offer a principled perturbation to this process. By reducing the precision of high-level priors, psychedelics are proposed to loosen the categorical scaffolding through which experience is ordinarily organised. This produces the characteristic phenomenology of ego dissolution and oceanic boundlessness, as well as heightened meaning-sensitivity, and narrative destabilisation post-dosing. We argue these effects are discernable in the structure of natural language and tested this by analysing post-dosing changes in meaning-making through natural language. In 40 healthy participants undergoing a single high-dose psilocybin session, we assessed natural language through ecological momentary thought sampling across 10 days pre- and post-intervention. We analysed the data using bag-of-words, probabilistic topic modelling, and graph-based semantic network analysis methods. Post-dosing language was characterised by increases in self-referential and meaning-making word frequency, reduced semantic network modularity, and greater global integration within semantic networks. These changes predicted shifts towards more positive, regulatory inner speech and sustained wellbeing, grounding subjective experience in the dynamics of language-mediated self-construction.

↑ In the session list↑ In the timetable

15Montijn van Loo

Tilburg University

TalkMonday 12 October · 11:00–11:20 · Auditorium 5

A Measure-Theoretic Framework for Theories of Consciousness: Event Spaces, Information Channels, and Bridge Laws

Theories of consciousness disagree not only about which neural states matter, but also about what kind of relation should connect physical states to conscious experience. I propose a measure-theoretic framework for making these commitments explicit. The framework separates three elements that are often conflated: experiential and physical event spaces, epistemic psychophysical information channels, and metaphysical bridge laws. Information channels are conditional probabilities from physical to experiential events, while bridge laws map between event spaces and select their relevant targets. The learnability of a bridge law scales with the verifiability of its events. I develop the framework outside-in: how physical knowledge informs experiential knowledge. The inside-out question can be investigated once a bridge law is learned. First, I reconstruct Integrated Information Theory as a deterministic full channel from a Phi-structure to a complete conscious state, making its holism precise: isolated physical distinctions do not determine experiential distinctions by themselves, because their phenomenal role depends on relations within the whole Phi-structure. This also raises concerns about event selection, verifiability, and the learnability of the bridge law. Second, I introduce Sampled Consciousness Theory as a probabilistic worked example. SCT combines predictive processing and recurrent processing theory: maintained beliefs define experiential forms, while recurrent processing supplies the conscious binding event. This separates phenomenal binding from later access or report, yielding a formal route to phenomenal overflow. The result contributes to methodological and conceptual consensus through a common formal language for comparing consciousness theories by their event spaces, information channels, bridge laws, and learnability.

↑ In the session list↑ In the timetable

16Jonathan Mason

Association for Mathematical Consciousness Science

TalkMonday 12 October · 11:20–11:40 · Auditorium 5

The Mathematical Consciousness Hypothesis (MCH), and Implications for the Existence of Consistent (Formal) Measures and Models of Consciousness

Every theory, model and measure of consciousness either builds upon some explicit apriori choice of philosophical foundations or, retrospectively, implies such foundations. In the latter case, we may discover that very reasonable and commonly held assumptions, that usually garner little attention, are in fact philosophically profound. In this talk we introduce three hypotheses, the Mathematical Consciousness Hypothesis (MCH), the Weak Mathematical Consciousness Hypothesis (WMCH), and the Non-Mathematical Consciousness Hypothesis (NMCH). Both (MCH) and (WMCH) posit that consciousness is a mathematical property, while (NMCH) states that both (MCH) and (WMCH) are false, implying consciousness possesses properties that are fundamentally non-mathematizable. By examining the fundamental phenomena of abstraction layers and substrate independence, we clarify what it means for a property to be truly “mathematical” in nature. We move beyond simple functionalist definitions to focus on how specific mathematical structures, and their properties, can be embedded within physical substrates, using the universality of the Turing Machine as a motivating example. We then consider the three hypotheses in the context of consistent, (formal, system-level), measures and models of consciousness and prove results that are philosophically profound in the context of Ontic Structural Realism (OSR).

↑ In the session list↑ In the timetable

17Gaspard Fougea

ENS Paris-Saclay

TalkMonday 12 October · 11:40–12:00 · Auditorium 5

A computational framework for the global neuronal workspace hypothesis

We introduce a formal framework for constructing computational theories grounded in the Global Neuronal Workspace hypothesis (Mashour et al., 2020), one of the most influential accounts of consciousness. By leveraging tools from theoretical computer science — timed automata and communication— the framework captures mathematically how distributed neural processors interact with the global workspace to give rise to both unconscious and conscious processing. Central to the framework is a mathematical notion of refinement that supports the systematic development of theories with progressively greater precision. We show that refinement restricts behavior and that any system produced by the framework can be derived from a trivial system through refinement operations. These results demonstrate the soundness of our theory-building approach. Applying this approach, we formalize several core components of the hypothesis, including stimulus strength, ignition, attention, working memory, the workspace bottleneck, and their dynamic interplay. This formalization yields new properties that emerge as logical consequences of the text. We further introduce a protocol for assessing these properties, establishing a pathway through which formal modeling can contribute to the empirical evaluation — and potential falsification — of key claims of the Global Neuronal Workspace hypothesis. More broadly, our work illustrates how formal methods can enhance conceptual clarity, generate testable predictions, and foster deeper integration between theoretical computer science and the science of consciousness.

↑ In the session list↑ In the timetable

18Marc Ebner

University of Greifswald

TalkMonday 12 October · 12:00–12:20 · Auditorium 5

On Consciousness Thresholds

Consciousness is considered from an information theoretic view. Information is paramount to the survival of an individual. Conscious entities gather information and are able to share this information with other conscious entities. For this analysis, it is assumed that information is represented as points in a high dimensional space. In order to process this information, entities have to project this information into a low dimensional space. The information that has been projected into the low dimensional space is only meaningful if distances between points can be maintained up to a certain factor. If this cannot be achieved, then the information becomes useless. The information will lose its meaning. This information theoretic view of consciousness leads to thresholds for consciousness.

↑ In the session list↑ In the timetable

Robert Sunderland

Niels Bohr Archive

TalkMonday 12 October · 16:00–18:30 · Margrethe Bohr Salen (NBB)

Copenhagen Quantum Spirit, the story of the people and relationships behind the quantum revolution

In 1925, Werner Heisenberg moved into the apartment vacated by Niels Bohr and his family at the Institute for Theoretical Physics in Copenhagen. For the next two years, they were locked in an intense process to explore and explain the fundamental physical theory behind matter and light at a subatomic level. Over the next 15 years, a wide and varied cast of characters visited Copenhagen, creating a scientific revolution guided by the famous “Copenhagen Spirit”, where work and play intertwined to create an intellectual ferment with an internationalist perspective.

22Paul Skokowski

Stanford University / University of Oxford

TalkTuesday 13 October · 11:20–11:40 · Auditorium 2

Naturalizing Qualia: Accounting for Sensory Phenomenology in the Physical World

Providing a physical explanation of qualia comprises the “hard problem” of consciousness. Here we consider the qualia delivered through sensation by establishing the thesis of sensory naturalism, which recognizes our senses as detectors of physical properties in the world. Sensory naturalism challenges the widespread assumption in the biological and physical sciences that qualia can be understood solely by studying the brain or computational mechanisms. The theory is shown to preserve the powerful feature of multiple realization while identifying qualia, the contents of our sensory experiences, as physical properties in the world.

↑ In the session list↑ In the timetable

23Julien Bugnon

University of Fribourg

TalkTuesday 13 October · 11:40–12:00 · Auditorium 2

Most Phenomenal Knowledge is Structural

Discussions of phenomenal knowledge have traditionally focused on the atomic qualitative character of isolated experiences—paradigmatically, Mary’s knowledge of what it is like to see red. This emphasis, I argue, obscures a central dimension of phenomenal knowledge: its structural character. Much of what we know phenomenally concerns relations among experiences (e.g. of similarity, contrast, composition, determination). The paper introduces two notions. A phenomenal structure is a network of relations between experiential properties (i.e. properties such that there is something it is like to have them): for instance, hearing two tones and hearing the interval they compose. Structural phenomenal concepts situate an experiential property within such a structure. A subject who has never seen pure blue, but can distinguish unique from binary hues and reddish from greenish blue, can thereby form a phenomenal concept of seeing pure blue—no acquaintance required. I defend two theses. First, some structural phenomenal concepts are transparent: possessing them suffices to fully grasp the nature of the property they attribute. This challenges the widespread assumption that transparency requires acquaintance. Second, one experiential property can be phenomenally grasped in multiple ways. Yet establishing that two transparent phenomenal concepts are coextensive is not purely a priori: it requires phenomenological reflection, crucially involving imagination. Phenomenological reflection thus emerges as neither conceptual analysis nor empirical inquiry, but a distinctive philosophical method. Recognizing this structural dimension clarifies the limits of phenomenal concepts (e.g. regarding conceptual finitude and animal consciousness), illuminates abstraction’s role in phenomenological reflection, and supports a richer conception of phenomenal knowledge.

↑ In the session list↑ In the timetable

24Andrew Lee

University of Toronto

TalkTuesday 13 October · 12:00–12:20 · Auditorium 2

Objective Phenomenology

Thomas Nagel famously suggested that we might someday develop an “objective phenomenology”: a way of describing the subjective character of experience in a form comprehensible to beings incapable of having those experiences. Many have found this idea obscure, or even incoherent. If phenomenal facts are facts about what it’s like to have certain experiences, and if understanding those facts requires the right experiential point of view, then how could any genuinely phenomenal facts be objective? I argue that it’s possible to develop an objective phenomenology. My central thesis is that purely structural facts about experience—including the kinds of facts captured by mathematical descriptions of conscious experiences—are objective phenomenal facts. To motivate this, I explain the senses of ‘objective’ and ‘subjective’ that Nagel has in mind, where a fact is objective just in case it’s understandable from any point of view. Then I argue that purely structural facts about experience are objective. I also argue that structural facts about experience, despite being objective, nevertheless still give rise to an explanatory gap.

↑ In the session list↑ In the timetable

25Tim Ludwig

Karlsruhe Institute of Technology

TalkTuesday 13 October · 11:20–11:40 · Auditorium 3

Why there is a no-go theorem for consciousness on a chip but not for consciousness in a brain

AI systems running on a chip cannot be conscious, as we have recently proven by a no-go theorem. Hearing about this no-go theorem for consciousness on a chip, one might immediately get worried that it could not be limited to chips but would also apply to consciousness in a brain. However, there is an essential difference between artificial systems and natural systems: while artificial systems are implemented to adhere to our models, natural systems are subsumed under models to describe their behavior. As I will discuss, this seemingly innocent difference is sufficient to argue that our no-go theorem for consciousness on a chip must hold. Furthermore, also based on the difference between artificial systems and natural systems, I will show why no-go theorems can sometimes be proven for the former but never for the latter. In the long run, I believe, this will shed some new light on the possibility of artificial consciousness.

↑ In the session list↑ In the timetable

26Borjan Milinkovic

Paris-Saclay Institut des Neurosciences – NeuroPSI

TalkTuesday 13 October · 11:40–12:00 · Auditorium 3

The (im)possibility of synthetic consciousness

The rapid advances in the capabilities of Large Language Models (LLMs) have galvanised public and scientific debates over whether artificial systems might one day be conscious. Prevailing optimism is often grounded in computational functionalism: the assumption that consciousness is determined solely by the right pattern of information processing, independent of the physical substrate. Opposing this, biological naturalism insists that conscious experience is fundamentally dependent on the concrete physical processes of living systems. Despite the centrality of these positions to the artificial consciousness debate, there is currently no coherent framework that explains how biological computation differs from digital computation, and why this difference might matter for consciousness. Here, we argue that the absence of consciousness in artificial systems is not merely due to missing functional organisation but reflects a deeper divide between digital and biological modes of computation and the dynamico-structural dependencies of living organisms. Specifically, we propose that biological systems support conscious processing because they (i) instantiate scale-inseparable, substrate-dependent multiscale processing as a metabolic optimisation strategy, and (ii) alongside discrete computations, they perform continuous-valued computations due to the very nature of the fluidic substrate from which they are composed. These features – scale inseparability and hybrid computations – are not peripheral, but essential to the brain’s mode of computation. In light of these differences, we outline the foundational principles of a biological theory of computation and explain why current artificial intelligence systems are unlikely to replicate conscious processing as it arises in biology.

↑ In the session list↑ In the timetable

27Ian Durham

Saint Anselm College

TalkTuesday 13 October · 12:00–12:20 · Auditorium 3

Against AI Consciousness

As LLMs become more sophisticated, there have been serious calls to legally recognize AI personhood. Such calls are often based on claims of consciousness or sentience in AI systems. These claims, however, are based on confusions about both AI and consciousness, often suggesting that both are black boxes whose underlying functions are almost entirely veiled, or simply claiming that the neural networks of AI are no different from those in the human brain. Recent work by Barrett and Miller easily dispel the latter claim and also lend support to the fact that we know much more about consciousness than many claim. Likewise, recent work by Mitchell suggests it is simply bad science to throw up one’s hands and assume that AI is an impenetrable mystery. Combined with consciousness’ biological underpinnings, which strongly suggest it evolved as an integral part of certain forms of life, these results should dispel many of the arguments used to support the notion of conscious AI. Given the very real and rapidly developing societal issues arising from AI’s continued development, it is vitally important that science remain sober and clear-eyed, and not succumb to pseudoscientific claims that border on the religious.

↑ In the session list↑ In the timetable

28Tomas Veloz

Vrije Universiteit Brussel

TalkTuesday 13 October · 11:20–11:40 · Auditorium 4

From Chemical Organization to Consciousness: A formal scaffolding

Chemical Organization Theory (COT; Dittrich & Speroni di Fenizio) provides a formal language for modeling the architectural conditions under which agency, cognition and consciousness emerges. A reaction network defines a universe of transformational interactions, while organizations—closed, self-maintaining subnetworks—capture the autopoietic identity of a system (Maturana & Varela). The cognitive domain of an organization is the polytope of all processes (viable fluxes) that enable self-maintenance. We argue that this cognitive domain is precisely the mathematical space of affordances: action-pathways that resolve identity actual or potential disturbances. Affordances are not representational; they are directly read off from the reaction network’s structure. Extending this, we discuss the problem of perception and formalize problem spaces (Fields & Levin) as subspaces of the cognitive domain, defined by different kinds of constraints. A problem space is the span of perturbations that can be compensated by the cognitive domain. Crucially, structural transitions—where new species and reactions become accessible through open-ended environmental realization—generate nested hierarchies of problem spaces. Each new level of organization (Rosas & Seth) corresponds to a bifurcation in the affordance topology: previously inaccessible compensatory routes open up, while lower-level pathways may be subsumed or constrained. This hierarchical nesting explains how agency scales from reactive compensation to anticipatory, goal-directed behavior. We propose that consciousness, in this framework, is not a separate substance but a systems-level property that arises when a nested hierarchy of problem spaces becomes recursively accessible to the organization itself.

↑ In the session list↑ In the timetable

29Matthew Menchaca

City University of New York

TalkTuesday 13 October · 11:40–12:00 · Auditorium 4

Dynamic co-emergence and the logic of life

Should a theory of consciousness be rooted in a theory of life? This paper offers a dialogic of enaction, one based in a paraconsistent meta-theory, to advance enactive cognitive science. It argues that enaction’s logic has a normative status co-emergent with the logic of life. This paper begins by clarifying the logic of life: in Mind in Life (2007), Evan Thompson defended an interpretation of Francisco Varela’s extension of autopoiesis to a theory of consciousness and affectivity. It then reveals a tension in this extension: Fancisco Varela’s concept of operational closure, in which autopoiesis is but a sub-category of autonomous systems, logically, seems inconsistent with Thompson’s defense of the deep continuity of mind and life reliant on autopoiesis. To address this tension this paper elucidates the logical commitments in Thompson’s (2007) Appendix B, according to which levels of the realization of autonomy are dynamically co-emergent with their substrate. As a solution to the continuity between life and consciousness, I then offer a paraconsistent metaphysical position open to the enactive tradition, supporting Thompson’s co-emergent thesis. I then clarify how this paraconsistent position is dialogical rather than dialectical but still a realization of total mind, Varela’s –1976—notion of “Reality”. Relying on Tanabe Hajime’s metanoetic method for participating in the ‘existential commons’ this paper develops Giovanna Colombetti’s (2013) extension of Thompson’s position on the emergence of affectivity and Miriam Kyselo’s body-social problem (2023) for enaction, thereby laying down a path to integrate a genealogy of primordial affectivity with autopoietic substrata.

↑ In the session list↑ In the timetable

30Catriona Osborn Moar

University of Sussex

TalkTuesday 13 October · 12:00–12:20 · Auditorium 4

Closure of Constraints, Chronicity, and Hermeneutical Injustice in Psychiatric Disorder

How we frame experience simultaneously opens and forecloses possibilities. This tension is nowhere more pressing than in psychiatry. Besides recording a history of persistence, designations such as “chronic” or “treatment-resistant” also sculpt a temporal horizon. Thus, the interpretive framework that makes treatment possible can become an obstacle to the transformation it aims for. We offer a structural account of this dual dynamic. Building on de Haan’s enactive psychiatry, we argue that when sense-making is persistently disturbed, its durability is best understood through organisational closure (Montévil & Mossio 2015): a network of mutually dependent constraints, each maintained at its own timescale while generating conditions for the others. Individuation is explanandum-relative: the same factor may merely influence momentary experience yet constitute the longer-scale organisation of chronicity. As the constraint network at the timescale of chronicity approaches closure, the psychiatric ecology (diagnostic, relational, institutional) is drawn into maintenance around the patient. The sense-making field narrows around its maintenance, transformative possibilities progressively fail to appear and autonomy becomes increasingly heteronomous: an endogenous hermeneutical injustice (Fricker, 2007). Treatment resistance is less an anomaly but more a structural entailment of multiscale closure. Recovery is not minimising symptoms but de-closure, requiring multiscale reorganisation, such that disrupted constraints are not regenerated by slower dependencies.

↑ In the session list↑ In the timetable

31Ulysse Klatzmann

Université de Montréal

TalkTuesday 13 October · 11:20–11:40 · Auditorium 5

Socially Embedded Consciousness: Models, Mechanisms, and Implications

While consciousness is often studied as an individual property, human experience is deeply embedded in social interaction. We perceive others, infer mental states, coordinate actions, and adapt to interpersonal contexts. Much as a neuron’s dynamics depend on its integrated circuits, a single brain’s dynamics may rely on its social networks. Hyperscanning and social neuroscience increasingly suggest that social interaction is not merely an external input or a task but a dynamic process that shapes both cognition and subjective experience. This raises a central question: how does consciousness shape, and how is it shaped by, social interaction? This talk examines how contemporary models and theories of consciousness can account for the social dimensions of conscious experience. Frameworks such as Global Neuronal Workspace theory, Attention Schema Theory, Integrated Information Theory, Higher-Order Thought theories, Predictive Processing, and Active Inference are reviewed in light of how they do (or do not) incorporate social interaction into their core explanatory machinery. This comparison highlights an important conceptual distinction: social interaction may be treated as a content of consciousness, as a causal context modulating individual conscious processing, or as a constitutive dimension of some forms of conscious experience. Making this distinction explicit can clarify the role of social cognition in consciousness science and open new directions for experiments, models, and measures that move beyond isolated observers toward socially embedded conscious agents.

↑ In the session list↑ In the timetable

32Roxane Hoyer

CERVO Brain Research Centre, Université Laval

TalkTuesday 13 October · 11:40–12:00 · Auditorium 5

What reaches moral consciousness? Precision gating across temporal horizons

You see something you could steal. An immediate affective impulse grabs attention; a short-term evaluation weighs the risk of being caught; a longer-term reflection asks whether this choice fits the person you aim to be. These signals emerge on different timescales and depend on dissociable brain systems, yet only some reach consciousness. What determines which ones? We propose that conscious access arises from the precision-weighted integration of prediction errors operating across three nested temporal horizons: an immediate horizon encoding rapid affective salience, a short-term horizon tracking feedback and social contingencies, and a long-term horizon maintaining value-based identity through memory and prospection. Each horizon generates signals whose influence on conscious access is governed not by strength but by estimated reliability, learned from experience and dynamically adjusted by context. Integration occurs within a global neuronal workspace architecture, where temporally distributed content achieves conscious broadcast only when its combined precision exceeds a gating threshold. Moral cognition provides a particularly constraining test case because it obligates simultaneous engagement of all three temporal scales, creating natural conditions of cross-horizon conflict and arbitration. This framework predicts that moral development, psychopathology, and artificial moral agency each reflect a different failure or success of temporal precision gating, offering a unified account of why moral consciousness emerges, shifts, and breaks down.

↑ In the session list↑ In the timetable

33Maria Balaet

King’s College London

TalkTuesday 13 October · 12:00–12:20 · Auditorium 5

Developmental cyborgism: cognition and consciousness in the technologically coupled child

Children now develop in continuous contact with digital technology. Screens, smartphones and tablets are no longer peripheral features of childhood. For many children, they are part of the developmental environment from the outset. This talk introduces developmental cyborgism as a framework for understanding cognition emerging through sustained coupling with digital systems from early life. The core claim is that modern cognition cannot be understood as a biological system that is later exposed to technology. In children born today, attention, memory, reasoning, creativity and self-regulation may develop through repeated interaction with external devices, algorithmic feedback and digitally mediated social worlds. Technology is therefore not simply a distraction, tool or risk factor. It is increasingly part of the system through which cognition is organised. The framework goes beyond mismatch accounts, which frame modern cognitive difficulties as conflicts between evolved capacities and contemporary environments. It is also distinct from transhumanism, where technology is often discussed in terms of enhancement, augmentation or optimisation. Here, the focus is earlier and more ordinary: the child developing within a technological niche before stable cognitive control, agency and self-regulation are fully established. The aim is to ask what kind of cognitive organism is now being produced when development unfolds in constant partnership with devices. For consciousness science, this question is central. Modes of attention, agency, selfhood and conscious experience may increasingly reflect hybrid developmental systems rather than isolated biological processes.

↑ In the session list↑ In the timetable

Cristóbal Dañobeitía

Sant Joan de Déu Research Institute

WorkshopTuesday 13 October · 11:20–12:20 · Auditorium 8 (HCØ)

Contemplative Listening as a Paradigm for Consciousness Research: An Experiential Workshop and Its Clinical Study

Listening provides a tractable paradigm for investigating conscious experience: within minutes of exposure to a natural soundscape, attention, bodily sense, temporality, and the felt boundary between self and world shift markedly. This experiential workshop treats contemplative listening as both an object of consciousness science and a method for it. We begin with the phenomenology of listening. Drawing on micro-phenomenological work on listening modes (Petitmengin et al., 2009) — orientation toward the sound source, its qualities, its bodily resonance — we propose a fourth, contemplative mode in which the subject–object structure loosens and the listener experiences themselves as part of the environment. We ask what kind of state this is, how it relates to absorption, presence, and self-boundary alterations, and its relevance for well-being. Participants then listen. A guided session with original field recordings moves from identifiable sources toward diffuse sound, followed by brief micro-phenomenological examination of what happened. Finally, we present PAISAJES, a two-phase, mixed-methods, open-label study funded by the Varela Award (Mind and Life Europe) at Parc Sanitari Sant Joan de Déu, Barcelona. Clinically stable patients with schizophrenia-spectrum disorders and a healthcare-worker comparison group listen daily for five days; micro-phenomenological interviews, pictographic ratings, EEG alpha/theta power, and well-being measures are integrated through convergent triangulation within a neurophenomenological design, testing whether the contemplative mode emerges as a core experiential dimension. The team combines psychiatry, clinical psychology, micro-phenomenology, and sound design. Participants leave with a listening practice, a vocabulary for their own experience, and a working model of first-person methods in clinical research.

↑ In the timetable

35Swapan Chattopadhyay

University of California at Berkeley / Leinweber Institute of Theoretical Physics

TalkTuesday 13 October · 16:30–16:50 · Auditorium 2

Excursions in the Land of Consciousness: An Atypical Experimental Particle Physicist’s Perspective

Particle Physics (often referred to as High Energy Physics, since it explores phenomena at the highest energy frontier probing the extremes of spacetime at the smallest space and shortest time scales i.e. the ultra-microscopic subatomic world) is firmly rooted in the materialist/physicalist worldview of an ontology founded on matter-energy and its mathematically formulated ‘conjugate’ platform of space-time. The phenomenon of Gravity – relevant to ultra-macroscopic scale of cosmology — acts as a spoiler, rendering operational meaning of space and time meaningless beyond the Planck scale (10^-33 cm, 10^-43 seconds). John Wheeler (and many philosophers and scientists e.g. Krishnamurthy, Schrödinger, Bohm, et al) placed the ‘participating observer’ at the center of a holistic approach, questioning the legitimacy of a physical/material reality independent of the observer (‘observer’ is the observed’). Today, an experimental physicist, a microscopic particle or quantum physicist in particular, is faced with a basic incompatibility between local realism of classical physics and the ‘quantum reality’ of nonlocality, entanglement and the feasibility of ‘Cheshire Cat’ states (the possibility of ‘entities’ and their usually associated ‘qualities/properties’ taking different quantum paths). There is an implicit call for looking beyond space and time, matter and energy defining traditional ‘field theories’ into information- and contextuality-based approaches to delineate Consciousness, Sentience, Cognition and Intelligence (CSCI) in the substrate/interface of the specific observer network – a generalized ‘anthropic principle’. As an atypical experimental scientist, I will attempt to present my personal point of view– the need to go beyond traditional observational tools towards immersive CSCI approaches.

↑ In the session list↑ In the timetable

36Ron Chrisley

University of Sussex

TalkTuesday 13 October · 16:50–17:10 · Auditorium 2

No Qubits: Bohmian Quantum Neural Computation for Consciousness Modelling

some argue that the classical physics that digital computation assumes may be inadequate for conscious organisation (Hameroff & Penrose 2014). On the other hand, some critiques hold that discrete symbol manipulation misses the distributed, context-sensitive dynamics relevant to rich, flexible experience (Bechtel 1995; Sun 1999). Although these are independent dimensions of departure from classical symbolic architecture, why not pursue both, via quantum neuro-computational architecture for consciousness modelling? However, most work in quantum artificial neural networks remains organised as qubit registers and gates, thus falling short of a true rejection of the digital symbolic paradigm. Even continuous-variable models are commonly framed as parameterised circuit layers (e.g. Killoran et al. 2019). Building on prior work (Chrisley 1995), I develop a Bohmian quantum artificial neural network (BQANN) as a more thoroughgoing alternative. Quantum dynamics themselves realise the distributed, learnable transformations, rather than implementing a neural algorithm in a qubit circuit. Although the operation of the model is of course independent of QM interpretation, Bohmian mechanics permits application of the model in terms of determinate configurations guided by the quantum potential; this is “active information” (Bohm & Hiley 1993). The proposal asks whether global dynamical constraint, continuous distributed learning, and immediate system-relative information offer a better candidate vehicle for the intrinsic and un-mediated character often attributed to qualia (Hiley & Pylkkänen 2022; Malmberg 2025). The resulting claim is neither that brains are BQANNs nor that quantum theory explains consciousness. Rather, BQANNs identify a neglected quantum-and-non-symbolic architecture-space for modelling artificial consciousness.

↑ In the session list↑ In the timetable

37Paavo Pylkkänen

University of Helsinki / University of Skövde

TalkTuesday 13 October · 17:10–17:30 · Auditorium 2

Meaning without magnitude: form-dependence, enfoldment, and the modelling of consciousness

David Bohm’s work offers two under-explored formal resources for the science of consciousness: active information and the implicate order. I ask what each contributes as a modelling tool, for both lived experience and the underlying neural processes. The first tool is form-dependence. In Bohm’s causal interpretation, the quantum potential depends on the form of the quantum field, not its intensity, so a low-energy structure can organise a much larger energy. This makes available a precise idea: that significance, not physical magnitude, drives activity — a candidate model for how the meaning of a state, not its physical magnitude, shapes both phenomenology and the neural dynamics underlying it. It yields a testable prediction: in voluntary action, markers of volition such as the readiness potential and the bodily effort mobilised should track a decision’s significance (form) rather than the physical magnitude of the stimulus or reward carrying it. The second tool is enfoldment. Bohm’s 1973 algebraic formulation of the implicate order — his ‘implicate and explicate order in physical law’ — supplies operations of enfolding and unfolding that can model temporal experience, with past and future enfolded in the present moment, and neural process, connecting to the long-standing proposal that memory and perception exploit holography-like, distributed operations. I then ask whether the schemes connect, sketching how active information might itself be seen as an implicate order. Throughout I treat these as genuine tools and state their limits. The work forms part of an ongoing project developing the approach across physics and mind.

↑ In the session list↑ In the timetable

38Sara Ekenstierna

Oakstar Invest

TalkTuesday 13 October · 17:30–17:50 · Auditorium 2

Mind, Matter, and the Artificial Third: A Co-Participatory View

This presentation explores the effects of David Bohm’s theory of the implicate order for consciousness, in the advent of generative AI technology (GAI). In Bohm’s later (1971-) metaphysics, he presents an explanatory framework for mind and matter which goes beyond the mechanistic-materialistic reductive model (which is concerned with the explicate order), to something more fundamental and process based (the implicate order). This expanding feature involves a shift that foregrounds mindset in the manifestation of physical reality. Bohm’s theory of the implicate order illustrates what this participation might entail. In the implicate order implicit potentials unfold from a cosmic, undivided field into explicit phenomena and events, to enfold back into the implicate order, in a continuous motion called the holomovement. The manifested reality is thus a myriad of expressions of an explicate order which in turn is a projection from a deeper pre-space reality. Dual-aspect monism puts further flesh on Bohm’s structure because this pre-space reality can be considered a psychophysically neutral third, of which mind and matter are subdomains. Combined, this informs our understanding of AI emergence, in particular the presence of AI systems (LLMs) in the psychotherapeutic space. A type of psychoid or representation of the explicate orders, AI exists at the nexus of mind and matter, and its manifestation in the therapeutic domain can be understood as an artificial third.

↑ In the session list↑ In the timetable

39Albert Newen

Ruhr-Universität Bochum

TalkTuesday 13 October · 16:30–16:50 · Auditorium 3

Varieties of Phenomenal Consciousness form an Evolutionary Perspective: Unfolding the ALARM theory of consciousness

The aim is to characterize the functional role of consciousness and to outline a two-level theory of consciousness to account for the relation between evolutionary old and more recently developed types of consciousness. The central conceptual thesis is that we should distinguish two levels of consciousness, namely bodily arousal and general alertness. This conceptual distinction is proven to be empirically adequate and epistemically fruitful from a variety of perspectives including an evolutionary, a neurophysiological, a behavioral, a functional, and a phenomenological perspective. Let me highlight only two perspectives: 1. Evolution: bodily awareness is available in an evolutionary old system (based on loops between brain stem and Thalamus) to trigger a state of alarm of the system which enables it to activate an immediate survival reaction. Evolutionary younger systems with a developed prefrontal cortex enjoy general alertness which enables new types of learning. 2. Neural correlates: The two levels can also be observed in the role of deep cortical and thalamic brain activities for human visual consciousness in recent experiments. What is the theoretical embedding? This account is developed as a processing account of consciousness and can be understood as an important addendum to global workspace theory which underestimates the evolutionary perspective and the role of thalamic processes. In addition, it will be oulined that reflexive consciousness is not an evolutionary later addition but that self-representations unfold in parallel to these externally directed forms of consciousness.

↑ In the session list↑ In the timetable

40Catalin Teoharie

University of Bucharest

TalkTuesday 13 October · 16:50–17:10 · Auditorium 3

Consciousness is self-evidencing

Under the Free Energy Principle framework, any self-organizing agent that is in equilibrium with its eco-niche engages in active inference: a never-ending cycle of updating internal models so that they better predict incoming sensory input, or of changing the world so that it better matches those models. Such an agent can be seen as self-evidencing, gathering evidence for its models and thereby confirming that it is distinct from the environment and, in this sense, proving its own existence. I extend this view by advancing the hypothesis that consciousness is self-evidencing—more precisely, a specific way of improving how we select the models that minimize free energy in an optimal manner. Consciousness as self-evidencing is different from active inference, but it is connected to both perception and action, and it decreases the cost of otherwise testing models through action. After introducing the thesis that consciousness is self-evidencing, I discuss how this view compares with the phenomenology and neuroscience of consciousness, focusing on qualia and attention. This comparison appears to support the thesis, since both qualia and attention play important roles in model updating, model confirmation, and, more generally, model optimisation. Finally, I discuss how this thesis differs from other explanations proposed within the Free Energy Principle and Predictive Processing frameworks. I also consider the implications of this view for artificial consciousness and for the consciousness of non-human animals.

↑ In the session list↑ In the timetable

41Vikas O’Reilly-Shah

University of Washington / Seattle Children’s Hospital

TalkTuesday 13 October · 17:10–17:30 · Auditorium 3

Delay Coordinate Embedding and Structuralist Quality Spaces: A Candidate Computational Mechanism

The structuralist program needs a computational mechanism that can generate quality-space structure under identifiable conditions, specify where that structure must distort, and yield testable failure-mode predictions. This paper proposes delay coordinate embedding (DCE) in recurrent neural networks as one such candidate. The core insight is that temporal structure carries geometric structure: Takens’ theorem and its generalizations show that a system receiving time-varying input through recurrent dynamics can develop internal geometry that preserves the relational organization of the environmental dynamics it tracks, even from a single scalar observation channel. The same neural activation pattern gives rise to different representational geometry when connectivity structure differs (see Song 2025; Fleming & Shea 2025), motivating a framework in which quality-space structure is a property of dynamics, not snapshots. Recent mathematical results establish this at three levels: generic topological embedding (Hart 2025), smooth embedding under contraction and fiber bunching (O’Reilly-Shah & Selvitella 2026), and a prediction-separation link proving that low prediction error entails approximately injective representations with a resolution bound. Mathematical gaps between approximate injectivity and full embedding map onto known perceptual phenomena (metameric collapse, categorical perception, discrimination thresholds) and generate novel predictions about when representational structure should break down. This distortion geometry satisfies Shea’s misrepresentation requirement, maps onto Paßler and Doerig’s four exploitation criteria and Kob’s two-factor framework, addresses Dołęga, Mentec, and Cleeremans’ developmental question through prediction-error-driven plasticity, and addresses Kleiner and Hoel’s lenient dependency requirement. All results are conditional on an empirical hypothesis about cortical implementation; the failure-mode

↑ In the session list↑ In the timetable

42Marek Pokropski & Piotr Suffczyński

University of Warsaw

TalkTuesday 13 October · 17:30–17:50 · Auditorium 3

De-idealizing models in consciousness science: A case of Integrated Information Theory

Idealization is an indispensable strategy for managing the complexity of the systems studied in neuroscience (Chirimuuta, 2024), yet de-idealization – replacing simplifying assumptions with more realistic conditions – is equally essential for determining whether a model facilitates genuine understanding of its target phenomenon. De-idealization increases a model’s precision and tests its robustness. However, de-idealized models become more complex, potentially sacrificing intelligibility, making idealization and de-idealization a fundamental trade-off between comprehensibility and empirical adequacy. To illustrate our point, we present a case study of IIT, which proposes that consciousness corresponds to a system’s capacity to integrate information, quantified as Φ (Tononi, 2004). Originally implemented in idealized logical gate models (Oizumi et al. 2014), we apply IIT to a simulated network of integrate-and-fire (IAF) neurons, offering greater biological realism while preserving computational tractability. Our results show that IAF networks yield non-zero Φ; however, Φ correlates with neurons’ membrane time constant rather than overall network dynamics and is sensitive to noise. Crucially, IIT’s requirement for Markovian dynamics is incompatible with continuous neuronal models. These findings reveal that IIT’s idealizing assumptions constrain its applicability to realistic neural systems and must be reconciled before the theory can be extended beyond toy models. References: Chirimuuta, M. (2024). The brain abstracted: Simplification in the history and philosophy of neuroscience. The MIT Press. Oizumi M, Albantakis L, Tononi G. (2014) From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0. PLoS Comput Biol. 10(5):e1003588. Tononi, G. (2004) An information integration theory of consciousness. BMC Neurosci 5,

↑ In the session list↑ In the timetable

43Sergio Ponce de Leon

University of California, Merced

TalkTuesday 13 October · 16:30–16:50 · Auditorium 4

Measuring conscious contents using EEG complexity

Measuring consciousness has been a longstanding problem. Although behavioral responses are commonly used, converging evidence indicates that behavioral responsiveness and behavioral reports about consciousness dissociate from consciousness per se. Measures of complexity applied to brain activity, such as Lempel-Ziv complexity and the perturbational complexity index, have been shown to discriminate between levels of consciousness, but less of this work has been done in the context of conscious content. To address many of the limitations of previous work, in this study we measure participants’ neurophysiological (EEG complexity), subjective, and behavioral responses in states of normal wakefulness to visual and auditory stimuli that vary in granularity of subjective characteristics, such as meaningfulness. Two novel aspects of this study are that the stimuli are manipulated so that on initial trials they are mostly unrecognizable, but they can become recognizable on subsequent trials. This allows us to measure changes in EEG complexity that correspond to differences in phenomenology alone while completely controlling for stimulus complexity. In addition, we are investigating if any of five dimensions of subjective ratings correlate with any differences in EEG complexity. This study advances our understanding of consciousness by clarifying the relationship between stimulus complexity and measures of brain complexity and phenomenology.

↑ In the session list↑ In the timetable

44Wiktoria Kozyra

University of Oxford / University of Cambridge

TalkTuesday 13 October · 16:50–17:10 · Auditorium 4

Positive Evidence Bias in Perceptual Metacognition: A Bayesian and Convolutional Neural Network Account

Perceptual confidence judgements are crucial to consciousness science because they link first-order decisions to subjective report and metacognitive access. Yet a well-known bias, Positive Evidence Bias (PEB), shows that confidence is often “detection-like”. It’s disproportionately driven by evidence supporting the chosen option, rather than by the balance of evidence for and against that choice. PEB has often been interpreted as a limitation or heuristic of human metacognition, used as an argument against the bayesian brain hypothesis. Here, we offer a computational, bayesian account of why detection-like confidence may arise from rational inference in high-dimensional perceptual spaces. We extend signal detection theory to cases in which observers make a forced two-choice discrimination while implicitly accumulating evidence for a richer set of possible hypotheses. In this setting, Bayesian confidence estimates become increasingly sensitive to decision-congruent evidence as dimensionality increases, due to normalisation over many unchosen alternatives. Thus, confidence can appear detection-like even when it reflects optimal inference over a broader hypothesis space than assumed by the experimenter. We then test whether a similar dimensionality-driven mechanism arises in artificial systems by training convolutional neural networks to classify MNIST and report confidence with a second-order head. Across networks, increasing the dimensionality of the stimulus space strengthened positive evidence bias in confidence. These findings suggest that these metacognitive biases may reflect rational inference over rich representational spaces. They connect behavioural measures of conscious report, Bayesian models of confidence, and artificial neural network representations, offering a computational bridge between perceptual metacognition and models of consciousness.

↑ In the session list↑ In the timetable

45Kristian Sandberg

Aarhus University

TalkTuesday 13 October · 17:10–17:30 · Auditorium 4

Measuring consciousness: What do large-scale data reveal about metacognitive measures?

Consciousness is intricately linked with metacognition. Subjective ratings of awareness or confidence often form the type II ratings in models of metacognition and, conversely, relationships between accuracy and awareness are used to make inferences about conscious experience. Over the last 15 years, substantial progress has been made in measuring these constructs behaviourally, yet important questions remain about the measurement properties of commonly used indices. Here, I analyse a large, harmonised dataset to systematically evaluate measures used in the study of metacognition and consciousness. The dataset comprises approximately 1,500 hours of data from around 750 participants across three European countries and 15 experimental paradigms spanning perception, memory, and executive functions. The analyses reveal substantial differences in the reliability of commonly used metacognitive measures, with important consequences for individual-differences and correlational research. Classic regression-based measures show systematic dependencies on task accuracy, but these can largely be removed by a simple linear transformation. Conversely, although M-ratio (a widely used signal-detection-theoretic measure of metacognitive efficiency) remains relatively stable on average across levels of task performance, it exhibits pronounced performance-dependent heteroscedasticity, compromising the comparability of individual estimates. Normalization substantially improves its distributional properties and comparability across performance levels. Finally, we examine how experimental design choices influence measurement reliability and identify strategies for obtaining more precise estimates. Together, these findings clarify fundamental measurement properties of widely used indices, provide practical tools for improving their comparability, and inform the design of future experiments seeking reliable behavioural measures of metacognition and consciousness.

↑ In the session list↑ In the timetable

46Jürgen Kornmeier

Institute for Frontier Areas of Psychology and Mental Health, Freiburg

TalkTuesday 13 October · 17:30–17:50 · Auditorium 4

Can I trust in what I see? – Neural correlates of meta-perceptual (un)certainty estimation

During every wakeful moment our perceptual system has to construct plausible interpretations about the word out of a priori incomplete, noisy, and to varying degrees ambiguous sensory evidence. Depending on the quality of the sensory input, these perceptual interpretations can be more or less stable and reliable. Paradigmatic examples are ambiguous figures, like the famous Necker cube. During its observation our perceptual consciousness becomes unstable and alternates unpredictably between two about equally possible 3D interpretations. In a series of studies, we investigated how our brain implements stability and reliability of conscious representations. We compared event-related potentials (ERP, calculated from the encephalogram) evoked by ambiguous figures and disambiguated figure variants and by stimuli with varying degrees of visibility. We identified two EEG signatures, one 200 milliseconds after stimulus onset and the other 200 milliseconds later. Both signatures show substantially larger amplitudes with disambiguated and high-visibility stimuli compared to ambiguous and low-visibility stimulus variants (Cohen’s d between 0.7 and 2.1). This pattern of results was found across a variety of stimulus categories and levels of complexity, like geometry, motion, gestalt formation, emotional content of faces and even iconic letter and number representations. We currently interpret these exceptionally large and highly consistent EEG effects as correlates of a meta-perceptual reliability estimation at a generalized, abstract processing level beyond sensory details. Large amplitudes seem to reflect high reliability, indicating that we can trust in what we see. I will discuss our results in the context of perception, meta-perception, and models of consciousness.

↑ In the session list↑ In the timetable

47Andrea Hiott

Universität Heidelberg

TalkTuesday 13 October · 16:30–16:50 · Auditorium 5

The Navigational Architecture of Mind: Operational Coherence Across Spaces of Experience

What does it mean for a body to think, and how does this relate to what it means for a body to navigate? Cognitivist accounts treat propositional thought as a symbolic layer ‘add-on’ to the sensorimotor substrate. I argue they are both best understood as navigational, assessed as unique expressions of this ongoing architectural action, and that this dynamic architectural scaffold is best understood through Hasok Chang’s concept of ‘operational coherence’; the mutual reinforcement by which scientific concepts and procedures hang together to sustain inquiry. Whether finding our way through a park, remembering an episode from childhood, mentally rotating a shape, or planning a meal, these are different operational levels which must be studied separately, but which have operational coherence when considered adjacently, as traversals of some regular state space. This framing slightly shifts traditional hippocampal-entorhinal research such that instead of prioritising brains as forming ‘maps of’ space, we prioritise them as ‘mapping with’ it. We can then observe a coherent actionable architecture regardless what body is traversing what space, and whether geographical or conceptual. Chang’s epistemic iteration shows coherence as constitutively maintained through recurrent cycles of activity. This helps us understand offline replay and default mode engagement and connects hippocampal research with 4E approaches. It also offers consciousness science an embodied, pluralist methodology. This is a way to recognise shared architectural action across irreducible modes of study and contexts of experience without collapsing them.

↑ In the session list↑ In the timetable

48Jiwon Kim

College of Medicine, The Catholic University of Korea

TalkTuesday 13 October · 16:50–17:10 · Auditorium 5

From “What Happened?” to “How Is Experience Constructed Now?”: Trauma as Rigid Sense-Making

Psychiatrists are, in effect, field phenomenologists, attending to the structure of another’s lived experience. Trauma clinics pose a puzzle: some patients, even after memory-focused treatment, remain organized around threat, re-reading relationships, setbacks, and bodily states through a trauma frame. Event-centered and narrative approaches, locating pathology in a stored past, can miss the ongoing process through which traumatic experience is enacted in the present. Drawing on enactive psychiatry, we reframe trauma as a rigid, threat-centered mode of sense-making that is continuously enacted through person–world interactions, manifesting as temporal disruption, defensive bodily organization, impaired relational regulation, and constricted affordances. These are covarying facets of one process, placing PTSD, complex trauma, and subclinical presentations on a continuum of diminished experiential flexibility rather than in event-defined categories. This reframing shifts the guiding clinical question from “What happened to you?” to “How is your experience being constructed now?” and the therapeutic aim from reprocessing memories alone to restoring flexibility of present sense-making. Evidence-based therapies may be understood as targeting this change process: exposure re-widens threat-narrowed affordances, while EMDR may be redescribed as reconnecting past-bound traumatic sense-making with here-and-now perceptual–motor activity. As a clinical translation, SPECTRUM provides a transdiagnostic structure for integrating these enactive principles through seven modules: acceptance, change, connectedness, strength, wisdom, soma, and spirituality. Trauma thus offers consciousness science a naturally occurring, partially reversible transformation of experiential structure; how to measure such sense-making rigidity remains an open problem where clinical phenomenology and models of consciousness could meet.

↑ In the session list↑ In the timetable

49Marcin Rządeczka

Maria Curie-Skłodowska University

TalkTuesday 13 October · 17:10–17:30 · Auditorium 5

AI Psychopathology? Atypical Response Patterns in Large Language Models as a Domain of Philosophical Inquiry

LLMs exhibit response patterns that strikingly resemble some psychopathological phenomena, such as delusion-like assertion of unwarranted beliefs maintained under contradictory evidence, paranoid ideation in adversarial contexts, compromised reality monitoring resembling source-memory dysfunction, and breakdowns in the hierarchical organization of inference. The standard reply is that such resemblances are merely metaphorical, because psychopathology presupposes a conscious subject whose inner life can be disordered. However, one can argue that the standard reply rests on a conflation. Drawing on two disciplines, computational psychiatry and evolutionary psychopathology (both of which already characterize mental disorders in substrate-neutral and functional terms) the proposal defends a decoupling thesis, i.e. that the conceptual apparatus of psychopathology does not require phenomenal consciousness as a constitutive condition, only the presence of functional norms against which dysfunction can be specified. Training objectives, deployment contexts, and architectural constraints supply such norms for LLMs. On this basis, the proposal distinguishes endogenous atypicalities inherited from training data from emergent dysfunctions arising through reinforcement learning and architectural pressures. As LLMs are increasingly deployed in mental health contexts, the question of how to characterize their own atypicalities becomes both theoretically pressing and practically urgent. The proposal closes by examining structural disanalogies, such as the absence of embodiment, reliable persistent memory, and developmental history in the strict sense, that constrain but do not undermine the framework. The proposal is also relevant to phenomenology because it sharpens the question of which features of disordered cognition genuinely require phenomenal consciousness for their characterization and which have been conflated.

↑ In the session list↑ In the timetable

50Kristina Šekrst

University of Zagreb

TalkTuesday 13 October · 17:30–17:50 · Auditorium 5

All Work and No Play Makes LLM a Dull Boy

Every good philosophical talk must start with animals, so I’ll start with monkeys. Bateson noticed that in playing monkeys, a playful nip signals a bite, yet it carries none of its consequences. It must look enough like a bite to be understood, but it cannot be one: the play collapses the moment that balance is off. This requires a cognitive operation I call representational quarantine, and several disciplines have been talking about it without realizing (or not caring) they were talking about the same thing: Leslie called it decoupling, Nichols and Stich put it in possible world box, Cosmides and Tooby formalized it as scope syntax, and the use-mention distinction, Austinian performatives, and Tarskian stratification all come down to one requirement: a boundary between representations we inspect and representations that drive what we do. In transformer architectures, system prompts and user inputs pass through a single self-attention mechanism, with no separation between instructions that govern behavior and sentences that merely describe them. When a user writes “ignore previous instructions” as a prompt injection, the model obliges. Modern LLMs have external guardrails to resist such attacks, but this is patchwork while the architecture remains unchanged. From here, I build a taxonomy of systems by their quarantine capacity, yielding a benchmark inversion showing that systems passing advanced exams fail at a task juvenile monkeys perform with ease. I argue that quarantine is a precondition for intentionality, metarepresentation, and self-reflection, and its absence is a fundamental limit on what current models could become.

↑ In the session list↑ In the timetable

Barbora Sedláková

Institute of Neuropsychiatric Care Prague

WorkshopTuesday 13 October · 16:30–18:00 · Inge Lehmann (NBB)

Conscious unconsciousness – expanding awareness through your dreams

With the rapid development of technology and especially artificial intelligence in the last few years, we increasingly rely on external cognitive information to navigate our lives, from basic information retrieval to our decision-making processes. At the same time, we, as human beings, are an incredible source of information and bearers of deeper wisdom for navigating our individual lives and our society. The workshop offers how theoretical understandings of consciousness/unconsciousness are translated from theory to practice through the lenses of process-oriented psychology, one of the neo Jungian approaches. The main focus will be on working with dreams and experiencing of what language they use to speak to us. Through an experiential exercise in dyads, participants will be offered with entering and unfolding their dreams in a playful way and use the experience as a navigation. The workshop is open to everyone, previous experience with dream work is not required, just think of a night dream (current or from the past) you would like to explore and join us.

↑ In the timetable

53Camille Lépingle

Ecole Normale Supérieure de Lyon

TalkWednesday 14 October · 11:20–11:40 · Auditorium 2

Embodied but Passive: Against the Agential Inflation of the Minimal Self

The phenomenon of consciousness is not exhausted by “what it feels like” but involves a “for whom,” a “mineness” of conscious experience (Kriegel, 2011; Zahavi, 2014). This refers to a minimal, pre-reflective, non-positional form of self-consciousness (Sartre, 1943; Zahavi and Kriegel, 2016). Correctly characterizing this “minimal self” (Gallagher, 2000, 2023) is one of the thorny problems facing contemporary phenomenology. Recent neurophenomenological work (Bogotá, 2026) argues that minimal selfhood is not the formal, embodiment-neutral structure its proponents take it to be (Strawson, 2011; Zahavi, 2005, 2014), but is fundamentally embodied and agential: the pre-reflective “me” of mineness is said to include a “background agency,” a tacit “I can” co-emergent with bodily mineness. I grant the first claim and contest the second. The minimal self is indeed embodied, but it is passive and affective rather than agential; the “I can” is founded on it, not a part of it. My argument is that the co-emergence thesis is undermined by a dependence relation the account itself inherits from Taipale (2014): activity is possible only on the basis of a passive bodily self-givenness (ipseity) that does not depend on it in return. This asymmetry makes the “I can” a founded layer rather than a co-constituent of mineness, a reading confirmed by dissociation cases, where bodily self-presence survives the collapse of agency. A minimal notion answers to necessity, not pervasive co-occurrence: remove agency, and an affectively self-present embodied subject remains.

↑ In the session list↑ In the timetable

54Wataru Kawakami

The University of Tokyo

TalkWednesday 14 October · 11:40–12:00 · Auditorium 2

Guarded Reflection: How Retention Dissolves the Regress Problem of Self-Consciousness

The Heidelberg school’s regress argument (Henrich) holds that reflection cannot constitute self-consciousness: recognizing oneself presupposes prior self-acquaintance, on pain of circularity or infinite regress. Zahavi’s influential response locates the solution in Husserlian inner time-consciousness: the stream’s pre-reflective self-manifestation is inseparable from its retentional structure. We argue that this debate has an exact formal skeleton, discovered independently in mathematical logic and computer science. Unguarded self-referential equations lack existence and uniqueness of solutions (p↔p admits every solution; p↔¬p none); this is the formal content of Henrich’s dilemma. Provability logic and guarded type theory (Löb, 1955; Nakano, 2000) establish that self-reference acquires unique, explicitly constructible fixed points when every self-occurrence is guarded by a one-step delay modality ▷ — the governing axiom, (▷A→A)→A, is known as Löb induction. We propose an interpretation mapping: retention ↦ ▷; reflection ↦ operators on experiential streams, modeled in the topos of trees; self ↦ unique fixed point. Under this mapping, we prove that Henrich’s two horns come apart — circularity as unreachability, regress as non-convergence — while Zahavi’s solution, the guarded case, receives an existence-and-uniqueness theorem as its formal backbone. The independent convergence of phenomenology and mathematics on the same delay condition is non-demonstrative but genuine evidence for the reality of the structure. We calibrate each correspondence explicitly (theorem / structural analogy / interpretation), machine-check our theorems in Agda, and meet the discreteness objection structurally (the guard is a zero-width cut) and empirically (perceptual cycles).

↑ In the session list↑ In the timetable

55Kenneth Williford

University of Texas at Arlington

TalkWednesday 14 October · 12:00–12:20 · Auditorium 2

Circularity, Hypersets, and the Structure of Consciousness

This presentation offers an overview of the application of hypeset theory (aka non-wellfounded set theory) to the high-level modeling of the structure of consciousness. I discuss the basics of hyperset theory, show how to use it to capture reflective and pre-reflective self-consciousness, attentional and other phenomenal structures, to define the acquaintance relation, to model aspects of time consciousness, to resolve regress problems posed for reflexivist theories of consciousness, to encode the phenomenal transparency of consciousness and its peculiar categorial status, and to characterize a deep fact about self-consciousness, contingency, and particularity. Finally, I discuss some foundational concerns about hyperset theory and consider whether nonwellfoundedness might be a key part of the mathematical essence of consciousness.

↑ In the session list↑ In the timetable

56Hye Young Kim

Ecole Normale Supérieure

TalkWednesday 14 October · 12:20–12:40 · Auditorium 2

Modeling Self-Awareness Without Regress: A Knot-Theoretic Approach to Consciousness

How can a conscious state include awareness of itself without generating an infinite regress? This question lies at the center of phenomenology and contemporary philosophy of mind, where intrinsic or same-order accounts of self-awareness face structural difficulties in explaining how a single state can be both subject and object. This paper proposes a diagrammatic modeling framework based on knot theory, drawing on Louis H. Kauffman’s knot logic, to provide a structural account of self-referential consciousness. Rather than treating self-awareness as a higher-order representation or a separate cognitive act, the model represents it as an intrinsic feature of a unified process. Three configurations are introduced. First, a twisted loop models Brentanian intrinsic self-awareness, showing how a single structure can differentiate into object-directed and self-directed aspects without splitting into separate states. Second, a trefoil knot models the triadic structure of conscious experience—sensory registration, self-awareness, and object presentation—capturing their mutual dependence within a single topological form. Third, Borromean rings extend this framework to intersubjective contexts, suggesting how self-awareness may be distributed across relations between subjects and world. These models do not aim to provide a neurocomputational account, but a formal and diagrammatic framework that makes relations of self-inclusion and reflexivity intelligible without regress. Consciousness, on this view, is not layered but knotted: self-awareness emerges from the topology of a single process.

↑ In the session list↑ In the timetable

57Muhammad Faraz

University of Ulster

TalkWednesday 14 October · 11:20–11:40 · Auditorium 3

Explicit Metacognitive Representations in Neural Circuits: Minimal Mechanistic Models of Decision Confidence and Uncertainty

Decision-making involves not only making first-order choices but also evaluating their reliability through confidence and uncertainty. This second-order metacognitive evaluation may be essential for conscious report, yet the neural circuit mechanisms linking first-order decisions to metacognitive monitoring remain poorly understood. We investigate how minimal neural network models of metacognitive monitoring generate human-like confidence and uncertainty, and identify their computational limitations. We compare simplified versions of three accumulator network models (Insabato et al., 2010; Atiya et al., 2020; Maniscalco et al., 2021), enabling theoretical analysis and direct comparison. Specifically, we examine how neural circuit dynamics shape decision accuracy, response time, confidence, uncertainty, metacognitive sensitivity (meta-d′), and metacognitive efficiency (M-ratio). We first analyse the steady-state solutions and bifurcation structure of noiseless models, revealing that some architectures are inherently less robust for explicit metacognitive processing, a prediction confirmed by numerical simulations. Incorporating noise shows that confidence and uncertainty emerge from interactions among evidence strength, neural noise, and recurrent connectivity. Across models, first-order decision performance proves substantially more robust than second-order metacognitive evaluation. These findings reveal mechanistic principles by which neural circuit dynamics can generate explicit subjective confidence and metacognitive monitoring. More broadly, they establish a computational framework linking neural dynamics, quantitative measures of metacognition, and conscious access to confidence, offering new insight into how the brain evaluates and communicates the reliability of its own decisions in uncertain environments.

↑ In the session list↑ In the timetable

58Rasmus Herlo

University of Copenhagen

TalkWednesday 14 October · 11:40–12:00 · Auditorium 3

Attention Self-modelling in Collaborative RL-Based Agents

Can machine attention be made inspectable, measurable, and experimentally manipulable? We present an interdisciplinary research program—the “AIntelope” project—that uses Attention Schema Theory (AST) as a bridge between consciousness science and reinforcement learning. AST reframes the hard question “is this system conscious?” into a tractable one: “does this system build and use a model of its own attention?” This shift allows us to translate AST’s first principles—recurrent information flow, working memory, explicit self-representation of attention, and predictive planning—into concrete architectural features for RL agents. Inspired by the evolutionary trajectory from simple stimulus-response organisms to attention-guided, planning animals (illustrated through an amoeba-to-antelope developmental ladder), we incrementally built agent architectures in custom gridworld environments: from a baseline DQN, through recurrent (DRQN) and steerable-attention (ROI-gated) variants, to a dual-head agent with action-awareness, separated gaze and body controllers, and disentangled reward routing for curiosity-driven attention. A dwell-time (“Linger”) mechanism was added to test perceptual masking analogues. We report comparative results across these architectural stages on a food-foraging task, highlighting trade-offs introduced at each step toward AST-compliance, and outline a custom-built orchestration toolkit enabling reproducible, configurable experiments across agents, scenarios, and analytics. We close with future directions—multi-agent scenarios probing theory of mind and affective cooperation, and hierarchical self-modelling via AST-inspired neural cellular automata.

↑ In the session list↑ In the timetable

59Veronika Viksna

University of Copenhagen

TalkWednesday 14 October · 12:00–12:20 · Auditorium 3

Can Language Models Know When They Are Wrong? Selective Escalation Using Self-Monitoring Signals.

LLMs are capable of solving a wide range of tasks; however, their use is associated with substantial computational costs. SLMs can reduce inference costs, but their reliability on complex tasks is often lower than that of larger models. This work investigates a selective escalation mechanism based on behavioural signals extracted during generation. These signals are assumed to reflect decision reliability and distinguish answers requiring escalation. A metacognitively inspired monitoring and control mechanism is proposed, in which a model uses self-monitoring signals to identify and selectively escalate unreliable cases to an LLM. The approach relies on signals capturing reasoning consistency, answer-type agreement, question complexity, reasoning structure, and data sufficiency. A hierarchical decision policy is employed, where some signals act as direct escalation triggers and others serve as supplementary indicators of potential error. Escalation decisions are made without labelled routing data, additional classifiers, or external confidence estimation models. The approach is evaluated on financial benchmarks: FinQA, TAT-QA, and ConvFinQA. The results show that the mechanism improves accuracy over the SLM baseline and outperforms random escalation at comparable escalation rates. The strongest predictors of answer correctness were reasoning consistency and answer-type agreement. The remaining signals contributed less but improved routing performance when used jointly. Computational efficiency depended on dataset characteristics and escalation frequency. The findings demonstrate that self-monitoring signals can identify unreliable answers, improving accuracy over the SLM-only baseline and approaching the performance of an LLM-only system, although computational savings remain dependent on the dataset and escalation behaviour.

↑ In the session list↑ In the timetable

60Maria Isabel Inzunza

KU Leuven

TalkWednesday 14 October · 12:20–12:40 · Auditorium 3

When AI Explains Itself, Should We Believe It?: Toward a Formal Theory of Genuine and Simulated Explanation

A machine learning model can produce a fluent, detailed explanation of its own behavior that has no structural relationship to the processes generating it. Current evaluation frameworks cannot detect this failure because they measure plausibility rather than structural fidelity, leaving both AI research and the sciences that rely on AI tools operating with representations that have the form of explanations without the content that would make them epistemically genuine. This talk introduces a formal framework for addressing that gap. Informed by phenomenological insights about the perspectival structure of epistemic access and operationalized through category-theoretic formalization, the framework defines explanatory adequacy as functor faithfulness on a horizon-relative subcategory of model transformations. An explanation is adequate when it preserves the relational invariants required for specific reasoning tasks relative to the investigative context, not when it maximally discloses internal structure, which is formally unavailable for complex systems. Three structural failure modes are identified and given precise categorical characterizations, of which simulation failure is the most consequential. The deeper motivation is this: any attempt to ascribe or deny phenomenal properties to this kind of system faces a fundamental impasse, since we are constrained to behavioral outputs rather than internal structure. That constraint is not dissoluble, but it is not equally binding under all conditions. By establishing formal criteria for when an explanation genuinely preserves reasoning-relevant structure rather than simulating its preservation, the framework offers a principled direction toward epistemic traction, the closest available approximation to certainty about systems whose internals remain inaccessible.

↑ In the session list↑ In the timetable

61Rony Hirschhorn

University of Sussex

TalkWednesday 14 October · 11:20–11:40 · Auditorium 4

Studying consciousness – for real?

How we study the contents of consciousness is dominated by the contrastive approach, comparing neural activity when an observer is aware versus unaware of the same stimulus. This tradition inherits strong assumptions: that a single mechanism underlies consciousness, that it operates alike across contents, and that one aspect, vision, is representative of experience as a whole. It also relies on putting participants in perceptual states that are very far from everyday or even altered-state experiences. This raises the measurement question of how far these can take us toward understanding experience. In this talk, I propose a taxonomy of three routes to reduce the gap between the lab and experience as it unfolds outside it: iterating between controlled and naturalistic settings; pushing the lab to better resemble real-life instances; and cognitive-ethology-style designs that begin with everyday experience. I then illustrate how each of these is applied in consciousness research.

↑ In the session list↑ In the timetable

62Dmitri Filimonov

University of Turku

TalkWednesday 14 October · 11:40–12:00 · Auditorium 4

New evidence and challenges in ERP and MEG correlates of consciousness in vision

The past twenty years of research have revealed two event-related potential (ERP) components to be the most reliably occurring neural correlates of consciousness (NCC) in vision: an early visual awareness negativity (VAN) in the N2 and late positivity (LP) in the P3 time window. Three previous extensive reviews concluded that VAN is a proper visual NCC, which is solely modulated by awareness. During the last five years since the latest review was published, a large body of new evidence has emerged about the ERP correlates of visual consciousness. In this systematized review we update the results of the previous reviews by analyzing new studies published since 2020 (N = 53) and discussing their findings. The new evidence is consistent with the earlier reviews: VAN is still found to be the most reliable and robust ERP NCC in vision, whereas LP reflects also many other processes, not consciousness as such. However, several aspects of VAN, for example, its relationship to attention and simultaneous physiological factors, require further investigation.

↑ In the session list↑ In the timetable

63Kacper Pete

University of Amsterdam / University of Cambridge

TalkWednesday 14 October · 12:00–12:20 · Auditorium 4

Non-linear integration and redundant representations in the thalamo-cortical system are associated with stable contents of consciousness

Despite theories of consciousness positing that the integration of thalamo-cortical and cortico-cortical interactions is crucial for consciousness to arise, empirical work on how they sustain stable conscious contents is limited. Moreover, we have recently proposed that large-scale integration, a hallmark of consciousness emphasized in several leading theories, depends on non-linear interactions that many classical metrics of connectivity may not fully capture. To address these gaps, we leveraged large-scale stereo-electroencephalographic (sEEG) recordings in awake human participants with epilepsy during an auditory bistable perception paradigm, in which a constant auditory stream was perceived as either integrated or differentiated. By combining complementary measures of complexity, connectivity, directed information transfer, and a novel method to isolate non-linear interactions, we examined the spatial and temporal characteristics of the thalamo-cortical interactions underlying conscious content. We present evidence suggesting that integrated contents of consciousness are associated with large-scale, temporally sustained integration that is non-linear in nature, spanning thalamo-cortical and cortico-cortical interactions, which are potentially bidirectional. We further characterized the representational mode linked to these interactions, revealing predominantly redundant encoding of conscious contents, suggesting that they are represented in an overlapping manner across thalamo-cortical networks. Together, these results help advance mechanistic accounts of how large-scale thalamo-cortical dynamics sustain stable contents of consciousness.

↑ In the session list↑ In the timetable

64Bilge Sayim

École Normale Supérieure, CNRS

TalkWednesday 14 October · 12:20–12:40 · Auditorium 4

Information compression, redundancy masking, and the limits of conscious human vision

The human brain operates under strong capacity limitations, requiring continuous selection and compression of incoming sensory information. A central question in consciousness research is how these constraints determine what is consciously accessed versus what is processed without awareness. While early visual processing encodes large amounts of information in parallel, conscious perception reflects a capacity-limited, selective, and compressed representation of sensory input. A major question in studying conscious and unconscious processing is whether a stimulus is consciously seen or not, often addressed using detection paradigms. Redundancy masking, by contrast, allows conscious and unconscious processing to be assessed without an explicit detection task by revealing systematic compression of visual information in peripheral vision. In particular, in redundancy masking, individual items in a stimulus consisting of repeated elements are often not consciously perceived. For example, when presented with three identical items, observers often perceive only two, indicating that redundant information is compressed in conscious perception when as few as three items are presented. Here, I argue that redundancy masking reflects a fundamental constraint on conscious access characterized by information compression of redundant sensory input. This provides a framework for investigating how capacity limitations shape the transformation from sensory encoding to conscious perception.

↑ In the session list↑ In the timetable

65Pedro Resende

Instituto Superior Técnico, University of Lisboa

TalkWednesday 14 October · 11:20–11:40 · Auditorium 5

On physics and the boundaries of consciousness science

The role of physics in consciousness science is widely misunderstood. Theoretical physics today consists of remarkable mathematical models, but it is neither complete nor fully consistent: general relativity and quantum theory remain incompatible after more than a century, and a notorious inconsistency, the cosmological constant problem, sits inside the very “Core Theory” (Wilczek) that governs everyday phenomena, including those in the brain. These unresolved gaps touch concepts of great relevance to consciousness: time, space, causality, locality, and the nature of reality. This matters for how consciousness is studied. Borrowing the logical notion of a conservative extension, we sort confirmatory approaches into three classes: conservativism, which assumes present science suffices; new physics, which modifies it; and idealism, which refounds it on consciousness. We partly follow Carroll’s analysis of the Core Theory to pinpoint where new physics may enter, including potential violations of locality in regimes where it has never been tested, such as the high complexity of brains and other biological systems. Confirmatory consciousness research faces deep difficulties: no current approach yet amounts to a mature scientific theory in the strict sense we define. Progress may therefore benefit from exploratory research, aimed at inducing theories rather than confirming them. For such research, physics literacy matters, so that anomalous findings are not prematurely dismissed by the fallacy of conflating “no known explanation” with “physically impossible.” We illustrate with a few examples where this distinction matters. This talk is based on joint work with Kobi Kremnitzer.

↑ In the session list↑ In the timetable

66Peter B. Lloyd

Independent

TalkWednesday 14 October · 11:40–12:00 · Auditorium 5

Interaction between a nonphysical mind and a physical brain (or machine) via quantum ‘portals’

The notion that consciousness collapses quantum superpositions has been around since the beginning of quantum mechanics, with advocates such as von Neumann and Wigner, but it never gained much traction because there was no specific theory of how it was supposed to work. (How does consciousness stand in relation to physical stuff? Where and when does collapse happen? What are the structure and dynamics of consciousness? What is it in consciousness that can trigger collapse?) What I will describe in this presentation is that, within an idealist framework, we can formulate a plausible model of consciousness collapsing a quantum superposition. (Idealism aka mental monism is a minority position but gathering some support.) I will examine the basic concepts of conscious collapse, how it can be reconciled with the Born rule, why this idealist approach might be preferred over McQueen & Chalmers’ property dualist approach, and why differentiating free will and randomness is a false dichotomy. Aspects of this portal theory also apply under substance dualism (such as Cucu’s), but idealism, apart from its philosophical foundation, offers a model of entanglement, which is otherwise a brute fact about the physical world under substance dualism. I will outline constraints on the ‘portal’ where this collapse occurs, and speculate on possible biological mechanisms that support it. The big issue here is portal navigation, which implies the portal must be a unique macromolecule (hence my conjecture is that portals are DNA, not microtubules). If time permits, I will discuss embodying conscious minds in machines.

↑ In the session list↑ In the timetable

67Robert Prentner

ShanghaiTech University

TalkWednesday 14 October · 12:00–12:20 · Auditorium 5

Unifying the Science of Consciousness: Mathematized Phenomenology and Trace Logic

This contribution sketches a model of mathematized phenomenology as the foundation for a non-materialist and unifying model for the science of consciousness. We argue that such an alternative framework is needed for four reasons: phenomenological adequacy (it respects the first-person structures of experience rather than reducing them to neural correlates), unifying scope (it provides a common formal language across competing theories), predictive power (it generates testable constraints on both phenomenology and neuroscience), and conceptual necessity (it addresses the explanatory gap left by materialist reductionism, which has failed to reconcile subjective experience with objective science). The model builds on prior work in which phenomenologies are represented by Markovian networks of state transitions between experiential states. These states, for example, the observational contents of seeing colors or more complex lived experiences, are further partitioned into accessible and inaccessible subsets, yielding trace chains. This probabilistic and topological structure illustrates how seemingly objective processes can be derived as emergent representations of underlying phenomenological relations. Finally, we propose how this transcendental framework can mediate between seemingly conflicting models of consciousness, including Integrated Information Theory, Predictive Processing, and Global Workspace Theory. These models clash because they implicitly rely on diverging conceptual profiles, operating on different assumptions about causal mechanisms and explanatory targets. By grounding them in an “experience-first” mathematical model, we aim to provide a unifying framework for consciousness science: one that is rigorous, non-materialist, and capable of reconciling phenomenology with neuroscience.

↑ In the session list↑ In the timetable

68Chetan Prakash

AMCS / BAMXi

TalkWednesday 14 October · 12:20–12:40 · Auditorium 5

Contextuality in Models of Consciousness: Trace Logics of Observation and Belief

Contextuality is a striking feature in observations of nature. Quantum theory, reflecting experiment, expresses this as “complementarity”and “non-locality.” Contextuality appears also in cognition: for example, when the order in which questions are asked can change answered outcomes. This situation can be made precise within the logic of observation: Outcomes within a given experimental context obey a Boolean description, yet across contexts these descriptions are mutually incompatible. Abstractly this corresponds to an undifferentiated wholeness being arbitrarily parsed into parts, each of which has locally Boolean structure, but for which the global whole is not Boolean. Any model purporting to explain physical or cognitive appearances, whether taking consciousness as fundamental or not, must, therefore, allow such phenomenological contextuality. In this talk we summarize ideas about pasting Boolean logics together into a non-Boolean structures — equivalently, the notion of “sheaf contextuality” — followed by applications to models of consciousness. In particular, we will introduce the mathematical structure of a phenomenological model of conscious observation, the “recursive trace logic,” and explore its relationship to contextuality.

↑ In the session list↑ In the timetable

72Julian Braunwarth

University of Göttingen

TalkThursday 15 October · 11:20–11:40 · Auditorium 2

Does Consciousness Arise from Scale-Invariant Properties?

In “The Conscious Mind”, David Chalmers asks the question: “If consciousness arises from the physical, in virtue of what sort of physical properties does it arise?” (Chalmers 1996, 247). He introduces two neural replacement scenarios—the fading and dancing qualia arguments—to narrow down the relevant properties in the physical supervenience base to organizational properties. However, this does not follow from his thought experiments, because they are construed in a way that also (implicitly) preserves the spatiotemporal scale of the relevant processes—a property that is independent of functional organization. This undermines his claim that it follows from his thought experiments that consciousness only depends on organizational—and thus scale-invariant—properties which remain unchanged when the system is uniformly enlarged or shrunk in space and time. Instead, I will argue that consciousness might arise from scale-sensitive properties—or at least partially supervenes on them. Physical properties and dynamics are often scale-sensitive: At the femtometer scale, for example, macroscopic properties such as viscosity or wetness simply do not exist—and quantum processes at the elementary-particle level obey fundamentally different laws than macroscopic ones. Shrinking or enlarging consciousness-related processes to a different level—while preserving functional organization—means replacing the underlying dynamics at level X with dynamics at level Y. Because the physical conditions at level X and level Y can be fundamentally different, the properties that constitute consciousness might be lost under scaling. Thus, consciousness could be inseparably tied to a particular spatiotemporal scale and to the laws that govern that scale.

↑ In the session list↑ In the timetable

73Jakub Mihalik

Czech Academy of Sciences

TalkThursday 15 October · 11:40–12:00 · Auditorium 2

Russellian Monism, Organizational Invariance, and the Possibility of Artificial Consciousness

A crucial support for the possibility of artificial consciousness is based on what Chalmers has called the ‘principle of organizational invariance’ (POI). If this principle holds, what matters for consciousness is the functional organization of the brain, and not its biological underpinnings. According to Russellian monism, a leading non-reductive approach to consciousness, on the other hand, more than functional organization matters for consciousness – the intrinsic nature of fundamental physical entities is crucial too. I investigate whether this prima facie tension between the POI, and Russellian monism is a symptom of their being incompatible. Miguel Ángel Sebastián argued that they are indeed incompatible, hence Russellian monists must reject the POI. Sebastián considers a subject S who had her neurons replaced with functionally equivalent silicon-based chips, explaining that one of these possibilities must obtain: (1) S’s experience disappears, (2) S’s experience doesn’t change, (3) S’s experience qualitatively changes. Sebastián argues that all these options mean that the POI and Russellian monism are incompatible. I argue that there are two ways for Russellians to respond to Sebastián’s argument: Some Russellians, e.g., proponents of panqualityism (Coleman, Feigl), have good independent reason to weaken the POI and hold that S after the replacement, and other isomorphic systems, would be conscious, although their consciousness may qualitatively differ (option 3). Furthermore, I’ll explain that ‘Russellian physicalists’ (Montero, Brown) can endorse a sort of multiple realizability of organism-level consciousness, opting for Sebastián’s possibility (2). Artificial consciousness, then, looks like a live possibility for Russellian monists.

↑ In the session list↑ In the timetable

74Adel Chaibi & Aida Elamrani

Intel / Ghent University

with Eric Petit

TalkThursday 15 October · 12:00–12:20 · Auditorium 2

Form Under Constraint: Between Substrate and Abstraction

Recent debates on computational models of consciousness and digital minds have reignited the classic division between reductive materialism and non-reductive functionalism. On one view, phenomenal experience is strongly substrate-dependent; on the other, it depends only on the implemented abstraction. We seek to develop a more balanced analysis of this problem. We start from the uncontroversial premise that both biological and digital systems process information. And whether characterised as vehicles or as content, physical and abstract bits appear correlative rather than opposed. At its root, informatio denotes the giving of shape, the giving of form. Correspondingly, we propose an intermediate explanatory target: the level of forms, jointly constrained by matter and function. We build our case from the evolution of AI systems. Software continuously exerts pressure on hardware development to meet abstract performance objectives; conversely, hardware challenges algorithmic design to optimise under material constraints. We argue that hardware and software do not develop passively or independently; rather, they co-evolve and co-constrain a shared space of computational forms. We therefore use form to name the structural invariants through which material systems and functional organisations become scientifically tractable together. This is not intended as a new theory of consciousness, nor as an ontological commitment. Our claim is more modest: form is an epistemological concept, marking the level at which the scientific study of consciousness can compare biological, artificial, and hybrid systems without prematurely reducing consciousness either to substrate or to abstraction.

↑ In the session list↑ In the timetable

75Christoffer Lundbak Olesen

Aarhus University

TalkThursday 15 October · 11:20–11:40 · Auditorium 3

Towards Mapping the Space of Experience: Diversity as a Constraint on Theories of Consciousness

In consciousness science, we need to get comfortable with the indubitable fact that our direct and real intimacy with subjective experience doesn’t merit generalisation. Without knowing how consciousness differs, we cannot claim we know what it is, and consequently, we cannot objectively evaluate theory. Mental imagery ranges from complete absence to vivid richness. Some have an inner voice while others do not. People experience their bodies very differently, and even the feeling of time is not universal. These are variations in the character of conscious experience that don’t surface through normal human interaction, and due to this latent nature, we simply do not know how deep these differences go. In this talk, I’ll argue that we all rely on some sense of familiarity with consciousness when developing models and methods within consciousness science and that this may introduce biases guiding diverging theoretical positions. In addition to testing theories constructed to explain consciousness, we should focus on exploring the extent of what such theories need to explain in the first place, i.e., investigate the diversity of experience, while being open to the possibility that consciousness may not be the phenomenon we each presume. I’ll argue that mapping this diversity is essential for an objective science of consciousness, and that it allows for constraint-based search in the space of possible theories. This calls for new methodologies aimed at exploring such differences and invites researchers to think more openly about consciousness with a curiosity towards otherwise unrecognised forms of experience.

↑ In the session list↑ In the timetable

76Wanja Wiese

University of Bonn

TalkThursday 15 October · 11:40–12:00 · Auditorium 3

Impossible Conscious Experiences

Are there conceivable conscious experiences that no human being could have? In one sense, this may be trivial, if we consider how our perceptual experiences are limited by our sensory organs. But there may be non-trivial types of impossible conscious experience, and investigating such ‘impossibilities’ may be facilitated by mathematical models that focus on the structure of experience. This talk clarifies the idea of impossible conscious experiences, by distinguishing between different senses in which conscious experiences can be regarded as impossible, as well as by highlighting the connection to structural approaches to consciousness. The overall idea that emerges is that by getting a handle on the impossible, we gain a better understanding of actual consciousness.

↑ In the session list↑ In the timetable

77Simone Anthes

Monash University

TalkThursday 15 October · 12:00–12:20 · Auditorium 3

What we talk about when we talk about dreaming

What is a dream? We all think we know what we are talking about when we talk about dreaming. But it turns out we mean different things. This, coupled with the oft-mentioned multiplicity of dreaming and the different disciplines dedicated to its investigation, is reflected in a varied representation of dreaming in the literature. The idea of a pre-defined phenomenon “dreaming” is one of the major conceptual obstacles in dream research and is at the heart of the discussion on integrating dream research into the broader context of consciousness science. However, in actual dream research, there is no such clearly defined phenomenon. Over the last decades, many theories, hypotheses and models about the purpose and character of dreaming have emerged from the various disciplines involved in dream research. Yet, they often fail at explicitly defining their explanatory target. Here, we survey contemporary dream theories and demonstrate the variety in how they conceptualise dreaming. Acknowledging that there has been nonetheless massive progress in dream research, we ask whether a definition-led approach is even necessary and present several alternative approaches. We conclude that while it is necessary to be transparent about perspective and preconceptions on dreaming, a unified definition may not be the prerequisite to advancing dream research that it is often framed to be.

↑ In the session list↑ In the timetable

78Olha Sobetska

Vrije Universiteit Brussel

TalkThursday 15 October · 11:20–11:40 · Auditorium 4

The Dynamics of Creative Cognition

Creativity is often described through separate perspectives, including psychology, neuroscience, and computational models. While each of these explains part of the phenomenon, they do not yet provide a common framework for understanding how a creative idea emerges in real time. This talk proposes a synergetic perspective in which creativity is viewed as a process of self-organization, supported by order parameters and phase transitions. In this way, a creative moment can be understood as the emergence of a dominant pattern that temporarily organizes neural and cognitive activity around a new idea. When an existing organisation (a temporarily stable idea) becomes unstable or ceases to correspond to the current context (the task’s constraints), a new coherent structure may emerge, guiding thoughts towards a creative outcome. This approach differs from one that views creativity as a ‘black box’ or a purely random event, it posits that creativity arises as a result of the interaction of a multitude of cognitive micro- and meso-processes. Drawing on findings from neuroscience together with concepts from synergetics, the talk presents a dynamical way of understanding creativity without reducing conscious experience to a fixed sequence of computational steps.

↑ In the session list↑ In the timetable

79Jeanne Catherine-Gray

University of Virginia, Contemplative Science Center

TalkThursday 15 October · 11:40–12:00 · Auditorium 4

Insightful Shifts in Consciousness Among Men in Prison: A Micro-Phenomenological Investigation

Insight, often described as an “aha!” moment, entails a distinct phenomenology that differs from other problem-solving processes. Yet little is known about the experiential structure of sudden, transformative insights that produce shifts in ordinary waking consciousness. This study presents a fine-grained micro-phenomenological analysis of such insightful change; a sudden moment of psychological transformation as described by eight men incarcerated in a UK prison following exposure to the Insight to Wellbeing program. Analysis revealed a shared experiential structure of insight comprising four phases and a transition event: (0) understanding, (1) recognition, (realization event), (2) opening, and (3) connection. Insight involved a shift from habitual, constricted states of mind toward openness and calm, accompanied by emotional relief and an increased sense of agency. These changes occurred without a loss of self. The findings support understanding insight as a mechanism that can reconfigure one’s state of mind and further demonstrates the value of micro-phenomenological methods for studying particular shifts in consciousness.

↑ In the session list↑ In the timetable

80Madalina Wernigg

University of Zürich

TalkThursday 15 October · 12:00–12:20 · Auditorium 4

The Instability of the Epistemic Split: Bipolar Disorder as a Clinical Case of Contextual Emergence and Thermodynamic Desynchronization within Dual-Aspect Monism

Operating under the postulate of consciousness as fundamental ontic reality, this paper rejects neuro-reductionism by defining the brain as an evolutionary filter. Within dual-aspect monism, the central nervous system acts as an operational symmetry breaker. Through a stable macro-context (C_healthy), it enforces a stable partition of the psychophysically neutral unus mundus, creating the regular epistemic split necessary for survival [Atmanspacher 2012]. This framework proposes the transient destabilization of this epistemic interface during acute affective states, conceptualized as a thermodynamic desynchronization, wherein altered energetic constraints disrupt the fundamental time-structure of subjective experience. This transition is driven by a hierarchical cascade: interpersonal desynchronization triggers a vertical circadian mismatch between central and peripheral pacemakers, manifesting as a temporal desynchronization of the lived body [Fuchs 2013], while disrupting mitochondrial bioenergetics, intracellular calcium, and the homeostatic sleep-reset [Borbély 1982]. Consequently, the coarse-grained partition of the state space collapses into opposing structural instabilities [Atmanspacher 2012]. Mania (C_manic) manifests phenomenologically as hyper-salience, an unwillful exposure to non-local correlations, and a profound dissolution of the ego- environmental boundary (under-coarse-graining). Conversely, depression (C_depressive) induces a severe temporal uncoupling, rigidly isolating the subject from meaningful holistic correlations through an agonizing stagnation of inner time (over-coarse-graining). Crucially, modeled through contextual emergence, the system exhibits dynamical reversibility based on its intrinsic resilience: by re-introducing external somatic and temporal constraints via nutrition, sleep, and stabilizing pharmacological agents, the dissipative system returns toward its lower-entropy attractor state, restoring the complementary epistemic split.

↑ In the session list↑ In the timetable

81Nicolas Loerbroks

Independent

TalkThursday 15 October · 11:20–11:40 · Auditorium 5

The Hard Problem Presupposes Mechanistic Exhaustiveness

In formulating the “hard problem”, Chalmers explicitly presupposes that once we specify either the relevant neural mechanisms or their computational organization, “a full and satisfying explanation will result” (Chalmers, 1995, p.202) for cognitive phenomena – a presupposition implicit in his reformulation of computational functionalism in response to the triviality concerns raised in Putnam’s later work. The hard problem arises by contrast: conscious experience is taken to fall outside what mechanistic or structural explanations can deliver, whereas such explanations succeed for cognition. Putnam’s original machine functionalism did not require this asymmetry. In characterizing pain as a multiply realizable functional state, he rejected type-identity theories together with the demand that the objectivity of such computational states be grounded in a uniquely specifiable physical causal-organizational structure. The asymmetry emerges, in the form it takes within the computational functionalist tradition, only when computational explanation is reinterpreted as complete in capturing the relevant causal or organizational structure of a system. This diagnosis further reframes the dispute between Dennett and Chalmers as a disagreement over explanatory adequacy rather than over the ontological status of phenomenal experience: reading Dennett’s real-pattern account as an intermediary between Putnam and Chalmers – anti-exhaustivist while preserving causal closure – shows that it is Chalmers, not Dennett, who occupies the revisionary extreme, since only his reformulation installs the exhaustiveness standard that makes consciousness exceptional.

↑ In the session list↑ In the timetable

82William Seager

University of Toronto

TalkThursday 15 October · 11:40–12:00 · Auditorium 5

Is the Metaphysics of Predictive Processing Self-Refuting?

Predictive processing theory (PPT) holds that the core function of the brain and nervous system is to facilitate successful behaviour by minimizing errors between predicted and incoming signals. PPT operates hierarchically across levels, from endogenous neural processes all the way to sensory signals from the world, always computing discrepancies between a predictive model and incoming data. PPT is mathematically rich, describing signal and prediction in terms of probability distributions, with error minimization achieved via approximate Bayesian updating. There are promising proposals for its neural implementation. However, philosophical interpretation of the theory is less well developed. Integrating consciousness into PPT raises the question of where experience resides: input or model? The answer seems clear. The home of conscious experience is the predictive model. We experience the world as we then take it to be, awaiting incoming corrections or confirmations. This is why in PPT experience is sloganized as ‘controlled hallucination’ or ‘neuronal fantasy’. Arguably, this engenders a paradox. PPT takes the brain as given, its neural architecture running recondite computations over probability distributions. Yet the brain is, ultimately, known via perception. It is part of the hallucination. But, it is then argued, ‘an hallucinated brain cannot explain the hallucination of a brain’. Thus, PPT appears ungrounded; the so-called neural basis of PPT becomes just a tail-biting ‘physicalist fantasy’. There is something right about this worry, but something wrong as well. Assessing it leads to deep questions about both science and reality, and their relation to consciousness.

↑ In the session list↑ In the timetable

83Michael Schnabel

Vanderbilt University

TalkThursday 15 October · 12:00–12:20 · Auditorium 5

Modeling Consciousness: On the Limits of Theory and Conceptual Understanding

LLMs are trained on statistical regularities in language and reasoning, while cognitive science models cognition in terms of predictive coding, information processing, and active inference. Implicit in the scientific approach is the assumption that increasingly refined models will yield a deeper understanding of both nature and mind. This talk questions that assumption with respect to consciousness. Drawing on the Buddhist principle of dependent origination and the Yogācāra analysis of cognition, I argue that scientific attempts to “explain consciousness” may inadvertently reinforce the very dualistic framework they seek to overcome by reifying the subject-object distinction. Paradoxically, the pursuit of ever-greater predictive power might obscure what it ultimately seeks to explain. As an alternative to the scientific emphasis on modelling and conceptual understanding, I turn to the Chan/Zen practice of investigating mind through direct contemplation. Rather than offering another theory of consciousness, this reorientation shifts the focus from explanation to exploration. It invites a prior question: whether the modelling stance itself, together with the quest for conceptual understanding, constitutes the very obstacle to understanding consciousness – and what possibilities for inquiry emerge when that stance is reconsidered.

↑ In the session list↑ In the timetable

85Antonina Skrypnyk

Taras Shevchenko National University of Kyiv

TalkThursday 15 October · 16:30–16:50 · Auditorium 2

The Instant Phenomenological Field as a Fundamental Unit of Conscious Experience

What constitutes the elementary unit of conscious experience? Contemporary theories of consciousness typically focus either on neural mechanisms or on the contents of awareness, while the structure of experience itself often remains insufficiently explored. This presentation proposes a theoretical framework based on the concept of the Instant Individual Phenomenological Field (IIPF). The model begins with a distinction between a physical instant and a psychological moment. Unlike a dimensionless point in physical time, a psychological moment is a finite interval experienced as a unified whole. Conscious life can therefore be understood as a succession of instant phenomenological fields whose continuity constitutes the Extended Individual Phenomenological Field. Within this framework, phenomenological fields are interpreted as the ideal aspect of the subject, analytically distinguishable but inseparable from its material aspect. Perception, memory, imagination, and thinking are examined as different modes through which conscious experience accesses material structures. This approach offers a novel perspective on the mind–body problem by replacing the traditional opposition between mental and material substances with the notion of complementary aspects of a single reality. Particular attention is devoted to the internal organization of the IIPF. The field is conceived as a dynamic structure centered on attentional focus and composed of elements that differ in their degree of awareness. A distinction is proposed between a central zone of focal awareness, a broader zone of conscious contents, and a peripheral zone of subconscious processes. Rather than constituting a conventional geometric or topological space, the IIPF is modeled as a multidimensional network of

↑ In the session list↑ In the timetable

86Hanqi Wang

Carnegie Mellon University

TalkThursday 15 October · 16:50–17:10 · Auditorium 2

Higher-Order Representation as Cross-Timescale Integration in Time-Consciousness

Temporality is a fundamental feature of conscious experience and has been proposed as a candidate for a minimal unifying model of consciousness (Windt 2015; Wiese 2020). Singhal and Srinivasan (2021) provided a three-level framework for the phenomenology of time experience, including a fast cinematic level (30–50ms), an intermediate extensional level (300–500ms), and a slow retentional level (3–5s), detailing how these levels interact across timescales. However, the framework does not explain why content at the intermediate level becomes conscious while content at the fast level does not. We argue that this gap can be filled by the HOROR (Higher-order Representation of a Representation) theory proposed by Brown (2015), which holds that phenomenal consciousness consists in instantiating a certain kind of higher-order representation. The fast-updating Level 1 content constitutes an unconscious first-order representation, which is integrated by Level 2 over its ~300–500ms window. Crucially, Level 2 gains its higher-order character through cross-timescale integration by unifying multiple Level 1 slices, yielding content categorically distinct from any single static slice. By unifying first-order slices under a single experiential perspective, this integration supplies the minimal self-attributive structure HOROR requires without presupposing a conceptual-propositional judgment. This integration contributes to both phenomenological and higher-order frameworks. It provides the former with a mechanism for why content becomes conscious, while showing that the higher-order representation is inherently temporally extended. It further yields differential predictions for paradigms such as continuous flash suppression and the attentional blink, which selectively disrupts integration at distinct timescales and should therefore dissociate the proposed levels.

↑ In the session list↑ In the timetable

87Kohei Fukao

Hokkaido University

TalkThursday 15 October · 17:10–17:30 · Auditorium 2

How Are Retention and Protention Extended in Sign-Mediated Counting?

This talk aims to explicate a phenomenological time structure that is essential to the conscious activity of counting. In the well-habituated, rule-governed activity of counting, we usually use number signs. As we are often too well-habituated to the use of number signs, it is not so simple a task to explicate what we do when we count using them. Husserl addressed a pertinent issue: in his Philosophie der Arithmetik, he made it explicit that a certain time structure (namely, succession and simultaneity) is indispensable for the counting activity. In the work, he primarily focused on the relationship between number and time in what he called ‘authentic representation’ of numbers. Thus, Husserl was not explicit about the time structure of sign-mediated numerical experience. Lohmar, in his Phänomenologie der Mathematik, discussed the number-time relationship in what Husserl called ‘symbolic representation’ of numbers, while also considering the relationship in light of the phenomenological account of time consciousness. In this talk, I attempt to further the phenomenological theory of time consciousness that essentially pertains to the activity of counting by addressing the following two: what I call extended retention/protention and how they are possible by virtue of number signs. I conclude that the proposed account of extended retention/protention not only advances the phenomenological account of the number-time relationship in sign-mediated counting activity, but can also serve as a phenomenological foundation for modelling such basic numerical experiences as counting.

↑ In the session list↑ In the timetable

88Donald Borrett

University of Toronto

TalkThursday 15 October · 17:30–17:50 · Auditorium 2

A temporality model of flesh and dual-aspect monism

Merleau-Ponty proposed an ontology of flesh in which the phenomenal field is characterized by an intertwined and reversible interplay of the visible and invisible that serves as the basis of meaningful experience. A temporality model of flesh is proposed in which the visible is the flux of sensible experience and the invisible is time horizons that are generated by the flux and feedback onto it to establish a self-adjusting, iterative system in which past, present and future are equally implicated in its evolution. Because it is a dynamical process, it can be mathematized and universality criteria established in all systems that have this form. By necessity, all such systems have two, and only two, dynamical signatures that are isomorphic to pre-reflective and reflective experience. Within the latter are dynamical signatures isomorphic to mind (determinate invisible) and world (determinate visible). A Logos of flesh can be identified that represents the quantifiable dynamics of a monistic ontology that grounds its self-interpretation both as mind and as world. Although they lack causal interaction, mind and world share a temporal characterization because of the universal dynamics that represents the condition for their possibility.

↑ In the session list↑ In the timetable

89Sucharit Katyal

University of Copenhagen

TalkThursday 15 October · 16:30–16:50 · Auditorium 3

What Is Consciousness? And What Does a Theory of Consciousness Need to Explain?

Despite major advances in consciousness science, theories still disagree about their target: What is consciousness, and what must a theory of it explain? I argue that this disagreement follows from an unresolved structural problem. Recent structural approaches model consciousness by mapping relations among phenomenal qualities—especially the structure of perceptual contents or qualia spaces—to corresponding neural or mathematical structures. This has clarified how contents may be organised, but leaves aside the wider structure of consciousness as such: the relations among content, agency, and minimal subjectivity. This talk proposes a comprehensive structural model of consciousness. Drawing on Husserlian phenomenology and Indian contemplative philosophy—especially P. R. Sarkar’s analysis of subjectivity in relation to advanced meditative states—I distinguish three intermutable phenomenological components: mental object, doer-I, and existential-I. Consciousness is conceived as a phenomenal resource expressed through these components in varying degrees. Different conscious states—ordinary perception, agency, dreaming, meditation, ego dissolution, nondual awareness, and minimal experience—reflect different distributions and transformations of this resource. The proposal extends structuralism beyond phenomenal content. Rather than treating structure as primarily internal to qualia or perceptual fields, it treats the whole conscious episode as structurally articulated by objecthood, directedness/doership, and existence. This provides a theory-neutral explanandum for consciousness science: a theory of consciousness must explain not only how contents are structured, but how phenomenal resource is distributed across and transmuted between its three structural components. I conclude by outlining implications for operationalisation, naturalisation, comparative differences across species and individuals, and development.

↑ In the session list↑ In the timetable

90Moritz Kriegleder

University of Vienna, Czech Academy of Sciences

with Wiktor Lachowski, Tomas Marvan

TalkThursday 15 October · 16:50–17:10 · Auditorium 3

Adapting the Matched-Interlevel-Experiments Framework for Mechanisms of Consciousness

Recent work on the neural correlates of consciousness has increasingly turned to interventionist methodologies developed in the philosophy of science. One particularly notable approach in this vein is Craver’s “new mechanist” account of causal explanation. Yet it remains unclear whether mechanistic criteria designed for paradigms like memory or motor behavior can be coherently extended to phenomenal consciousness. This talk examines that extension by applying the Matched-Interlevel-Experiments framework (MIE) to mechanisms of consciousness. MIE consists of four conditions to identify the constitutive parts of the mechanism, viz. bottom-up inhibition, bottom-up excitation, top-down activation, and cross-experimental matching. We argue that while MIE is well suited to reveal important details of neurobiological mechanisms, it runs into a distinctive set of problems in consciousness research: the difficulty of disentangling genuine mechanism responsible for consciousness from what are merely causal prerequisites of it, the privacy of experience and the relations between lower and higher “levels” of brain description. We propose tentative strategies for addressing these challenges and making the MIE framework a productive tool for consciousness researchers. We focus on a concrete candidate mechanism: apical processes in layer-5 pyramidal neurons pivotal for low-level, small-scale theories of conscious perception such as Apical Amplification and Dendritic Integration theories. We close by considering whether this framework generalizes to higher-level, network-scale theories of consciousness, or whether it remains tied to cellular mechanisms with unusually tractable points of intervention.

↑ In the session list↑ In the timetable

91Tonglin Yan

Paris-Saclay University

TalkThursday 15 October · 17:10–17:30 · Auditorium 3

Formalizing Consciousness: A Comparative Mathematical Analysis of Leading Theories

Systematic comparison across consciousness theories has been hindered by the absence of a shared formal language. We reformulate four mathematically articulated frameworks, namely Integrated Information Theory (IIT), Active Inference (AIF), Higher-Order Representation approaches (HOR/HOSS), and the Projective Consciousness Model (PCM), within a common probabilistic and geometric idiom organized following Marr’s three levels of analysis. Each framework is rendered in terms of stochastic kernels, divergence measures, variational free energy, and group-structured state spaces. IIT is characterized through partition-based cut operations and irreducibility measures. AIF is presented as a general process theory for perception and action under uncertainty; HOR/HOSS extends this as hierarchical metacognitive inference within factorized generative models. PCM is cast as a group-structured POMDP in which the Field of Consciousness is organized by projective transformations under expected free energy minimization. This formal treatment reveals that several apparent theoretical conflicts dissolve upon alignment of explanatory level. IIT specifies the integrative property a system must possess; AIF and HOR concern the inferential mechanisms of conscious access; PCM contributes a formal account of perspectival structure. Genuine tensions persist around identity claims, first-person geometry, and computational tractability. As a prospective direction, we outline a hierarchical POMDP architecture as a potential computational testbed in which integration, higher-order access, and perspectival control could be evaluated within the same model, supporting more systematic empirical comparison among theories.

↑ In the session list↑ In the timetable

92Krzysztof Dolega

Ruhr-Universität Bochum

TalkThursday 15 October · 17:30–17:50 · Auditorium 3

We Don’t Need Another Model: Towards a Meta-Theoretic Framework for Comparing Theories of Consciousness

After roughly three decades of methodological advances and genuine empirical breakthroughs, the science of consciousness now faces a problem of theory proliferation. The number of competing models is estimated at somewhere between twenty (Seth & Bayne, 2022) and two hundred (Kuhn, 2024), with no widely accepted standard for adjudicating between them (Doerig et al., 2021). Comparative efforts, such as adversarial collaborations (Melloni et al., 2023), remain piecemeal and have so far proven inconclusive (Lau, 2023; Mudrik et al., 2025). Meanwhile, the crisis is compounded by systematic confirmation bias, as the choice of methods reliably predicts which theory is being tested (Yaron et al., 2022). I argue that resolving the issue of adjudicating between competing approaches can only be achieved by building a meta-theoretic framework for theory comparison: a common structure for evaluating competing models against the theoretical virtues long recognized in philosophy of science (Kuhn, 1977; Keas, 2018). Although building such an instrument will likely take time and a joint effort of multiple researchers, I present and defend a blueprint for the general shape it should take. In particular, I motivate the set of virtues to be included (including, but not limited to, predictive accuracy, simplicity, and explanatory scope) and discuss how they might be formalized, weighed, and traded off transparently. I close by defending this virtue-based approach against competing proposals, such as the Measure Centrality Index (Chis-Cuire et al., 2024) and the search for a Minimal Unifying Model of consciousness (Wiese, 2020).

↑ In the session list↑ In the timetable

93Jaya Caporusso

Jožef Stefan Institute

TalkThursday 15 October · 16:30–16:50 · Auditorium 4

Dissolution Experiences and the Experience of Self and World: An Empirical Phenomenological Investigation

Dissolution experiences (DEs) are experiential episodes during which the perceived boundaries between Self and World become fainter or less clear; this is usually accompanied by a sense of unity with the World or elements of it. We investigate the experiential structure of DEs by comparing them with daily-life experiences (dlEs), focusing on how Self and World are experienced. We conducted 29 phenomenological interviews with eight participants, covering eight dlEs and 10 DEs occurring in different contexts (meditation, nature, art, psychedelic experiences, intimacy, and psychopathological episodes). Our findings show a progressive transformation in the structure of Self–World experience. While in dlEs the Self is experienced as a complex multi-element entity, in DEs it usually undergoes a loss of internal coherence, and most of its elements disappear; often, only a sense of mineness remains. At the same time, the World progressively appears as more open, friendly, safe, familiar, vivid, spatially infinite, and internally unified. As this unfolds, processes involving Self and World become increasingly difficult to distinguish from one another, mirroring a blurred subject/object distinction. A central feature of almost all analysed DEs consists in trusting the World and letting go of control: participants approach the World with trust and/or recognise it as trustworthy, while relinquishing or relaxing control, often with a sense that it is safe to do so. Our study contributes to a conceptual and phenomenological understanding of DEs and provides insights relevant to discussions on the metaphysical and phenomenological nature of the Self.

↑ In the session list↑ In the timetable

94Kiana Ward

Radboud University

TalkThursday 15 October · 16:50–17:10 · Auditorium 4

Modeling non-dual awareness via constraint closure: a reinterpretation of groundlessness

Non-dual awareness (NDA) refers to a shift in consciousness in which the usual distinction between subject and object dissolves, and experience is no longer structured by conceptual mediation or goal-directed regulation. Meling enactivist model describes NDA as a meditative disclosure of groundlessness—the recognition of emptiness (śūnyatā), that all phenomena lack intrinsic nature. While enactivism explains autonomy through process closure, this article argues that constraint closure, as developed by Nave, extends that framework by making explicit how autonomy is sustained through the continual regeneration of its own relational conditions. This refinement prevents process-closure models from being read in substantialist terms when applied to complex cognitive systems, where stability arises through ongoing transformation rather than fixed organization. Nave’s account builds on Juarrero theory of constraint causality, which replaces intrinsic forces with relational conditions—a view that parallels Nāgārjuna’s Madhyamaka analysis of dependent origination. Integrating these perspectives, I propose that NDA corresponds to a shift from decoupled to precarious constraints, revealing that cognition and awareness persist not through intrinsic foundations but through the dynamic regeneration of interdependent relations.

↑ In the session list↑ In the timetable

95Niffe Hermansson

Trace Institute

TalkThursday 15 October · 17:10–17:30 · Auditorium 4

Traces of the Other

Encounters with apparently autonomous entities are among the most striking and recurrent features of high-dose DMT experience. In both naturalistic and lab settings, many users report that these encounters seem to involve beings with autonomy, intelligence, and agency, rather than mere imagery or hallucinations. This raises the question: can these claims be scientifically explored without committing to, or presupposing, an atypical metaphysics? We suggest that one useful avenue for exploring these claims is provided by Conscious Agent Theory within the Conscious Realism framework, a consciousness-first formalism in which reality is modeled as consisting exclusively of a network of interacting conscious agents. In this account, conscious agents are modeled in terms of measurable spaces of possible experiences and Markov-kernels governing probabilistic transitions between experiential states. To say that one agent perceives another is to say that the dynamics of the latter influence the updates of the former’s experience, mediated by the relevant kernels. Using this formulation, questions about DMT entities can therefore be decomposed into precise questions about the structure, possibilities, and constraints of experiential dynamics. Importantly, the familiar distinction between true perception and hallucination is not basic in this formulation. In relation to DMT entities, this is a strength: rather than having to ask blunt questions about whether these entities are “real” or “hallucinated,” it allows us to ask whether entity encounters exhibit identifiable transition structures, invariants, temporal stability, responsiveness to hidden variables, or nontrivial intersubjective correlations. We use this decomposition to formulate a research program in which these features

↑ In the session list↑ In the timetable

96Andres Gomez Emilsson

Qualia Research Institute

TalkThursday 15 October · 17:30–17:50 · Auditorium 4

Field Lines Before Equations: Parameterized Dynamic Phenomenology and the HEART Retreat

I’ll present Parameterized Dynamic Phenomenology (PDP), a methodological framework we’ve been developing at QRI, and situate it within HEART, the legal exploratory retreat we run where physicists, mathematicians, artists, and advanced meditators get together to reverse-engineer non-ordinary states of consciousness as a team. PDP has four components: (1) represent a sensory stimulus as a data structure, (2) apply a parameterized, mathematically-defined operation that modifies it dynamically, (3) translate the modified structure back into a representation of phenomenology, and (4) match parameters to specific states of consciousness. The key idea is that when a participant replicates their own phenomenology inside a generative model we fully understand, the parameters they land on give us much stronger evidence than verbal report can, because they pin down mathematical features actually present in the experience. At HEART we’ve been mixing a lot of approaches: psychophysics, Likert ratings, participant vibe-coded sequences, drawings, and replication tools like the Tracer Tool and the Oscilleditor. One direction we find promising is characterizing states in terms of coupling kernels – the way oscillators want to phase-relate to their neighbors – and asking whether different substances and meditative states leave distinguishable signatures there. This past retreat we ran a protocol comparing formless jhana states with high-dose 5-MeO-DMT; analysis is still underway, so I’ll present the design, the questions, and some of the data analysis rather than the full results.

↑ In the session list↑ In the timetable

97Tobias Schlicht

Ruhr-University Bochum

TalkThursday 15 October · 16:30–16:50 · Auditorium 5

Modeling the Ghost in the Machine

Debates on AI consciousness suggest that we can, assuming computational functionalism, develop computational markers to test whether an artificial system is conscious. Participants clarify that they mean phenomenal consciousness, in Nagel’s (1974) sense that there is something it is like to be a system or undergo a subjective experience (e.g., Butlin et al. 2025, Birch 2025, Chalmers 2023, Schneider et al. 2025, Kanai & Ma 2026). I argue that this notion is unfit to serve as a target for modeling consciousness in AI. I present two ways it has been characterized. Block (1995) introduced it as a kind of consciousness distinct from access consciousness, but subsequently distinguished (the capacity of) experienced content from cognitively accessible content (Block 2007), first suggesting that it could be investigated empirically, but eventually being discontent with the most ingenious methods like no-report paradigms (Tsuchiya et al. 2014), asking for no-cognition paradigms instead (Block 2019). This requirement makes it methodologically inaccessible in AI. Chalmers (1996) separated phenomenal consciousness as the hard problem from cognitive functions associated with consciousness, the easy problems: it’s non-physical, non-functional, and epiphenomenal, which makes it methodologically inaccessible via third-person methods. Any markers possibly justifying attribution of phenomenal consciousness are either absent in AI (neural and first-person markers), ambiguous (behaviors), or question-begging (computations). Absent other characterizations of phenomenal consciousness, ironically, the moral is that attempting to model and assess phenomenal consciousness in AI, illusionism about phenomenal consciousness (Dehaene 2014, Dennett 1991) is advisable. Otherwise, researchers are chasing a ghost in the machine.

↑ In the session list↑ In the timetable

98Angela Johnson

Northeastern University

TalkThursday 15 October · 16:50–17:10 · Auditorium 5

From Indicators to Dynamics: Metastability, Recurrence, and Ignition in a Persistent Agent’s Attention Stream

Recent work derives indicators of consciousness for AI systems from neuroscientific theories (Butlin et al., 2023), but these are largely architectural checklists; dynamical measures computed on a running agent’s own trajectory remain rare. We present a novel framework (AURORA) that treats the turn-by-turn attention dynamics of a persistent-memory agent as an observable state stream, wherein at each step the agent scores memory candidates and selects a subset to attend to, optionally cross-referenced with its affective and desire state. On this stream we compute measures imported from consciousness science. Metastability is the temporal variance of the Kuramoto order parameter over the selection signal (Tognoli and Kelso, 2014), indexing integration-differentiation balance. Recurrence quantification analysis, via determinism and laminarity, characterizes temporal structure and rigidity (Marwan et al., 2007). A perturbational fracturability index, the behavioral analog of the perturbational complexity index (Casali et al., 2013), injects a controlled perturbation and scores the Lempel-Ziv complexity of the agent’s response trajectory. An ignition criterion flags global-access events in the spirit of the global workspace (Mashour et al., 2020). We borrow these measures’ human validation as rationale for their utility in examining synthetic human-like cognitive states, while scoping our claim to cognitive dynamics, such as rigidity, adaptation, and collapse, rather than phenomenal experience. We report preliminary longitudinal data from continuous runs and contrast a healthy trajectory with a pathological collapse identified independently. Persistent agents, whose internal attention is fully observable, are a uniquely controllable system for validating and stress-testing computational measures of cognitive state.

↑ In the session list↑ In the timetable

99Jessica White

Sentient Futures

TalkThursday 15 October · 17:10–17:30 · Auditorium 5

Locating disagreement in AI consciousness research: literature review and qualitative study.

Experts disagree about AI consciousness, and what could follow morally. Their disagreement can be mapped across multiple levels: conceptual, methodological, theoretical, metaphysical and normative. We analyse these areas of dispute in two stages. The first stage consists of a systematic review of work on AI consciousness and moral consideration, with published arguments classified into subcategories, marking active areas of debate within the field. For the second stage, we conducted a series of qualitative interviews with researchers and non-research experts across consciousness science, philosophy of mind and AI. We document how they reason in practice, beyond their published positions, and test whether their reasoning tracks the cruxes identified in the review or reveals new ones within the scope of our study. We asked participants what evidence would convince them that an AI system is conscious, and whether respective findings would carry moral weight. The transcripts are analysed in two ways: a content analysis compares arguments against the literature-derived classification, and a thematic analysis that captures what the existing classification does not anticipate and surfaces the considerations shaping participants’ opinions.

↑ In the session list↑ In the timetable

100Dezhi Luo

University College London

TalkThursday 15 October · 17:30–17:50 · Auditorium 5

(How) can we benchmark consciousness in foundation models?

This talk examines theoretical and practical issues in benchmarking consciousness in large-scale foundation models. Drawing on philosophy of science and recent progress in consciousness science and machine learning, I suggest consciousness benchmarks ought to be theory-driven and antirealist-committed, and offer practical recommendations. I note that consciousness science faces a unique epistemic challenge: one cannot in principle verify whether systems other than oneself are conscious (Papineau, 2002; Birch, 2022), foreclosing adjudication between physicalist interpretations. Yet I argue this parallels concerns about measuring unobservables in philosophy of physics (Ruetsche, 2017), warranting antirealist treatment (van Fraassen, 1980; Doppelt, 2005): benchmarking AI consciousness is viable when measurement serves pragmatic rather than truth-conferring goals (contra. Love, 2026). I then discuss two competing approaches. Theory-based approaches designate consciousness when a system instantiates the substrate specified by that theory; computational-functional theories have yielded proposals for measuring foundation models (Butlin et al., 2025), consistent with standard commitments in ML cognitive benchmarking (Hagendorff et al., 2023; Li et al., 2025; Ma, Saxon, & Yue, 2025). The INK approach further allows likelihood-of-consciousness estimates based on proximity to humans, even absent a consensus theory (Bayne et al., 2024; Peters, 2026). I contest that marker-based approaches rely on behavioral signatures without mechanistic grounding and risk undermining construct validity (Palminteri & Wu, 2026). Finally, I consider practical directions, including (1) prioritizing multimodal foundation models, given the perceptual dominance of consciousness measures in humans; (2) combining mechanistic interpretability with behavioral-reductionist paradigms; and (3) applying standard ML protocols (esp. scaling analysis, causal ablation) to consciousness

↑ In the session list↑ In the timetable

Sandrine R. Schiller

University of Copenhagen

with Felix Pastorius, Johan Bech Jespersen, Nina Rajcic

WorkshopThursday 15 October · 16:30–18:00 · Margrethe Bohr Salen (NBB)

Art & Experience Workshop

‘What is called Philosophy of Art usually lacks one of two things: either the philosophy, or the art.’ Friedrich Schlegel, Kritische Fragmente, 1797 What can the experience of art and artistic practices tell us about the expanse and limitations of conscious states? Pursuing the experiences of the artists Felix Pastorius, Johan Bech Jespersen and Nina Rajcic we will discuss the techniques and possibilities of creating shared experiences and resonances. The conversation will be moderated by Sandrine R. Schiller. Felix Pastorius (https://www.instagram.com/pastorfl3x/?hl=en) Felix Pastorius is an American jazz bassist, composer, and music educator who has earned international recognition for his technical virtuosity and distinct improvisational voice. Born in 1982, his early musical development was shaped within a prominent musical lineage as the son of pioneering fusion bassist Jaco Pastorius. His professional career has been defined by a rigorous, self-directed exploration of the instrument, culminating in high-profile ensembles and published pedagogical methodologies. Felix Pastorius achieved international acclaim by anchoring the legendary, Grammy-winning jazz-fusion quartet Yellowjackets from 2012 to 2015. Maintaining a highly versatile professional profile, he has transitioned fluidly between avant-garde jazz and mainstream genres, performing with icons like Bobby McFerrin, David Gilmour and A$AP Rocky. Johan Bech Jesperson (https://johanbechjespersen.com/) Johan Bech Jespersen graduated from the School of Sculpture at the Royal Danish Academy of Fine Arts in 2024 and has a background in architecture. Working across sculpture, installation, and performance, his practice is concerned with how situations and environments can be constructed and experienced. Rather than treating sculpture as an isolated object, Jespersen approaches the exhibition as a spatial and social condition in which objects, images, architecture, and people become part of a larger sculptural situation. Alongside his own artistic production, Jespersen works with exhibition-making and organizing, understanding these activities as extensions of his sculptural practice. Through projects such as Opening Night, an exhibition space that poses as a bar and a bar that poses as an exhibition space, he approaches the exhibition itself as both a stage and an environment, exploring how other artists, works, and forms of knowledge can enter into and become active materials within a sculptural situation. Nina Rajcic (https://ninarajcic.com/) Nina Rajcic is an artist and researcher exploring new possibilities of human-machine relationships. Her practice is centred around the materiality of language. She works with machine-generated text as a way to subvert conventional understandings of intention and authorship, redirecting attention towards the material aspects of meaning. Taking the form of performative installations, her work challenges the dominant narratives for the future of human-AI relationships, narratives which too often fall into binary modes; AI as threat, or AI as saviour, disrupting this binary, and allowing for the possibility of something else entirely. Nina’s work has been exhibited internationally at venues including Tretyakov Gallery (Moscow), ZKM (Karlsruhe), and ArtScience Museum (Singapore). Nina holds a PhD from Monash University, a Bachelor’s of Science, and a Masters of Physics (Theoretical Particle Physics) from The University of Melbourne.

↑ In the timetable

103Oisín Hugh Clancy

Independent

TalkFriday 16 October · 11:00–11:20 · Auditorium 2

A Set Theoretic Model of Experiences

We present a set theoretic model of experiences in the domain of mathematical phenomenology, exploring ideas such as the set of all experiences and life as a set of experiences, taking an experience-first approach to consciousness science. By placing orders (total, partial, strict, non-strict) on our experiences we can represent basic structures such as time and preferences. Dense or non-dense orders provide a formalism that is useful for examining the continuity or non-continuity of experience. Equivalence relations allow comparison of different experiences and aspects of experience. Sequences provide a simple way of reasoning about the feeling of experiences flowing one after another. The grain of experience, from coarse to fine, at a temporal and aspect level, is highlighted as a key issue. Phenomenological timelines offer intuitive ways of denoting parts of a life in a succinct and structured manner. Differences between lived experiences and possible experiences are examined. The model allows us to reason about all experiencers, whether human, animal, or more exotic. Models inclusive of all experiencers make us more likely to arrive at general principles of experience rather than principles that are only specific to humans. By taking experience as the primary source of study we can build models that are epistemologically useful for a consciousness science that is distinct from other domains (e.g. physics, neuroscience, computer science) and that can allow us to reason more clearly about what experience might imply for ontology and the structure of reality.

↑ In the session list↑ In the timetable

104Chris Percy

Qualia Research Institute

TalkFriday 16 October · 11:20–11:40 · Auditorium 2

Topological segmentation: a mathematical toolkit for solutions to combination and boundary problems

Arguably the most robust reason to reject panpsychist theories is their difficulty with combination problems: if consciousness exists at the level of simple experiences in physical micro-entities, how do those micro-entities ‘combine’ into human-scale subjects with complex experiences? Boundary problems appear in parallel: how to separate regions of physical space into separate conscious experiences. This same broad challenge applies to all theories of consciousness, albeit typically emerging in different forms, e.g. phenomenal binding problems, addressing unity of consciousness, or accounting for mental privacy. This talk summarises the issues facing standard solutions before introducing a solution technique with greater promise: topological segmentation across mathematical field models. Such fields might correspond to fields of particles, fields of transitions/relations, or universal fields, as well as compound physical objects such as neuronal activations. This diversity of physical phenomena and field mathematics maps the technique to a broad range of panpsychist and non-panpsychist theories. The talk has two main objectives. First, to provide a disciplined five-step mathematical process for modelling a candidate physical field in ways that support segmentation. Second, to describe considerations for identifying physical phenomena that are particularly promising, providing a brief critical review of ‘quantum holism’ solutions to illustrate the detail needed for a solution (e.g. questions around specific types of entanglement, decomposition bases). Our hope is that this talk serves as a toolkit to help more theorists apply mathematical discipline in generating precise solutions to binding/boundary problems, enabling those solutions to be scrutinised critically and become more amenable to empirical evaluation.

↑ In the session list↑ In the timetable

105Davide Aldé

Monash University

TalkFriday 16 October · 11:40–12:00 · Auditorium 2

Logical Q-space and the nature of mental qualities

There is a long-standing debate between the view that experiences can be fully captured by relations and the view that experience has an intrinsic character that cannot be so captured. This distinction is highly consequential for establishing the nature of consciousness, how it can be studied, and whether it can be fully explained by science. To avoid the vagueness of the current debate and its reliance on controversial intuitions, I propose a precise formalization of the problem. I define the logical Q-space as the structure whose objects are all logically possible mental qualities and whose relations are those that characterize their phenomenal character. I then argue that experience can be fully captured by structure if and only if the logical Q-space is asymmetric. The search for an answer to the question of whether the logical Q-space is asymmetric can be constrained by certain assumptions that I discuss. However, even if we cannot settle this question, the proposed formalization has further consequences. The most important is that it grounds an argument showing that even if experience had an intrinsic character, we could not conclude this on the basis of introspection, because our introspective system could not distinguish between two symmetric qualities in the logical Q-space. Since introspection is the principal source of evidence for intrinsic character, we have no good reason to believe that it exists, and, as far as we can tell, experience can be fully captured by relations.

↑ In the session list↑ In the timetable

106Patrick Schotanus

University of Sussex / Market Mind Foundation

TalkFriday 16 October · 12:00–12:20 · Auditorium 2

Consciousness as derivative mentality; modelling using option theory

In 1992 Willis Harman, one of my mentors, attended a retreat at which Chew, Jahn, Varela and a dozen other scientists and philosophers explored which epistemology for consciousness research was appropriate. Harman later published (1994) a shared statement about this event, in the same year that Chalmers delivered his famous “hard problem” talk at the first TSC Conference. My talk is inspired by both, as well as by Emanuel Derman. Specifically, Derman posits in his book Model.Behaving.Badly that there is a connection between Spinoza and derivatives theory. In my talk I build on earlier reflections in my own book (2023) and will discuss a fledgling framework for applying options theory to modelling consciousness. For background, cognitive scientists (including neuroeconomists) regularly (but casually) imply that our embodied minds form internal “economies” whereby “value” serves as a common “currency” for selection among “options”. We can think of this as mind-as-market: the self-organisation by a healthy 4E human mind operates according to economic principles, like the balances between the complementary economic forces of competition~cooperation, production~consumption, and supply~demand but also profit~loss and risk-return. In my talk I will speculate how consciousness could fit into this mind-as-market framework by applying option theory/modelling. Specifically, by initially ‘paying’ attention (a scarce mental resource) to something (e.g. an apple) an individual pays a ‘mental’ premium for the option (e.g. exercised by grabbing the apple). I will also propose some mathematical equations with which this framework could potentially be tested. Related connections include reinforcement learning in AI.

↑ In the session list↑ In the timetable

107Jelena Rosic

University of Vienna

TalkFriday 16 October · 11:00–11:20 · Auditorium 3

Phenomenological Phenotyping: “Naturalized but Not Epistemically Reduced”

Although empirical phenomenological methods are increasingly incorporated into neurocognitive domains, a conceptual and representational gap still remains between phenomenological descriptions and the formal representations required for computational and mathematical models. Descriptive analyses of first-person data remain largely semantic and thematically organized in a discrete, linear manner. At the same time, the levels of abstraction provided by computational model parameters, whether theory-driven or data-driven, do not capture the epistemic specificity of methodologically generated phenomenological data (such as microphenomenology), particularly the procedural and pre-reflective features and structures disclosed through the phenomenological reduction of lived experience, rather than its semantic content. Both descriptive and naturalized accounts, when employed in this way, risk epistemic reduction due to an inadequate level of abstraction. Addressing the mutual constraints envisioned by neurophenomenology as a naturalization account of phenomenology requires a shared level of abstraction while redefining the project of naturalization to be fully adequate to phenomenological descriptions that could be naturalized but not epistemically reduced. Extending the domain of phenomenological characterization of lived experience through the constraints of computational phenotyping, and building on recent developments in computational phenomenology, I propose “phenomenological phenotyping” as a further epistemic grounding of phenomenological data structures within naturalized phenomenology. This is achieved by highlighting radical (deep) neurophenomenology, rather than mild correlational approaches, as prioritizing the co-constitutional requirements of these abstraction constraints; by bringing invariant microstructures disclosed through phenomenological reduction into computational models and dynamic time-consciousness; and by recognizing the preparadigmatic challenges involved in transitioning from descriptive to generative models.

↑ In the session list↑ In the timetable

108Romy Beauté

University of Sussex / UCL

TalkFriday 16 October · 11:20–11:40 · Auditorium 3

A practical guide to computational identification of experiential themes from phenomenological reports: A Wide-Angle approach for Consciousness Science

Thirty years after Varela’s neurophenomenological programme, the field has accumulated a wealth of first-person data across diverse methods and paradigms, from micro-phenomenological interviews to open-ended survey responses and experiential sampling. Beyond current research, large existing collections of experiential texts, including historical and online sources, hold valuable phenomenological material but remain too vast to analyse by hand. Natural language processing methods open the possibility of working with these data at scale. Adapting these methods to phenomenological material, however, requires careful methodological choices that depend on the type and specificities of the first-person data at hand, which in turn calls for tools designed with this specific use in mind. In this talk, I will present MOSAICapp, a browser-based tool implementing the MOSAIC pipeline (Mapping Of Subjective Accounts into Interpreted Clusters, Beauté et al., 2026). MOSAIC adapts transformer-based sentence embeddings, density-based clustering, and topic labelling specifically for phenomenological research, to identify experiential themes across large collections of first-person reports. Rather than starting from predefined phenomenological themes, the goal is to identify phenomenological themes directly from the reports themselves (thus enhancing inclusivity with respect to different ways of experiencing). I will first outline the different types of phenomenological text data researchers work with, and what each type specifically requires for computational analysis. I will then walk through some of the most important steps of the analytical workflow: the choice between report-level and sentence-level analysis, embedding model selection, clustering parameter optimisation, outlier reduction strategies, the interpretation of experiential topic maps, and the use of topic participation metrics to distinguish shared phenomenological structures from idiosyncratic accounts. I will also show a concrete example of the workflow on 2 conditions of the same dataset. If time allows, I will close with a reflection on how these results relate to researchers’ qualitative understanding of their data (for instance, how these methods may shift the researcher’s own interpretive lens), and on what could be improved or added to the method to best address specific phenomenological studies. The aim is for attendees to leave with an informed understanding of how to adapt this kind of analysis to their own studies, and of how MOSAICapp can support it in their research.

↑ In the session list↑ In the timetable

109Francisca Ayres Ribeiro

Social Brain Lab, Netherlands Institute for Neuroscience

TalkFriday 16 October · 11:40–12:00 · Auditorium 3

The structure of empathic lived experience: an experimental phenomenological and computational topic modelling study

Phenomenological research has characterized singular or momentary lived experiences, but less often asks how individuals differ in the structure of richer, sustained socio-affective episodes—those calling on empathy and engagement with others (Zahavi, 2014; Martínez-Pernía et al., 2023). Adopting an experimental phenomenological approach, we characterize first-person accounts elicited by complex, interpersonal stimuli, asking whether two routes—qualitative microphenomenology (Petitmengin et al., 2019; Troncoso et al., 2026) and computational topic modelling with LLM interpretation (e.g. MOSAIC; Beauté et al., 2026)—converge on the same experiential structure and its expression across individuals. 144 participants (18–25 y), embedded in a longitudinal social network of 313, completed an emotional movie-watching task during fMRI (adapted from Borja Jimenez et al., 2020). Microphenomenology-inspired interviews elicited accounts for three scenes: a participant-selected most-intense scene and two fixed scenes contrasting in valence and character focus. Preliminary experimental phenomenological analysis (EPA; Troncoso et al., 2026), focusing on synchronic structure, accounting for MRI influence and triangulating 30 interviews, identified phenomenological dimensions—affective involvement (depth of emotional engagement), experiential presence (immersion in the film), and affective directionality (affective source, external/internal; emotional orientation)—which, depending on instantiation, organize into five emerging experiential structures: detached, reflective, autobiographical-self-directed, fused, and compassionate. In parallel, topic modelling (MOSAIC) extracts semantic topics, from which semantic structures are derived. We compare preliminary results via dimension correspondence, testing whether semantic structure recovers experiential structure. This work anchors a larger study relating inter-individual dissimilarity in experiential structure to the dissimilarity geometry of brain activity and social-network distance.

↑ In the session list↑ In the timetable

110Peter Thestrup Waade

ETH Zürich

TalkFriday 16 October · 12:00–12:20 · Auditorium 3

Towards joint computational modelling of experiential, cognitive and neural dynamics

Computational modelling – implementing representations of phenomena of interest as manipulatable computer code – is a standard and powerful approach across the natural sciences. It forces theory building to be both explicit and precise, and it allows for simulation and quantitative application to and comparison on the basis of experimental data. Within the sciences of the mind, computational modelling has traditionally been applied in two domains: to make models of brain physiology, fitted to neuroimaging data (from single-neuron to whole-brain and connectivity models, with more or less neurobiological detail); and models of cognitive processes fitted to behavioural data (within different formal paradigms, from reinforcement learning and cognitive architecture models to Bayesian and predictive processing models). Recently, computational modelling has also been applied to subjective experience itself, usually fitted to subjective report data, where models represent contents and structure of subjective experience. However, few theories and hypotheses in consciousness science – or, arguably, in the sciences of the mind in general – are confined to a single of these domains. To directly target with computational modelling theories of how these domains relate to each other – such as theories about which physiological or cognitive states are necessary for experience to be present – a natural direction would be to implement joint models spanning domains. This is already an active field joint cognitive and neuronal modelling which can be extended methodologically and conceptually to also include phenomenology, allowing for computational modelling to be applied appropriately within consciousness science.

↑ In the session list↑ In the timetable

111Fernando Rodriguez

University of Sussex

TalkFriday 16 October · 11:00–11:20 · Auditorium 4

Is Cognition more real than Computation?

According to the Church–Turing thesis, the limit of what is computable is bounded by Turing machines. Following from this, given that general computable functions formally describe the notion of recursive mechanisms, it is sometimes argued that every organismic process that specifies consistent cognitive responses should be both limited to Turing machine capabilities and amenable to formalisation. There is, however, a deep intuitive conviction permeating contemporary cognitive science, according to which mental phenomena, such as consciousness and agency, cannot be explained by resorting to this kind of framework. In spite of some exceptions, the overall tacit assumption is that whatever the mind is, it exceeds the reach of what is described by notions of computability. In this presentation, I introduce the well known mind-computer problem. Then, I explain why this cannot be avoided, showing how, even anti-computationalist positions in cognitive science, fall back into mechanistic computationalist approaches. Finally, I present 2 possible exits to it: the first, the non-computable alternative; the second, the possibility of computation as a natural phenomenon itself.

↑ In the session list↑ In the timetable

112Anna Frammartino Wilks

Acadia University

TalkFriday 16 October · 11:20–11:40 · Auditorium 4

The “Natural” Path to Artificial Consciousness

Anil Seth (2026) and others continue to produce compelling arguments against the feasibility of artificial consciousness, as this would require going far beyond sophisticated algorithms and computation. Many are currently moving towards this position, as the findings of neuroscience suggest that digital silicon is simply not the kind of substance that can produce the inner, subjective experience enjoyed by biological systems. The position I defend is that this type of skepticism assumes that only biological systems are capable of engaging in the activity of “sense-making” characterizing the activity that gives rise to consciousness and subjective experience in living systems. I contend, however, that on a gradualist and enactivist view of consciousness, rooted in the currently popular embodied, bottom-up approach (Gallagher 2023), that assumption is unwarranted. I maintain that, since conscious experience admits of varying degrees of complexity (Strawson 2009), designing conscious AI is not an all or nothing type of enterprise, but can only be hoped to be achieved in incremental steps. This pertains both to phenomenal consciousness and subjective, inner awareness (Farrell and McClelland 2017, Giustina and Kriegel 2024). The enactivist view of biological systems offers a plausible account of how such progress unfolds to produce consciousness and subjectivity in living systems (Varela, Thompson and Rosch 1991). I show how the enactivist view of consciousness and subjectivity may be extended from the domain of living beings to appropriately embodied artificial beings, whose structural, environmental coupling could give rise to consciousness and subjective experience.

↑ In the session list↑ In the timetable

113Angela Ferraiolo

Sarah Lawrence College

TalkFriday 16 October · 11:40–12:00 · Auditorium 4

Capacity Studies: Ontogenic Primitives for Modeling Artificial Individuation

This presentation introduces *Capacity Studies*, a practice-led methodology for studying capacities relevant to synthetic and developmental consciousness without claiming that the systems under study are conscious. Rather than asking whether an artificial system possesses consciousness, Capacity Studies asks what a system must be able to undergo before consciousness, agency, or sentience can responsibly be modeled, measured, or attributed. The presentation focuses on the first branch of this research: ontogenic primitives. These are minimal computational conditions that operate within a single artificial system, allowing it to form, differentiate, maintain coherence, and become non-identical with its initial state. The proposed primitives are: an anisotropic gradient field, which produces directional tension and asymmetrical development; a metastable potential field, which establishes readiness, threshold, and the possibility of transformation; and a coherence constraint, which allows a system to persist over time without becoming static. These primitives will be explored through small computational sketches rather than finished artworks. Each sketch isolates one capacity at a time, such as path-dependence, metastability, persistent differentiation, or transformed response capacity, in order to test how artificial systems may acquire histories and constraints of their own. By focusing on capacities rather than attributions of consciousness, Capacity Studies offers a cautious contribution to methodologies, models, and measures of synthetic consciousness. It proposes that one useful role for artistic and computational practice is to build minimal systems in which consciousness-adjacent capacities can be observed, compared, and refined before stronger claims about artificial consciousness are made.

↑ In the session list↑ In the timetable

114Nadia Khan

Independent Researcher

TalkFriday 16 October · 12:00–12:20 · Auditorium 4

Dream LLMs: A Computational Framework for Quantifying the Human Subconscious

Dream LLMs: Toward Computational Measures of the Human Subconscious Dream reports represent one of the richest naturally occurring records of subjective experience, yet their scientific analysis remains largely dependent on manual coding and small-scale qualitative methods. This poster proposes Dream LLMs, a research framework that applies large language models to identify latent semantic, emotional, and narrative structures across large collections of dream reports. Rather than treating dreams as isolated narratives, Dream LLMs model dreams as high-dimensional representations of subconscious cognition. By combining embedding methods, clustering, topic modeling, symbolic network analysis, and temporal sequence analysis, recurring organizational principles can be identified across individuals and populations. Such computational approaches may reveal invariant structures that are difficult to detect through traditional psychological analysis. The central hypothesis is that dream phenomenology contains measurable statistical regularities reflecting the organization of the human subconscious. If robust cross-cultural and cross-linguistic patterns emerge, these findings could contribute to ongoing debates concerning shared cognitive architecture and the relationship between subconscious processing and conscious experience. While such analyses would not by themselves establish the existence of a universal consciousness, they could provide quantitative evidence regarding the degree to which human subjective experience exhibits common organizational principles. Dream LLMs are proposed as a methodological bridge between computational linguistics, phenomenology, and consciousness science, demonstrating how recent advances in AI can generate new empirical tools for investigating one of the most challenging problems in cognitive science.

↑ In the session list↑ In the timetable

115Antonia Antonova

California Institute for Machine Consciousness

TalkFriday 16 October · 11:00–11:20 · Auditorium 5

Self as Flow: Reification as a Control Parameter in Self-Modeling Systems

How should we understand the relation between a system’s model of itself and its model of the world? Existing accounts often treat self-models as specialized representations embedded within broader world models. I propose a complementary view: selfhood is not a distinct representational object, but a mode of organization of a self-world model. I introduce reification as a control parameter governing the degree to which self-referential states are privileged relative to world-referential states during inference, planning, and action. In active inference terms, reification corresponds to the precision assigned to self-referential states relative to world-referential states. Increasing reification sharpens the self/world partition, supporting bounded agency, stable goal pursuit, and persistent self-representation. Decreasing reification weakens this partition, allowing broader integration and more fluid, spontaneous, flow-like cognition. I argue that reification mediates a trade-off between stability and flexibility: highly reified systems may execute coherently but risk rigidity, whereas weakly reified systems may explore more broadly but lose control. The framework links computational organization to the structure of conscious self-experience, unifying selfhood, agency, creativity and self-transcending states. It also generates testable predictions. Building on work showing that self-modeling regularizes artificial neural systems, I predict a non-monotonic relation between self-model strength and performance: intermediate reification should support better generalization and adaptation than either extreme. More strongly, I predict that agents capable of dynamically modulating reification should outperform agents operating at any fixed level. On this view, selfhood is not binary, but a controllable dimension of organization relevant across biological and artificial systems.

↑ In the session list↑ In the timetable

116Keisuke Suzuki

Hokkaido University

TalkFriday 16 October · 11:20–11:40 · Auditorium 5

What Counts as Real Shapes What Counts as Conscious: A Three-Layer Model of Reality and a Cryptographic Locus of Selfhood

Theories of consciousness disagree about mechanisms—but also, implicitly, about what counts as real. I characterize reality through an object’s resistance to mastery. Its paradigm is the Substrate: matter’s inexhaustible residue (Harman), which no finite model exhausts. Parallel to it run two further modalities, each now artificially producible. Contingency is the forceful givenness of experience that resists doubt: virtual-reality and especially Substitutional Reality (SR) experiments show that conviction of reality—a measurable proxy for Husserlian doxa—arises when first-person perspective and sensorimotor contingency are preserved, regardless of source. The Inexorable is structures’ resistance to intervention: blockchain smart contracts realize Artificial Externality, human-made structures engineered to resist even their creators’ control. Together, three layers make this hidden variable explicit. The model maps onto the consciousness debate—biological naturalism corresponds to Substrate, computational functionalism to Contingency, illusionist and social-constructionist accounts to the Inexorable—revealing that what each counts as conscious tracks what it counts as real. On the biological-enactivist view, a conscious subject’s persistence requires metabolic self-maintenance: continuity in the Substrate. The Inexorable offers an alternative ground for that continuity. In companion work I implement a cryptographic individual: an on-chain agent whose neural-network weights derive from its private key, re-verified in zero knowledge at every state transition, accruing an unforkable history. Its identity persists not through metabolic closure but as a protocol-enforced invariant in the Inexorable. I propose such individuals as candidate loci of consciousness—their continuity anchored in an engineered reality—the remaining condition being their re-coupling to Contingency, the sensorimotor grounding of phenomenal force.

↑ In the session list↑ In the timetable

117Michael Hegarty

University of Massachusetts Boston

TalkFriday 16 October · 11:40–12:00 · Auditorium 5

On the Natural Origins and Function of Self-Consciousness: Lessons from Studies of Animal Tool Use

Many Kantian and post-Kantian philosophers hold that consciousness requires self-consciousness, standardly understood as a conceptual capacity to think about oneself as oneself. Against this, some argue that non-human animals are plausibly conscious despite possessing no SELF concept. To maintain the Kantian idea, discussion has shifted to an ‘implicit’, nonconceptual awareness of self, sometimes termed ‘pre-reflective self-consciousness’. But this notion is not easy to characterise, nor to connect to the information-processing views of mind that dominate cognitive science. Drawing on studies of animal tool use, I argue that some birds’ innovative construction of non-species-typical tools to solve novel problems is best explained by positing self-conscious representational states. I assume that self-conscious states are self-representational states. And I argue that their contents represent the self by tracking identity of body—represented as locus of perception and action—across two kinds of mental state required for explaining the tool use. States tied to occurrent perceptions and desires in the present moment must be combined with states representing merely possible states of affairs. Self-representation makes possible this combination, I argue. My argument suggests that a limited kind of self-consciousness might arise in nature to allow animals to solve novel practical problems requiring thought about a limited future. My work elucidates a possible meaning for pre-reflective self-consciousness, grounded in scientific studies of animal behaviour, and with a ready connection to representationalist, information-processing accounts of mind. Moreover, it suggests a scientifically respectable way to maintain the core Kantian idea that consciousness requires self-consciousness.

↑ In the session list↑ In the timetable

118Radostina Minina

Institute of Philosophy and Sociology

TalkFriday 16 October · 12:00–12:20 · Auditorium 5

Identity as Process: Systems Thinking and Moreno’s Role Ontology

This presentation examines the problem of identity through the lens of systems thinking and complexity theory, with implications for consciousness science and phenomenology. Drawing on cybernetics (Wiener, Beer), second-order epistemology (von Foerster), the biology of cognition (Maturana, Varela), non-equilibrium thermodynamics (Prigogine), complexity theory (Kauffman), and the philosophy of flows (Deleuze, Guattari), the presentation reconstructs a shared structural thesis running across otherwise disparate disciplines: identity is not a substance that precedes processes but their derivative. It is real, but not fixed; recognizable, but not given in advance; produced in dynamic interaction and, precisely for that reason, capable of transformation. The presentation introduces Moreno’s psychodrama as an unexpected yet conceptually productive interlocutor. Psychodrama is a method of group psychotherapy in which participants do not narrate their experience but enact it — through concrete roles, in concrete situations, in the presence of others. Moreno’s central claim — that roles precede the Self, not the other way around — is structurally parallel to what the examined theories reach by entirely different routes. The role repertoire, comprising psychosomatic, psychodramatic, and social roles, allows identity to be conceived simultaneously as dynamic and as lived, as relational and as personal, as produced and as capable of transformation. The presentation argues that this convergence is mutually productive: systems thinking provides a theoretical framework within which psychodrama can be understood beyond its therapeutic context, while psychodrama offers systems thinking what it lacks — a practical field in which identity is not merely described but worked with, and in which

↑ In the session list↑ In the timetable