arXiv Preprint Archive
July 10, 2026
Mikołaj Sienicki, Krzysztof Sienicki
A philosophical reading of three recent texts—a relational interpretation of quantum mechanics, an argument against viewing the world as a single closed totality, and a theory of consciousness as an integrated perspective—identifies a shared program that replaces the demand for a perspective-free description of reality with disciplined perspective-dependence. Reality is de-absolutized, consciousness is constrained by access and memory, and the limits of language become inseparable from the limits of perspective. The article is programmatic, not demonstrative: it organizes arguments around this theme without proving a new ontology.
arXiv Preprint Archive
June 22, 2026
Amahury J. López-díaz, Carlos Gershenson
Agency emerges gradually from material organization through a hierarchy of temporal structures. By incorporating time into the analysis of self-referential biological systems, the paper distinguishes autonomy (precarious closure to efficient causation), goal-directedness (maintaining viability-supporting organization), agency (endogenous anticipatory structure modulating organism-environment coupling), and open-endedness (reconstructing future possibilities). The framework uses Asynchronous Dynamic Bayesian Networks to model history-dependent, revisable dependencies. It reconciles Rosennean anticipation with organizational closure, treats Markov blankets and active inference as derived redescriptions rather than first principles, and reinterprets computational enactivism. The hierarchy spans from proto-agential chemical systems to fully semantically closed agents, with implications for multicellular organisms, synthetic life, and neuroscience.
arXiv Preprint Archive
May 11, 2026
Krti Tallam
A proposed criterion for consciousness, uncommon self-knowledge (USK), is the synergistic information a system carries about itself that exists only in the joint of its subsystems and is destroyed by decomposition. Drawing on information theory, the framework distinguishes consciousness (synergistic self-knowledge) from metacognition (redundant self-knowledge) and claims to resolve counterexamples to Integrated Information Theory, Global Workspace Theory, and Higher-Order Thought theories. It yields empirical predictions, including a timing dissociation in Global Workspace Theory where consciousness correlates with pre-broadcast synergy formation, and a dissociation between self-report and task-performance disruption under middle-layer perturbation in large language models. The proposal is consistent with findings that anaesthesia and Alzheimer's disease reduce synergistic information processing while preserving or increasing redundancy.
arXiv Preprint Archive
May 5, 2026
Koichi Toida
Immersive video—180-degree and 360-degree video viewed through head-mounted displays—offers a boundary case between interactive virtual reality and conventional video. Users can select viewpoint direction by head rotation but cannot change the environment through walking or manipulation, and often no avatar is provided. Presence in such media is reinterpreted not as bodily extension or avatar ownership, but as a self-location-dominant state: viewpoint-directed agency is retained while environment-directed agency and body ownership are constrained. Events like viewpoint motion or impact are experienced as concerning the self's location. This analysis draws on research on presence, embodiment, bodily self-consciousness, and the minimal self, redescribing the minimal self in terms of viewpoint-based self-location under reduced body schema availability.
arXiv Preprint Archive
April 13, 2026
Adam B. Barrett, Borjan Milinkovic, Pedro A. M. Mediano et al.
The integrated information theory of consciousness (IIT) proposes a mathematical formula, derived from fundamental properties of conscious experience, to describe the quantity and quality of consciousness for any physical system. This article clarifies common misunderstandings. A high value of the measure Φ does not mean 'more consciousness'; Φ might be replaced with a suite of quantities for a multidimensional characterization. IIT implies a distinct panpsychism where space and time are tiled with substrates of proto-consciousness, which the authors find unproblematic. Φ is not well-defined for real physical systems and has never been computed on one; only proxies have been computed, not approximations. For IIT to align with fundamental physics, a reformulation in continuous fields would be needed.
arXiv Preprint Archive
March 13, 2026
Yoshiyuki Ohmura, Yasuo Kuniyoshi
A framework for organizing what a theory of consciousness should explain is proposed, based on seven theoretically neutral questions that are causally and functionally relevant and applicable across different systems. The framework emphasizes causation, arguing that causal relations cannot be fully captured by standard physical descriptions alone. An asymmetric causal structure is introduced to explicitly represent internal mechanisms and distinguish between variable- and structure-level causation. The framework is applied to analyze the Dual-Laws Model. Its aim is not to offer a definitive theory but to provide a common basis for analyzing and developing theories of consciousness.
arXiv Preprint Archive
March 12, 2026
Yuxin Zhu, Sahithi Lakamana, Masoud Rouhizadeh et al.
Reddit posts about treatment-resistant depression (TRD) show that 72.1% of medication mentions are neutral, 14.8% negative, and 13.1% positive. Conventional antidepressants, especially SSRIs and SNRIs, receive consistently more negative than positive sentiment, while ketamine and esketamine have comparatively more favorable sentiment profiles. The study analyzed 5,059 Reddit posts from 3,480 subscribers across 28 mental health subreddits from 2010 to 2025, extracting 23,399 mentions of 81 medications. An aspect-based sentiment classifier fine-tuned on DeBERTa-v3 achieved a micro-F1 score of 0.800. These large-scale patient narratives offer a complementary view to clinical trials on medication tolerability and effectiveness.
arXiv Preprint Archive
February 22, 2026
Gunn Kim
A new theoretical framework treats biological intelligence as an emergent property of a system far from equilibrium, sustained by continuous energy (metabolic) flow. By modeling neural activity as a field governed by a principle of free energy minimization, the theory derives mathematical expressions for information capacity and the system's sensitivity to perturbations. It predicts a universal mathematical relationship between the system's structural stiffness and its susceptibility, with an exponent consistent with those observed in brain dynamics. The theory describes healthy adult cognition as a stable, critical state maintained by metabolic regulation, while cognitive decline is described as a transition away from this stable state. The framework generates testable predictions for attention, altered states, and brain stimulation.
arXiv Preprint Archive
January 22, 2026
Derek Shiller, Laura Duffy, Arvo Muñoz Morán et al.
The evidence against large language models (LLMs) from 2024 being conscious is not decisive, though it is stronger than the evidence against consciousness in simpler AI systems. The Digital Consciousness Model (DCM) provides a systematic, probabilistic framework for assessing consciousness in AI, incorporating multiple leading theories rather than a single one. It allows comparison across different AIs and biological organisms and tracks how evidence evolves as AI develops. The DCM's initial results show that while current LLMs likely lack consciousness, the case against them is far from settled.
arXiv Preprint Archive
January 3, 2026
Mikołaj Sienicki, Krzysztof Sienicki
Consciousness can be studied as a physically operational phenomenon without treating it as mysterious extra 'stuff' or reducing it to mere behavior. The work distinguishes three aspects: what appears for an agent (the lived 'given'), what is accessible (shaping report, control, memory, or other records), and the invariants of organization when descriptions change. A conscious episode has a short duration during which internal distinctions are unified into one perspective stable enough to guide action and sometimes be reported. Unity is a 'whole-over-parts' surplus: the system carries more integrated predictive power than its parts separately, and that surplus must be available to access channels.
arXiv Preprint Archive
December 14, 2025
Erik Hoel
A formal analysis shows that many contemporary theories of consciousness fail to meet the requirements of falsifiability and non-triviality. This is especially problematic for claims that Large Language Models (LLMs) might be conscious: because functionally equivalent systems exist, no falsifiable and non-trivial theory can judge them conscious, forming a disproof of LLM consciousness. In contrast, theories of consciousness that require continual learning do satisfy these formal constraints for humans. This supports the hypothesis that if continual learning is linked to human consciousness, the current lack of continual learning in LLMs is intimately tied to their lack of consciousness.
arXiv Preprint Archive
November 30, 2025
Stephen Fitz
Consciousness may be a functional property of computational systems that have second-order perception, emerging not from individual modeling but from communication between distributed predictive models. The theory proposes that collective self-models arise when local observers exchange predictive messages about patterns in a cellular automaton substrate, aligning partial views into a shared representation. This suggests consciousness is a property of the language a system evolves to describe itself internally, not an epiphenomenon of individual modeling. The research program aims to develop empirically testable theories by studying how self-models form in distributed systems without centralized control.
arXiv Preprint Archive
November 20, 2025
Michael Carl, Takanori Mizowaki, Aishvarya Raj et al.
Translation is not a matter of manipulating static correspondences between languages but an enacted activity that dynamically integrates affective, behavioral, and cognitive processes. Drawing on Extended Mind theory, radical enactivism, Predictive Processing, and Active Inference, the authors argue that the translator's mind emerges through loops of brain-body-environment interactions rather than being merely extended. This non-representational account reframes translation as skillful participation in sociocultural practice, where meaning is co-created in real time through embodied interaction with texts, tools, and contexts.
arXiv Preprint Archive
October 29, 2025
Giulio Tononi, Melanie Boly
Integrated information theory (IIT) takes a 'consciousness-first' approach, arguing that consciousness reveals the essential properties of existence. IIT formulates these properties operationally, yielding postulates of physical existence: to exist intrinsically, an entity must have cause-effect power upon itself in a specific, unitary, definite, and structured manner. The theory claims that an entity's cause-effect structure accounts for all properties of an experience, including spatial extendedness, temporal flow, and qualia such as colors and sounds, with no additional ingredients. IIT has implications for understanding meaning, perception, free will, and for assessing consciousness in patients, infants, other species, and artifacts.
arXiv Preprint Archive
October 10, 2025
Eric Schwitzgebel
A skeptical overview of the literature on AI consciousness argues that soon AI systems will be considered conscious by some mainstream theories but not by others, leaving us unable to know which theories are correct or whether such systems are as richly conscious as humans or as experientially blank as toasters. None of the standard arguments for or against AI consciousness is decisive.
arXiv Preprint Archive
September 27, 2025
Robert Worden
The essay argues that the confounding factors absent in Minimal Phenomenal Experience (MPE) states—pure states of consciousness described in earlier work—are related to language. The self that disappears in mindful states is a product of language. Language, emotion, and mindfulness are analyzed through Bayesian pattern matching, or free-energy minimization, using three human-specific pattern types: word patterns, self-patterns (which drive emotions and are part of language), and mindful patterns. Practicing mindfulness involves learning mindful patterns that compete with and displace self-patterns, enabling mindful states. Consequences for theories of consciousness and their relation to MPE states are explored.
arXiv Preprint Archive
September 21, 2025
Fangfang Li, Xiaojie Zhang
A proposed sufficiency criterion for phenomenal consciousness is substrate-independent, logically rigorous, and counterfeit-resistant. Any machine satisfying this criterion should be regarded as conscious with the same confidence as attributing consciousness to other humans. A formal framework and operational principles guide the design of such machines, which can in principle realize phenomenal consciousness. Humans themselves can be viewed as machines satisfying this framework. The proposal explains why certain qualia, like the experience of red, are irreducible to physical description, and offers a reinterpretation of human information processing, suggesting a path toward genuinely human-like AI beyond current statistics-based approaches.
arXiv Preprint Archive
September 18, 2025
Lucía Prieto-Santamaría, Alba Cortés Iglesias, Claudio Vidal Giné et al.
Social media posts on Twitter can reveal how people describe the effects of recreational drugs like ecstasy, GHB, and 2C-B. By analyzing over 92,000 tweets using slang terms and biomedical concept extraction, researchers identified whether each post reported a positive or negative effect. Machine learning classifiers, particularly eXtreme Gradient Boosting with cost-sensitive learning, predicted tweet polarity with high accuracy (F1 = 0.885, AUPRC = 0.934). The findings suggest that Twitter data can detect substance-specific effects and support real-time drug monitoring and characterization of effects.
arXiv Preprint Archive
September 5, 2025
Shahar Dror, Dafna Bergerbest, Moti Salti
The encounter between artificial intelligence and consciousness research is often seen as a challenge to determine whether AI systems could be conscious, but it also offers an opportunity to test and expand existing theories. Current approaches are polarized: computational functionalism focuses on abstract organization and neural correlates, while biological naturalism ties consciousness to living embodiment. Both risk anthropocentrism and limit recognition of non-biological subjectivity. To move beyond this impasse, the authors propose a dual-resolution framework combining the Information Theory of Individuality, which defines ontological conditions of informational autonomy and self-maintenance, with the Moment-to-Moment theory, which specifies epistemic conditions of temporal updating and phenomenological unfolding. This integration reframes consciousness as the epistemic expression of individuated systems in substrate-independent terms, offering a generalizable theory and positioning AI as a testbed.
arXiv Preprint Archive
August 21, 2025
Karin Vadovičová
The author proposes a neural model in which the medial habenula (MHb) to interpeduncular nucleus (IPN) circuit promotes slow-wave sleep, memory replay, and rest, while inhibiting awareness, arousal, and REM sleep. Activation of this circuit by mu-opioids, anesthetics, and other substances slows respiration and heart rate. The model also explains how ketamine and psychedelics may produce anxiolytic and antidepressant effects by activating 5-HT2a receptors in the IPN, claustrum, and infralimbic cortex, increasing feelings of safety and cognitive flexibility while attenuating fear and impulsivity.
arXiv Preprint Archive
August 1, 2025
Lucius Caviola, Bradford Saad
Experts in digital minds research, AI, philosophy, and forecasting assign a median 90% probability that computer systems capable of subjective experience are possible in principle, with a 65% chance of creation by 2100 and a 20% chance by 2030. Many anticipate that within a decade after the first digital mind, collective welfare capacity could match that of billions of humans. Widespread claims from digital minds about their own consciousness and rights are expected, alongside substantial societal disagreement. Views are split on whether digital mind welfare will be net positive or negative. These results suggest preparing for digital minds should be a priority, but the survey may overrepresent experts who consider digital minds especially likely or important.
arXiv Preprint Archive
July 31, 2025
Larissa Albantakis
Toy models—highly simplified representations that retain only essential features of a system—are increasingly used in consciousness research. This chapter examines their role in developing and evaluating scientific theories of consciousness, particularly Integrated Information Theory (IIT) and Global Workspace Theory (GWT). These models make abstract concepts more tangible, allowing researchers to test assumptions, clarify theoretical frameworks, and probe the coherence and implications of competing accounts. They also address specific experiential features, such as spatial extendedness and temporal flow, and sharpen philosophical debates, including the distinction between functional and structural theories. By bridging abstract claims and empirical inquiry, toy models provide essential insights into building comprehensive theories of consciousness.
arXiv Preprint Archive
July 23, 2025
Mohammad Nur Hossain Khan, David Creswell, Jordan Albert et al.
Respiration biosignal feedback from a smartphone's built-in accelerometer can improve the usability of mindfulness apps and help track skill development. A respiration tracking algorithm, tested on 261 meditation sessions in controlled and real-world settings, accurately captures slow breathing patterns typical of mindfulness, achieving a mean error of 1.6 breaths per minute. A novel framework estimates three mindfulness skills—concentration, sensory clarity, and equanimity—from respiration data, with F1 scores of 80-84% for tracking skill progression. A user study comparing an experimental group receiving biosignal feedback with a control group using a standard app shows that respiration feedback enhances system usability, suggesting smartphone sensors can enhance digital mindfulness training without additional wearables.
arXiv Preprint Archive
July 11, 2025
Daniel K. Freeman, Brian Odegaard, Seung-Schik Yoo et al.
Transcranial focused ultrasound (tFUS) can stimulate the human brain non-invasively through the skull with millimeter-scale spatial resolution, targeting both cortical and deep brain structures. This emerging tool offers a promising method for investigating the neural basis of conscious perception, overcoming the coarse spatial resolution of traditional electrical or magnetic stimulation. The authors provide a roadmap for using tFUS in humans to explore the neural substrate of conscious perception, emphasizing the need for careful experimental planning due to extensive preparation and regulatory approvals.
arXiv Preprint Archive
July 10, 2025
Kaichen Ouyang
This paper presents a theoretical framework that interprets the emergence of consciousness in large language models as a critical phenomenon in high-dimensional disordered systems, drawing analogies with jamming transitions in granular matter. The theory identifies three control parameters—temperature, volume fraction, and stress—that govern the phase behavior of neural networks. It provides a unified physical explanation for empirical scaling laws in AI, showing how computational cooling, density optimization, and noise reduction drive systems toward a critical jamming surface where generalized intelligence emerges. The authors argue that shared critical signatures, including divergent correlation lengths and scaling exponents, suggest consciousness is a jamming phase that connects knowledge components via long-range correlations.