Confidence, Metacognition, and the “Hard Problem” of Consciousness
Megan Peters
Cognitive Science, UC Irvine
10:30 (EST) September 11, 2025
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Video: https://youtu.be/D2VcTCDOPYU
Abstract: Conscious feeling—the “hard problem”—remains a central challenge for cognitive science. This talk explores whether computational models of metacognitive confidence and uncertainty, grounded in introspective psychophysics and higher-order representations, can illuminate phenomenal experience. I will discuss recent work showing how the brain encodes uncertainty, how confidence can be modeled computationally, and how subjective reports can be formalized. These approaches point toward canonical computations linking perception, decision-making, and metacognition, offering a possible path to studying subjective experience—and perhaps a way to rethink the hard problem itself.
References:
Peters & Azimi Asrari (2025). How brains build higher order representations of uncertainty. arXiv.
Peters (2025). Introspective psychophysics for the study of subjective experience. Cerebral Cortex.
TOPIC: Behavioral evaluation of language models as models of human sentence processing
Roger Levy
Department of Brain and Cognitive Sciences, MIT
September 18, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Video: https://youtu.be/t61u1SiSOik
Abstract: This talk examines how large language models can serve as computational models of human sentence processing, focusing on behavioral evaluation methods that compare model predictions with human psycholinguistic data. I will discuss recent work showing that direct probability measurements from language models often provide better insights into linguistic knowledge than prompting-based evaluations. The talk will cover methodological considerations for using surprisal theory and other information-theoretic measures to validate LLMs as cognitive models, examining both the promises and limitations of current neural language models in capturing human sentence processing mechanisms. I will present evidence on how model-derived measures of processing difficulty align with human reading time data and discuss implications for both cognitive science and natural language processing.
References: Hu, J., & Levy, R. (2023). Prompting is not a substitute for probability measurements in large language models. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, 5040–5060.
Shain, C., Meister, C., Pimentel, T., Cotterell, R., & Levy, R. P. (2024). Large-scale evidence for logarithmic effects of word predictability on reading time. Proceedings of the National Academy of Sciences, 121(10), e2307876121.
Wilcox, E. G., Futrell, R., & Levy, R. (2023). Using Computational Models to Test Syntactic Learnability. Linguistic Inquiry, 1–44.
Futrell, R., Gibson, E., & Levy, R. P. (2020). Lossy-context surprisal: An information-theoretic model of memory effects in sentence processing. Cognitive Science, 44, 1–54.
Meaning in Large Language Models: Bridging Formal Semantics, Pragmatics, and Learned Representations
Christopher Potts
Departments of Linguistics and Computer Science Stanford University
Thursday, September 25, 2025 10:30am – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Video: https://youtu.be/t61u1SiSOik
English Abstract: In its modern form, semantics (the study of the conventionalized aspects of linguistic meaning) is firmly rooted in symbolic logic. Such logics are also a cornerstone of pragmatics (the study of how people create meaning together in interaction). We can trace this methodological orientation to the roots of these fields in mathematical logic and the philosophy of language. This origin story has profoundly shaped both semantics and pragmatics at every level. How would these fields have looked had they instead been rooted in connectionism? They would have been radically different: the distinction between semantics and pragmatics would fall away, the range of relevant empirical phenomena would expand, and the theories themselves would have greater predictive force. This is not to say that there would be no role for symbolic logic in this hypothetical connectionist “semprag.” Large language models do learn solutions that reflect existing symbolic theories of meaning, and this is key to their success. This points to a future in which the fields of semantics and pragmatics embrace much more of what is happening in AI – without, however giving up their roots in symbolic logic.
Christopher Potts is Professor of Linguistics and, by courtesy, of Computer Science at Stanford, and a faculty member in the Stanford NLP Group and the Stanford AI Lab. His research group uses computational methods to explore topics in context-dependent language use, systematicity and compositionality, model interpretability, information retrieval, and foundation model programming. This research combines methods from linguistics, cognitive psychology, and computer science, in the service of both scientific discovery and technology development. Chris is also Co-Founder and Chief Scientist at Bigspin AI, a start-up focused on collaborative development of AI systems.
References:
Epistemological challenges in the study of “Theory of Mind” in LLMs and humans
Sean Trott
Department of Cognitive Science, University of California San Diego
October 9, 2025 10:30am – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Abstract: Humans reason about others’ beliefs—a key aspect of Theory of Mind. Can this emerge from language statistics alone? I present evidence that large language models show some sensitivity to implied belief states in text, though consistently below human levels. This suggests distributional learning is partly but not fully sufficient for Theory of Mind. I then examine epistemological challenges in treating LLMs as “model organisms,” including construct validity and distinguishing genuine generalization from pattern-matching. I argue that addressing these challenges opens opportunities for methodological innovation in both Cognitive Science and AI.
References:
Jones, C. R., Trott, S., & Bergen, B. (2024). Comparing humans and large language models on an Experimental Protocol Inventory for Theory of Mind Evaluation (EPITOME). Transactions of the Association for Computational Linguistics, 12, 803-819.
Trott, S., Jones, C., Chang, T., Michaelov, J., & Bergen, B. (2023). Do large language models know what humans know? Cognitive Science, 47(7), e13309.
Jean-Baptiste Mouret
LARSEN Team, Inria Nancy – Grand Est / CNRS, Loria
October 16, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Abstract: Robots must operate in the real-world environment, which is far more complex and varied than controlled simulations or chess-like games. Early AI research in the 1960s attempted to address this challenge by representing the world through rules and symbols, enabling reasoning via formal logic, but grounding these symbols proved to be close to impossible. In response, the robotics community developed robots that learn their own representations through end-to-end learning or evolutionary methods, bypassing the need for explicit rules and symbols. In this talk, I will first present examples of this learning-based approach, including our work on robots that adapt after physical damage. I will then discuss a recent shift: the return of symbols to robotics through large language models. I will show how these models integrate a kind of common sense, which sidesteps many of the issues with logical reasoning, and describe the early efforts by the robotics community to integrate them with robotic actions and perception.
Jean-Baptiste Mouret is Director of Research at Inria Nancy – Grand Est and member of the LARSEN team (Lifelong Autonomy and interaction skills for Robots in a Sensing ENvironment). His research focuses on machine learning and evolutionary computation for designing adaptive robots that can learn and adapt through trial-and-error learning. His work “Robots that can adapt like animals,” published on the cover of Nature in 2015, demonstrated how robots can recover from damage in under two minutes using behavioral repertoires. He has held visiting positions at Cornell University, University of Vermont, and Technical University Darmstadt.
References
Cully A, Clune J, Tarapore D, Mouret JB. Robots that can adapt like animals. Nature. 2015 May 28;521(7553):503-7.
Amadio et al. From Vocal Instructions to Household Tasks: The Inria Tiago++ in the euROBIN Service Robots Coopetition. 2024. arXiv preprint arXiv:2412.17861
Mouret, J.-B. Large language models help computer programs to evolve. Nature, 2024, 625
(7995), pp.452-453. 10.1038/d41586-023-03998-0
Totsila D, Rouxel Q, Mouret JB, Ivaldi S. Words2contact: Identifying support contacts from verbal instructions using foundation models. In 2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids) 2024 Nov 22 (pp. 9-16). IEEE.
NeuroAI: The Convergence of Neuroscience and Artificial Intelligence
Terry Sejnowski
Salk Institute for Biological Studies, UC San Diego
October 23, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Abstract: This talk explores the revolutionary convergence of neuroscience and artificial intelligence in the emerging field of NeuroAI. Drawing from recent breakthroughs in large language models like ChatGPT, I will examine how computational principles derived from brain function are informing next-generation AI systems, while simultaneously showing how AI tools are advancing our understanding of neural computation. The discussion will cover the bidirectional flow of insights between transformer architectures and cortical traveling waves, demonstrating how self-attention mechanisms in AI parallel the brain’s encoding of temporal context. I will present evidence from our recent work on predictive sequence learning in the hippocampus and how neural prediction errors mirror computational processes in modern AI. The talk will address fundamental questions about the embodied Turing test and whether AI systems can achieve the sensorimotor intelligence that evolved over 500 million years. Finally, I will discuss the implications of this convergence for understanding consciousness, memory consolidation during sleep, and the future of human-AI collaboration in scientific discovery.
References:
Embodied Language: Evaluating LLMs in the Real World
Yonatan Bisk
Language Technologies Institute & Robotics Institute, Carnegie Mellon University
October 30, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Abstract: This talk examines the critical challenge of evaluating Large Language Models in interactive, embodied settings where language must connect to physical actions and environmental understanding. Drawing from recent research in embodied AI and language grounding, I will explore how current LLMs perform when tasked with interpreting language instructions that require spatial reasoning, object manipulation, and social interaction. The discussion will cover methodological frameworks for assessing language-to-action capabilities, including benchmarks that move beyond traditional text-based evaluation to encompass multimodal environments where language commands must be translated into executable actions. The talk will address fundamental questions about what it means for AI systems to truly understand language in the context of physical agency, examining both the successes and systematic failures of LLMs in interactive settings that require grounded communication and sensorimotor integration.
References:
Bisk, Y., Holtzman, A., Thomason, J., et al. (2020). Experience Grounds Language. EMNLP.
Thompson, J., Garcia-Lopez, E., Bisk, Y. (2025) REM: Evaluating LLM Embodied Spatial Reasoning through Multi-Frame Trajectories. In Proceedings of the Second Conference on Language Modeling.
Xie, Q. Min, S., Zhang, T. et al. (2024) Embodied-RAG: General non-parametric Embodied Memory for Retrieval and Generation ArXiv.
Do LLMs pass the Turing test? And what does it mean if they do?
Cameron Jones
Department of Psychology, Stony Brook University
November 6, 2025 10:30am – noon EDT/EST
Zoom: Université du Québec à Montréal (UQÀM)
Abstract: Large Language Models (LLMs) seem well designed for the Turing test in that they can produce fluid naturalistic text. Many have suggested that they would pass the test or implicitly already have. We addressed this question empirically by evaluating several LLMs in a standard three-party 5 minute Turing test. Two models, when prompted to adopt a humanlike persona, achieved a pass rate of 50%: suggesting that interrogators were no better than chance at distinguishing between humans and LLMs. One of these models (GPT-4.5) was judged to be human 73% of the time, significantly more often than the real humans it was being compared to. These results suggest that LLMs pass the Turing test, but what does that mean? I will discuss potential interpretations of these results, including whether they suggest that LLMs are intelligent, produce humanlike behaviour, or are merely exploiting superficial cues.
References:
Jones, C. R., & Bergen, B. K. (2025). Large language models pass the turing test. arXiv preprint arXiv:2503.23674. Jones, C. R., Rathi, I., Taylor, S., & Bergen, B. K. (2025). People cannot distinguish GPT-4 from a human in a Turing test. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (pp. 1615-1639).
The Global Latent Workspace: A model of cognition with AI applications
Rufin VanRullen
Artificial and Natural Intelligence (ANITI), Université de Toulouse
November 13, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Rufin VanRullen is CNRS Research Director in neuroscience and artificial intelligence at the Centre de Recherche Cerveau et Cognition (CerCo) and research chair at the Artificial and Natural Intelligence Toulouse Institute (ANITI). His research focuses on brain-inspired AI architectures, visual perception, attention, and consciousness. VanRullen has authored over 200 scientific papers and is a leading researcher in computational approaches to consciousness and neural oscillations in perception.
References:
Butlin, P., Long, R., Elmoznino, E., et al. [including VanRullen, R.] (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. arXiv preprint.VanRullen, R., & Kanai, R. (2021). Deep learning and the global workspace theory. Trends in Neurosciences, 44(9), 692-704.
Articulating the Ineffable: The Analytic Turn in Generative AI
Ari Holtzman
Computer Science and Data Science Institute, University of Chicago
November 20, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Abstract: Generative AI has taken an analytic turn: we now cultivate models from objectives and data, then try to understand what we’ve grown post-hoc. Current approaches to studying LLMs—focused on engineering progress or mechanistic explanations at the implementation level—are insufficient for grasping their emergent behaviors. I will discuss what it means for interpretability approaches to be predictive rather than mechanistic, the changing landscape of machine communication, and efforts to identify fundamental laws that govern LLM behavior. I will argue that developing precise behavioral vocabulary and conceptual frameworks is the only way to turn the ‘fieldwork’ of finding surface regularities in LLMs into a science of LLMs. The guiding questions are basic, empirical, and exploratory: what do models consistently do, what do they reliably miss, and how do they incorporate and store new information? Along the way I’ll make the case that AI has been given a new mandate—to articulate the ineffable, by describing aspects of communication and computation that we previously had no words for because they were stuck too deep inside human cognition to be easily referenced.
Ari Holtzman is Assistant Professor of Computer Science and Data Science at the University of Chicago, where he directs the Conceptualization Lab. His research is on developing new conceptual frameworks for understanding generative models, treating them as organically arising complex systems rather than traditional engineering artifacts. He believes the main bottleneck in LLM science is the lack of good vocabulary to describe LLMs. He introduced nucleus sampling, a text generation algorithm used in deployed systems including the OpenAI API.
References:
Holtzman, A. and Tan, C. (2025). Prompting as Scientific Inquiry. arXiv preprint arXiv:2507.00163.
Holtzman, A., et al. (2023). Generative Models as a Complex Systems Science. Journal of Social Computing.
Computational Models of Social/Emotional Interactions in the Era of LLMs:
The Challenges of Transparency
Chloe Clavel
ALMAnaCH Team, INRIA Paris
November 27, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Abstract: Research on AI and social interaction is not entirely new — it falls within the field of social and affective computing, which emerged in the late 1990s. To understand human socio-emotional states, the research community has long drawn on both artificial intelligence and social science. In recent years, however, the field has shifted toward a dominant focus on generative large language models (LLMs). These models are undeniably powerful but often opaque. In this talk, I will present our current work on developing machine learning approaches — from classical methods to LLMs — for modeling the socio-emotional layer of interaction, with a particular focus on improving model transparency. I will also briefly present some of the applications we are developing to support human skill development, particularly in the fields of education and health.
Chloe Clavel is a Senior Researcher in the ALMAnaCH team at Inria Paris, the French national research institute for digital science and technology. Her research interests lie in the areas of Affective Computing and Artificial Intelligence, at the crossroads of multiple disciplines including speech and natural language processing, machine learning, and social robotics. She works on computational models of socio-emotional behaviors — such as sentiment, social stance, engagement, and trust — in both human–human interactions (e.g., conversations in social networks or face-to-face settings) and human–agent interactions (e.g., conversational agents and social robots).
Chhun, C., Suchanek, F. M., & Clavel, C. (2024). Do language models enjoy their own stories? prompting large language models for automatic story evaluation. Transactions of the Association for Computational Linguistics, 12, 1122-1142
Hemamou, L., Guillon, A., Martin, J. C., & Clavel, C. (2021), Multimodal Hierarchical Attention Neural Network: Looking for Candidates Behaviour which Impact Recruiter’s Decision, IEEE Trans. of Affective Computing
Nouri, C., Cointet, J. P., & Clavel, C. Graphically Speaking: Unmasking Abuse in Social Media with Conversation Insights. ACL 2025
Wan, C., Labeau, M., & Clavel, C. EmoDynamiX: Emotional Support Dialogue Strategy Prediction by Modelling MiXed Emotions and Discourse Dynamics. NAACL 2025
Chadi Helwe, Tom Calamai, Pierre-Henri Paris, Chloé Clavel, and Fabian Suchanek. 2024. MAFALDA: A Benchmark and Comprehensive Study of Fallacy Detection and Classification. NAACL 2024
M. Hulcelle, L. Hemamou, G. Varni, N. Rollet and C. Clavel, “Leveraging Interactional Sociology for Trust Analysis in Multiparty Human-Robot Interaction,” International Conference on Human-Agent Interaction, HAI 2023
Is it really easier to build a child AI than an adult AI?
Est-ce vraiment plus facile de créer une IA enfant qu’une IA adulte?
Emmanuel Dupoux
Laboratoire de Sciences Cognitives et Psycholinguistique, EHESS
December 4, 2025 ,10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
Abstract: This talk reexamines Turing’s proposal to achieve machine intelligence by building an artificial child. With language acquisition as a testbed, I examine whether recent advances in self-supervised learning and large language models applied to child-centered audio or audio/video data take into account early phonetic and lexical developmental landmarks in real children. Focussing on the issues of robustness and data efficiency in child language learning, I will recast the long-standing controversy between statistical learning, social approaches and nativist hypotheses as an investigation of inductive biases in AI models in the light of ecologically realistic data.
Résumé : Dans cette présentation, je pars de l’idée formulée par Turing selon laquelle il serait plus facile de construire une “intelligence artificielle” en simulant un enfant apprenant qu’un adulte déjà éduqué. Prenant l’acquisition du langage comme banc d’essai, j’examinerai comment les avancées récentes en matière d’apprentissage auto-supervisé et de grands modèles de langue appliqués à des données audio ou audio/vidéo centrées sur l’enfant permettent ou non de reproduire les premiers jalons du développement phonétique et lexical chez l’enfant. J’aborderai les questions de robustesse et d’efficacité des données dans l’apprentissage du langage chez l’enfant, en reformulant les vénérables controverses entre l’apprentissage statistique, l’apprentissage social et le nativisme comme une investigation des biais inductifs dans les modèles d’IA confrontés à des données écologiquement réalistes.
Emmanuel Dupoux is Professor at the École des Hautes Études en Sciences Sociales (EHESS) and directs the Cognitive Machine Learning team at the Laboratoire de Sciences Cognitives et Psycholinguistique (LSCP, ENS/CNRS/EHESS). He is also a part-time scientist at Meta AI Research. His research focuses on the mechanisms underlying cognitive and linguistic development in infants, combining experimental psychology, brain imaging, and machine learning. He holds a PhD in Cognitive Science (EHESS), an MA in Computer Science, and a BA in Applied Mathematics (ENS). He is recipient of an ERC Advanced Grant and organizer of the Zero Resource Speech Challenge, developing computational approaches to understanding how children learn language from their environment.
Emmanuel Dupoux est professeur à l’École des Hautes Études en Sciences Sociales (EHESS) et dirige l’équipe Cognitive Machine Learning au Laboratoire de Sciences Cognitives et Psycholinguistique (LSCP, ENS/CNRS/EHESS). Il est également chercheur à temps partiel chez Meta AI Research. Ses recherches portent sur les mécanismes du développement cognitif et linguistique chez les nourrissons, combinant psychologie expérimentale, imagerie cérébrale et apprentissage automatique. Il détient un doctorat en sciences cognitives (EHESS), une maîtrise en informatique et une licence en mathématiques appliquées (ENS). Lauréat d’une bourse ERC Advanced Grant et organisateur du Zero Resource Speech Challenge, il développe des approches computationnelles pour comprendre comment les enfants apprennent le langage à partir de leur environnement.
References:
Robot Learning from Demonstration
Apprentissage robotique par démonstration
Sylvain Calinon
Robot Learning & Interaction Group, Idiap Research Institute
December 11, 2025 10:30 – noon EDT/EST
Zoom: https://uqam.zoom.us/j/86058204969
Université du Québec à Montréal (UQÀM)
English Abstract: This talk explores how robots can efficiently acquire complex manipulation skills from minimal human demonstrations, addressing one of the fundamental challenges in modern robotics. I will present approaches that exploit the inherent structure and geometry of demonstration data to enable few-shot learning, moving beyond traditional imitation learning that requires extensive datasets. The discussion will cover representations for manipulation skills that can capture task variations and coordination patterns, optimal control techniques that bridge learning and control, and intuitive interfaces for meaningful human-robot interaction. Key topics include learning on Riemannian manifolds to handle orientation and manipulability constraints, tensor methods for exploiting multidimensional sensorimotor data, and bidirectional interaction strategies that allow robots to actively collect better demonstration data. I will demonstrate applications ranging from industrial manipulation tasks to assistive robotics, showing how robots can adapt learned skills to new situations and perturbations. The talk will address both the theoretical foundations of demonstration-based learning and practical considerations for deploying such systems in real-world scenarios.
French Abstract: Cette conférence explore comment les robots peuvent acquérir efficacement des compétences de manipulation complexes à partir de démonstrations humaines minimales, abordant l’un des défis fondamentaux de la robotique moderne. Je présenterai des approches qui exploitent la structure et la géométrie inhérentes des données de démonstration pour permettre l’apprentissage à partir de peu d’exemples, dépassant l’apprentissage par imitation traditionnel qui nécessite des ensembles de données étendus. La discussion couvrira les représentations pour les compétences de manipulation qui peuvent capturer les variations de tâches et les motifs de coordination, les techniques de contrôle optimal qui relient apprentissage et contrôle, et les interfaces intuitives pour une interaction humain-robot significative. Les sujets clés incluent l’apprentissage sur des variétés riemanniennes pour gérer les contraintes d’orientation et de manipulabilité, les méthodes tensorielles pour exploiter les données sensori-motrices multidimensionnelles, et les stratégies d’interaction bidirectionnelle qui permettent aux robots de collecter activement de meilleures données de démonstration. Je démontrerai des applications allant des tâches de manipulation industrielles à la robotique d’assistance, montrant comment les robots peuvent adapter les compétences apprises à de nouvelles situations et perturbations.
French Bio: Sylvain Calinon est chercheur senior à l’Institut de recherche Idiap à Martigny, Suisse, et chargé de cours à l’École Polytechnique Fédérale de Lausanne (EPFL). Il dirige le groupe Robot Learning & Interaction à Idiap, avec une expertise en collaboration humain-robot, apprentissage robotique par démonstration, et optimisation basée sur des modèles. Son travail se concentre sur les applications robotiques centrées sur l’humain où les robots acquièrent de nouvelles compétences à partir de peu de démonstrations, développant des modèles qui exploitent efficacement la structure et la géométrie des données.
References:
Shetty, S., Lembono, T., Löw, T., & Calinon, S. (2023). Tensor Train for Global Optimization Problems in Robotics. IEEE RAS Best Paper Award.Jaquier, N., Rozo, L., Calinon, S., & Buerger, M. (2019). Bayesian Optimization Meets Riemannian Manifolds in Robot Learning. Conference on Robot Learning (CoRL) – Best Presentation Award.
Understanding LLM Understanding 