Maximilian Puelma Touzel
|
Staff Research Scientist in the Applied Machine Learning Research Team
@ Mila, Québec AI Institute Member of the Complex Data Lab I study how collective behaviour emerges in populations of learning agents, and how to infer what drives it from data. Trained as a theoretical biophysicist, I have applied the same toolkit, statistical physics, Bayesian inference and dynamical systems, to neurons, immune repertoires, and now people and AI agents. My group combines inference from large-scale text and social-media data, simulations of societies populated by LLM agents, and theory of how consensus, polarization, collusion or specialized roles emerge. The aim is to inform how human–AI societies are designed and governed, including the safety of multi-agent AI systems. Started STAMINA to build community around these directions. PhD, Physics (IMPRS Physics of biological and complex systems), University of Göttingen MSc, Physics, University of Toronto BSc, Physics & Mathematics, University of Toronto |
 cebeuq [tod] alim [ta] mtamleup |
News
Gave a talk on multi-agent theory to Surya Ganguli's group at Stanford.
Hosted the 2nd edition of our Workshop on Social Simulation with LLMs at COLM 2026 in San Francisco. Thanks to all the speakers and participants!
"From Rage to Refusal: Using Social-Media Data to Steer Hostility-Related Affect in LLMs" accepted to the BlackboxNLP 2026 workshop at EMNLP.
Two ICML main conference position papers! A call for better evals of LLM social sims, and a call to better evaluate collusion in market agents. Check them out and see us in Korea!
To celebrate 6-months, STAMINA is doing STAMINA Matters!, an info session on all our activities.
Welcome to two summer interns Shirley and Shiyan!
Gave a talk on Cognitive Effort and LLMs to the Otto lab at McGill Department of Psychology.
2nd edition of LLM social sim Workshop accepted to COLM 2026. See you in SF in October!
Participated in the IVADO Social Reasoning and the Ecology of Thought Workshop focussing on agent sociology as part of IVADO's Thematic Semester on Computational Ingredients of Reasoning. Great Event!
Kicking off the STAMINA working group.
Participated in the IVADO Agents Workshop focussing on the Current State of AI agents as part of IVADO's Thematic Semester on Autonomous AI agents.
Very happy to deliver our Workshop on Social Simulation with LLMs.
Participated in the week-long bootcamp Focussing on the Current State of AI agents as part of IVADO's Thematic Semester on Autonomous AI agents.
Presented my social simulation work and Veracity app demo papers accepted to main IJCAI conference.
Zachary presented our social simulation work at the NLP for Positive Impact workshop at ACL 2025 in Vienna.
Submitted TRUTH-RAG work to SOLAR workshop at COLM 2025.
The group welcomes Sneheel Sarangi as an incoming CS Master's student at McGill in the Fall. Sneheel will continue to work on the social simulation project. Exciting!
COLM workshop of LLM-based social simulations accepted! See you in October!
Presented Social Simulation project to the Multi-agent Seminar at Deepmind
Submitted BluePrint (BlueSky persona) work!
Social Simulation project has been awarded an FLI grant!
Social Simulation project has been awarded an IVADO exploratory grant!
Happy to present social simulation work at the Centre for the Study of Democratic Citizenship's seminar series. CRIM.
Happy to share my thoughts on misinformation as a panelist at the Information Integrity Lab's Climate Mis/Disinformation Workshop hosted at CRIM.
We present Social Simulation work at 3 NeurIPS workshops (Socially Responsible Language Modelling Research (poster), Safe Generative AI (poster), and Towards Safe & Trustworthy Agents (Oral)). Congrats to the team!
We got a Foresight Institute grant to study AI collusion.
Happy to be affiliated with the Complex Data Lab.
Happy to join the Centre for the Study of Democratic Citizenship as a research member.
Very happy on this first of a series: "A Simulation System Towards Solving Societal-Scale Manipulation" accepted at the NeurIPS Workshops: Socially Responsible Language Modelling Research, Safe Generative AI, and Towards Safe & Trustworthy Agents (selected for an oral presentation) Here is the preprint.
"Regional and Temporal Patterns of Partisan Polarization during the COVID-19 Pandemic in the United States and Canada" submitted! Here is the preprint.
ClimateMatch running again this year. Continuing as Day Lead for the socioeconomic curriculum, we've revamped the curriculum for the day. We leaned a fair bit on the Enroads simulator. Try it out.
Workshop paper on Scalable approaches to the theory of many minds problem accepted to the Agentic Markets Workshop at ICML .
Carbon tax topic modelling work published in Environmental Data Science .
Gave a talk on using topic models on survey responses to infer ideology underpinning carbon tax opposition. This was at the CCAI workshop AAAI on Artificial Intelligence and Climate: The Role of AI in a Climate-Smart Sustainable Future at the AAAI Fall symposium in Washington, DC.
Climate Match Academy is alive and running! Delivered a day of socioeconomic curriculum and organized our Discord server of 100s of students from over 100 countries! Check out our always-accessible-from-anywhere Jupyter Book.
Presented game theory of social norms work to Joel Leibo's group at Deepmind.
Presenting topic modelling at NeurIPS Workshop Tackling Climate Change with Machine Learning.
Giving a talk (open to the public) at University of Washington's data science seminar (recording).
Attended the IBM/DIMACS Workshop on Bridging Game Theory and Machine Learning for Multi-party Decision Making at Rutgers.
Giving a talk at the Montreal Computational & Quantitative Linguistics Lab .
Presenting carbon tax work for the first time at the Montreal AI Symposium!
Meriem's NoisET paper got accepted in Physical Chemistry A!
Our mixing times paper was accepted to NeurIPS!
RLDM2022 was great! Presented urgency work and ran our workshop on social alignment in human and machines. Workshop recording available on workshop website.
Urgency work published in PLoS Computational Biology.
Presented our agent abstraction paper at ICLR workshop: From cells to societies: learning across scales.
Gave a couple guest lectures in graduate-level math course on dynamical systems at UdeM. So nice to have the time in a talk to spell things out and discuss with students!
Lyapunov spectra for RNNs paper accepted to Frontiers in Applied Mathematics & Statistics
2 accepted submissions to COSYNE: the neurodata validation of our decision-making model; and new work on noise robustness in recurrent neural nets with Colin Bredenberg.
Gave a talk on computing with transients at the Banff workshop on Dynamical Principles of Biological and Artificial Neural Networks.
Presented polynomial mixing times work at EcoRL workshop at NeurIPS. See the preprint.
Participated in Montreal's MAIN neuroAI conference.
Presented poster at Montreal's AI Symposium.
post on Mila blog post on our NeuroAI reading group.
Gave a talk on Stochastic Thermodynamics of learning to the Physics of Machine learning reading group at Mila.
Urgency work peer-reviewed and accepted at COSYNE.
Join us for a exciting day of talks and a panel of top experts on the goals and challenges for robust scientific explanations in neural and artificial intelligence systems.
Urgency work peer-reviewed and accepted at the Biological and Artificial Reinforcement Learning workshop at NeurIPS.
Presented urgency work at the inaugural NeuroAI conference, NAISYS, at Cold Spring Harbor Labs.
Happy to lead the breakout session on higher cognition at UNIQUE's inaugural NeuroAI symposium.
Check out our preprint on Lyapunov spectra for RNN training.
Our work on inferring population dynamics from intrinsically variable, vastly subsampled, and indirectly accessed genetic sequences is published in PLoS Computational Biology.
We are organizing a NeuroAI workshop at NeurIPS2019.
nnRNN paper presented in main track of NeurIPS.
With yet another wonderful edition of the Montreal Artificial Intelligence and Neuroscience (MAIN) conference, Montreal is further establishing itself as the hotbed for incisive NeuroAI research. Honoured to have my new postdoc work recognized with two awards here.
We are organizing a Physics and AI Workshop in Montreal.
family_size+=1!
Summary
My research asks
- how resource limits and incentives shape beliefs and decisions at the individual level,
- how beliefs, narratives and social norms spread and compete in populations, and
- how collective behaviour, such as coordination, specialization and collusion, emerges in populations of humans or AI agents, and increasing both.
Four connected lines of research, with an example paper from each:
- Inferring beliefs: Ideology from topic mixture statistics: Inference method and example application to carbon tax public opinion
- Tracking changes in beliefs: Regional and Temporal Patterns of Partisan Polarization during the COVID-19 Pandemic in the United States and Canada
- Simulating societal-scale belief dynamics: SandboxSocial: A Sandbox for Social Media Using Multimodal AI Agents
- Theory of collective states in agent populations: Measurement-Induced Role Differentiation in Agent Populations
