June 2025 Newsletter 🎁🏄🐛 Active Inference Institute
Updates from June 2025 & Looking towards the second half of the year
See today’s quarterly roundtable for these updates in video form.
Activities at the Institute will mostly pause during July. From August through November 2025, we will have a second semester of activities, resuming with re-activating projects only, followed by a winter break for December 2025. Read on for many exciting updates!
Updates from the Institute
We are happy to share that in June 2025, the Survival and Flourishing Fund has awarded a $270,000 grant to the Active Inference Institute to support work on the AI Capabilities & Alignment Consensus Project (AICACP). AICACP is a multi-year initiative designed to reshape the conversation around AI capabilities, alignment, and regulation. By combining high-impact journal collections, in-person discussion-oriented workshops, and academic media content for public outreach, the project aims to bridge the divide between AI “doomers” and “accelerationists” through deeply exploring the meanings of “world models” and “agency,” and what these concepts mean for AI development.
Adam Safron (at Allen Discovery Center at Tufts University) is the creator, principal investigator, and organizing editor for this project. This grant supports an expanded team to assist with efforts to manage the special issues and workshops, including these initial contributors: Victoria Klimaj (Lead Assistant Editor), Michael Garfield (Media Lead and Workshop Organizer), Amelia Thomley (Workshop Project Supervisor), Caitlin Lynch (Workshop Experience Designer). For those interested in learning more, please check out this document.
We are currently in planning stages for the 5th Applied Active Inference Symposium, to be held in November 2025. At this point we are seeking Presenters (who would like the present work or lead a workshop), and Sponsoring organizations (who would like to support the work with financial or in-kind donations). Email us with subject line [SYMPOSIUM] if you have any interest or questions here.
The Theoretical Neurobiology (TNB) Group meetings continue. Information on how to join these weekly meetings can be found at the project page, and recordings of past meetings can be viewed at the YouTube channel.
Some highlight livestreams from the last month included:
GuestStream #110.1 with Takuya Isomura on Triple equivalence for the emergence of biological intelligence
GuestStream #112.1 with Hadi Vafaii on Brain-like variational inference
OrgStream #009.1 with Manuel Hoffmann on The Value of Open Source Software
Updates from Projects
From the Active Inference Account of Belief Updating in PTSD project, Andrew Pashea (programmer, co-lead), Jeremy Cooper (researcher, co-lead), Haeun Sun (programmer), Angelos Krypotos (researcher) report: We prepared, based upon our current draft work and preliminary model results, extended abstract and then (anonymously) submitted it to the 6th International Workshop on Active Inference (IWAI) 2025.
Ian Tennant reports: Following the 4th Applied Active Inference Symposium I discovered that one of the presenters, Leonardo Muller works here are ARU at the Cambridge Institute for Music Therapy Research (CIMTR). We discussed our shared interest in Active Inference and collaborated to record interviews for each other’s YouTube channels (on Inner Sense, and on Music Minded) and are now discussing a paper on Archetypes. We have met two milestones so far (recording interviews) and have unmet milestones for writing a paper.
Daniel Friedman reports: There have been major updates made to the open source Generalized Notation Notation (GNN) repository. Some of these updates were covered in Active InferAnt Stream #014.2 “GNN for Generative Model Supply Chains: A Golden Spike Moment for Multiagent Trajectory Planning with RxInfer.jl”, and Active InferAnt Stream #014.3 “The Sound of Uncertainty: Auditory Rendering of Generative Models in the Field of Streams”.
In the project The Einstein Model of a Solid as a Model of the Mental Apparatus from the Economic Perspective of Psychoanalytic Theory, Theodoros Aliferis, along with Dr. Tiziano Colibazzi (Associate Clinical Professor of Psychiatry at Columbia University) and Professor Konstantinos Anagnostopoulos (NTUA) reports: Continuous collaboration and exchange of ideas for manuscripts to be published to reputable journals.
Research Fellow updates:
From the Symbolic cognitive robotics project, Jean-François Cloutier reports: I updated the self-published paper on my implementation of an Apperception Engine. I split my sprawling Agency repo into separate Prolog repos for my implementation of an Apperception Engine, an Actor framework, logging and other utilities, and agency proper - it uses the Actor framework. I am currently coding the ground level of a society of mind (a collective intelligence), namely sensor and effector cognition actors (CAs). Coding brings clarity. I think I have a good handle on the lifecycles of sensor and effector CAs and on how they interact with the CA's one layer up in the agent’s society of mind. I am also iterating on the Actor framework whenever I find it to be lacking. A detailed function design of Artificial Agency was completed. Implementation is under way. The project is advised by Elliott Hauser and Avel Guénin-Carlut.
From the CogNarr Ecosystem: Facilitating Group Cognition at Scale project, John Boik reports: There have been group meetings and developments with John B., Tim L, Jesse, Jeff S., Lee E., Michael L. Updates include: reading project materials, participating in group discussions, co-drafting of follow-on plans and analysis). Dissemination of project materials, group discussion, development and implementation planning. Drafting and sharing of project/technology development and implementation plans in PDF form to inform and invite others regarding contribution and opportunities for CogNarr project impact/application. Refined scope and operational definition of "stick-figure" implementation prototype.
Intern updates:
Michele Vannucci, with Peter Bloem as thesis supervisor, reports: After an initial thesis proposal which focused of studying active inference in an agent whose internal states and dynamics are defined by a cellular automaton, we pivoted the research to a more theoretical discussion on intrinsic motivations as surprise minimization or knowledge seeking within the AIXI model setting. Updated versions of the thesis draft can be found at https://michelevannucci.me/mscthesis
Satyaki Maitra and Harshil Shah report: Over the course of the project, we have slowly crafted and improved an implementation of an active inference model deployed in a Microsoft AirSim drone with the goal of eventual real-world implementation. With the newly implemented sensor fusion and final improvements to the model, we have the tools necessary to begin moving to a real drone, and will begin this new phase of the project. Here is a brief summary of the project: We developed a hierarchical active inference framework in AirSim that uses LiDAR and/or depth camera data to infer latent suitability maps for navigation. The drone minimizes expected free energy (EFE) to select feasible, adaptive trajectories in complex environments, achieving high simulation success rates in GNSS-denied scenarios.
Allegra Grunberg reports: This research presents a novel theoretical framework for transforming the employment ecosystem through the integration of the Free Energy Principle (FEP), collective intelligence methodologies, and the Adaptive Socio-Ecological System (AdSES) framework, operationalized via the SenseMaker tool. The study addresses critical misalignments between educational provision, labor market demands, graduate preparedness, and recruitment processes within the UK employment ecosystem, utilizing UCL STEaPP as a paradigmatic case study. The proposed framework employs FEP-based modeling to minimize stakeholder uncertainty, optimize goal alignment, and enhance job-matching and skill development processes. Through systematic collection and analysis of micro-narratives from 100 participants—comprising students, alumni, faculty, and employers—the research establishes a continuous feedback mechanism that identifies emergent properties within the employment system. This approach enables both reactive adaptation to current economic conditions and proactive anticipation of future labor market demands.
Support and Engage
All activities at, and products from, the Institute are provided open source and without any financial barrier. Donate to the Institute to support our work and sustainability in the years to come.
Here is the entry point for 2025, giving ways to get involved with the Institute and Ecosystem this year.
Subscribe to or share the newsletter, or contact us with other philanthropic and grant ideas. The Affordances page lists targeted opportunities for volunteer contributions.
Get in touch if you have general or specific feedback, ideas, or questions for the Institute.
Officers 🎆