mailwal

October 2026 Newsletter ⏭⚖⛰ Active Inference Institute

Active Inference Institute

Read on to learn more, and complete the Measurement form yourself to have your research, education, and work updates included in a future newsletter.

Symposium in 2 weeks!

  • The 6th Applied Active Inference Symposium will be held online November 12–13, 2026 — free to attend; register here. The full program is live: 53 confirmed contributions from more than 50 presenters, including a keynote by Karl Friston (Nov 12, 13:00 UTC); 13 livestreamed sessions with Alison Gopnik, Mike Levin, Ryan Smith, Takuya Isomura, and many more; 3 hands-on interactive workshops; and 36 pre-recorded talks. Topics range from computational psychiatry and zebrafish decision-making to consciousness, narrative, design, planetary-scale agency, and global governance. Both days stream on YouTube from 12:00 UTC (Day 1, Day 2). We are still seeking Sponsors (more information), who will be acknowledged in the program. Contact us with any ideas or questions.

Updates from the Institute

  • The Fundamentals of Active Inference textbook group continues to meet weekly. All details and registration form here, the playlist of all meeting recordings is here, our interactive learning document is here, and all open source code examples are here. The textbook is “Fundamentals of Active Inference: Principles, Algorithms, and Applications of the Free Energy Principle for Engineers” by Sanjeev V. Namjoshi (2026). Great work to Andrew Pashea and Fraser Paterson for their skilled facilitation!

  • Do you have scientific or technical expertise you’d like to contribute to an open research and education community? The Active Inference Institute is accepting applications for its Scientific Advisory Board, a volunteer role that works alongside the Board of Directors and Officers and focuses on scientific direction and rigor. We welcome applicants from across the fields that Active Inference touches. Details are at http://sab.activeinference.institute/ and the application is at http://sabform.activeinference.institute/

  • We are also looking for Officers for 2027. This is an exciting, meaningful, operational volunteer role. We especially welcome those with financial, non-profit, and grant administration experience — and a passion for using those skills to help those in our ecosystem Act, Infer, Serve! Email [email protected] to share your interest and learn more.

Updates from the Active Inference Ecosystem

  • A Cladistic Taxonomy of Cognitive States: An Integrated Model of Active Inference and Epistemic Parity project (Ben Ruby) reports a new paper, “A Hierarchical Thermodynamic Taxonomy of Doxastic States: An Active Inference Model, Neuro-Computational Opportunity Cost Architecture, and EEG-fMRI Fusion Paradigm”. The paper models cognitive states (belief, disbelief, suspended judgment) as energy constrained reality engines using the Free Energy Principle. It details an fMRI paradigm to isolate the thermodynamic cost of maintaining objective neutrality (symmetrical precision weights via the dlPFC) against subcortical biases. The work aims to support independent, open science contributions to active inference frameworks.

  • Active Inference for Food Systems Transformation project (Ben Elers within the UNDP Food Systems group) seeks to leverage Active Inference / Predictive Processing to understand behaviour and change across global food systems, from farmers to policymakers. The approach is non-computational: it takes the underlying principles of Active Inference and Predictive Processing and explores whether they are accessible and useful to a non-technical audience. It is also a potential test-bed for Active Inference’s scale-free claims: does it have something equally relevant to say to people working with farmers in Indonesia, government ministers in Peru and research institutes in Uganda? Ben would welcome contact from anyone interested in collaborating or who is interested in the nexus between ActInf/PP and social change.

  • cpomdp — Continuous Active Inference for Python project (Dan Corva) reports: cpomdp v0.5.0 shipped on PyPI on 16 October 2026. The release marks the first industry example. It reports two measurements. 1) GATE-D4, the accuracy gate on the exact reference filter. 2) Anti-surge margin on a centrifugal compressor.

  • Two new video walkthroughs explore geospatial Active Inference, connecting executable models in GEO-INFER, explicit specifications in Generalized Notation Notation (GNN), and formal mathematics in FEP-formal. Homeward follows a simulated salmon through the Smith River watershed, pairing inference and action with Lean-checked model properties. Active Inference Across California examines how limited monitoring resources can be allocated across fire, seismic, and drought indicators, revealing tradeoffs between information gain and coverage. Together, the demonstrations offer a reproducible research foundation while clearly distinguishing measured geography, simulated behavior, and verified mathematical claims.

  • Daniel Friedman reports the open source Active Inference Battleship codebase, which brings active inference to a familiar game, using Generalized Notation Notation (GNN) to specify an agent that updates beliefs about hidden fleets and balances preferred outcomes with information gain. The accompanying paper derives the model and reports a reproducible 1,280-search comparison of five strategies, examining fleet completion, early hits, and uncertainty reduction within a finite catalog of legal fleets. Try the interactive dashboard to play against the agent, inspect its beliefs and action scores, and explore how preferences, uncertainty, and information seeking shape its decisions.

  • Daniel Friedman reports the Active Inference Language Model (AILM), with an accompanying paper and open source code. AILM uses language as a testbed linking uncertain perception, structured representation, and goal-directed action, via explicit discrete generative models at six levels: letters, phonemes, words, case roles, sentence types, and document position. Its probabilities, preferences, habits, and policy precision have declared meanings and inspectable sources, and the generative core uses no transformer, neural network, or learned embedding. GNN specifies the models; pymdp runs inference and, for generation, selects policies under expected free energy. A gate-audit method tests results against trivial baselines and deliberately broken variants. AILM offers a bounded, inspectable complement to large language models for studying specified linguistic mechanisms from an energy-based cognitive perspective.

Audio-Visual Production.

Big thanks to the many of you — Fellows, interns, and others in the ecosystem — who are doing awesome work and submitting updates!

This page gives a 1-page summary of the structure of the Institute — a helpful entry point for those who are new, or learning about our form and functions. Check out the ways to get involved with the Institute and Ecosystem this year.

Donate to the Institute to support our impact and sustainability. Contact us with other philanthropic and grant ideas.

Get in touch if you have general or specific feedback, ideas, or questions for the Institute.

More to come! Stay tuned via the newsletter and Discord.

Officers,
Alexandra & Daniel

© 2026 Active Inference Institute
751 2nd Street #82, Crescent City, CA, 95531
USA, Earth