May 2023 Newsletter 🌊📢🦌 Active Inference Institute
May 2023 updates from the Active Inference Institute.
Welcome to the May 2023 updates from the Active Inference Institute.
Read on to see updates from the Institute & opportunities for your engagement.
Institute scale
Applied Active Inference Symposium in August 2023 — Call for co-organizers. This year, we're thrilled to announce that the theme of our symposium is projected to be: “Enacting Ecosystems of Shared Intelligence: Active Inference as a Cognitive Kernel” (exact title and framing subject to the approval of the co-organizer team). If you are keen to contribute your expertise and become a part of our dynamic co-organizing team, we warmly invite you to reach out to us at ActiveInference@gmail.com before June 7th.
Our Officers are continuing to work towards obtaining 501c3 tax exempt status for the Active Inference Institute.
We are applying in the coming months for the NSF grant “Pathways to Enable Open-Source Ecosystems (POSE)”, for the Institute to serve as a managing organization in the growing Active Inference ecosystem. Please get in touch with us if you would be interested in partnering here!
We are active with our Scientific Advisory Board, an informal group that provides the Institute with insight and expertise. If you are interested in joining the Scientific Advisory Board for 2024, please fill out this form.
We have initiated the development of an AII qualification system that will be utilized to validate the educational results of participants.
The Activities page has updated timing information on all the ongoing projects at the Institute. We have about 10 synchronous meetings of different types throughout the week — there are many ways to learn and apply Active Inference by contributing to research projects, textbook groups, livestreams, and more.
Education
During early 2023 we have two ongoing cohorts of the Parr et al. Active Inference Textbook Group. To register for an upcoming cohort (Cohort 4 will begin in June), please see here for more information and the Textbook Group registration form.
The Chris Fields course on "Physics as Information Processing" is active. There will be 6 Lectures, alternating with 6 participatory discussions. All materials and information on the course can be found on our website.
We had many exciting livestreams in May and more are coming up in June:
June 1 at 22 UTC, GuestStream #015.3 with Bobby Azarian on “The Teleological Stance: The Free Energy Principle as a Basis for a Scientific Self-Help System”
June 13 at 0 UTC, GuestStream #044.1 with Naotsugu Tsuchiya and Hayato Saigo “Using category theory to assess the relationship between consciousness and integrated information theory”
June 19 at 18 UTC, GuestStream #045.1 with Maxwell Ramstead and Mahault Albarracin “The inner screen model of consciousness: applying the free energy principle directly to the study of conscious experience“
The Active Inference Journal continues to produce augmented transcripts and translations for every livestream, as a resource for education and research. To stay updated with the current state of the Journal, see the Github repository of the Active Inference Journal (and if you are interested to see or improve on our methodology, see the Github repository of the Journal Utilities). There is also a special Journal announcement below.
Research
The Active Blockference project continues on an exciting path, curating implementations of Active Inference and integrating them with the complex systems simulation framework cadCAD.
We continue development on the Active Inference Ontology, bringing accessibility and rigor to our growing field! Special progress during May 2023 was made on applications of the Active Inference Ontology to philosophical questions, and providing correct/incorrect examples for future learning.
Our Interns are pursuing their interesting research and application directions.
Let us know if you have any recent publications (papers, blogs, etc) that we can share in future newsletter updates.
Update from Robotics & Embodied Project:
The Robots & Embodied project sits at the intersection of Minsky's Society of Mind, Active Inference, and Symbolic Machine Learning. A prior iteration implemented hierarchical predictive processing onto Lego robots using a functional language (Elixir). Learning was limited to refining action policy selection from experience (see https://zenodo.org/record/6862626).
The current iteration aims at greatly expanding the learning capabilities of a robot. This is to be achieved by dynamically growing a robot's Society of Mind - a collective of cognition actors - and their aggregate predictive/policy selection capabilities from the experience they gain engaging in exploration/exploitation activities that they themselves cause.
The project considers Active Inference a normative (as-if) framework and not an implementation (as-is) framework. The project's implementation uses a form of Symbolic Machine Learning to discover, from a history of observations, logic programs that embody valid causal theories that the individual cognition actors then run to infer predictions. The discovery of these logic programs/causal theories is done by an Apperception Engine (see https://arxiv.org/abs/1910.02227) re-implemented in this project using Prolog + Constraint Handling Rules.
One of the key difficulties of using an Apperception Engine is discovering, in reasonable time, good-enough logic programs/causal theories hidden in a massive haystack of potential solutions. This is a hard problem. It so happens that Cognitive Science as a whole is hard for the very same reasons, according to the paper "How hard is cognitive science?" by Patricia Rich et al. (see https://psyarxiv.com/k79nv/).
In the upcoming Robotics & Embodied meeting (date and time TBD, reply to this message to be included in scheduling announcements), we will discuss why what makes discovering causal theories by a robot hard is also what makes Cognitive Science hard.
Update from the Active Inference Journal:
At the Active Inference Journal, we are eagerly seeking contributors who would like to help improve the transcripts, named entities list, and topics discussed in the ongoing Chris Fields course “Physics as Information Processing”.
If you are already planning to spend time on this course, we hope you’ll consider working with the Journal as well! This is a unique opportunity to build editorial skills & make high-leverage Open Science contributions to the Active Inference ecosystem.
Specifically, this work entails listening carefully to the course lectures, and making corrections to the AI-generated transcript (from AssemblyAI). For example, for the transcript of the first Lecture, see here. We are also building out the list of people and topics named which will improve the scholarship and relevance of the work.
All transcripts and materials related to the course will be published open source (e.g. see earlier work by the Journal on Mark Solms livestreams) so that this content can be archived, citable, and translated into different languages.
We have synchronous meetings for the Journal in the Institute Discord on Mondays at 13 UTC; most of the work is done asynchronously via the Github links above.
If you are interested in participating in the Journal on this project, please let us know by joining the Journal channel in our Discord, or replying to this message.
How to participate in 2023
See all participation onboarding information.
Join the AII Discord to get involved with our community and join meetings in the voice chat.
As a community-driven open science organization, we provide multiple opportunities for deepening your relationship with the Institute (Informal, Volunteer and Intern). All backgrounds, time zones, time availability, and level of familiarity with Active Inference are welcome and encouraged at AII.
Volunteers are active learners who want to contribute to ongoing projects at the Institute. There is no limit to the type or number of projects that Volunteers can be involved in and drive —see more information and fill out the Volunteer registration form here.
Interns are Volunteers who are looking for a more intense and structured experience with the Institute. Interns are assigned a mentor from the Institute. If you are interested in the internship, see more information and complete the internship registration form.
Let us know via reply to this email if you have any other ideas or questions.
Thank you for the regime of attention and see you around.
Active Inference Institute