Skip to content
3rd Applied Active Inference Symposium

icon picker

Enacting Ecosystems of Shared Intelligence

3rd Applied Active Inference
August 22, 2023,

See to get email reminders and calendar invites.
Symposium activities can be watched live and rewatched at the same links →

See the below ⬇️

2023 Symposium Program
Not synced yet
UTC Time
Event Name
8/22/2023, 1:00 AM
André Bastos
Multi-laminar, multi-area recordings in the non-human primate brain suggest Predictive Coding is implemented via Predictive Routing
To understand the neural basis of cognition, we must understand how top-down control of bottom-up sensory inputs is achieved. We have marshaled evidence for a canonical cortical control circuit that involves rhythmic interactions between different cortical layers. By performing multiple-area, multi-laminar recordings, we've found that local field potential (LFP) power in the gamma band (40-100 Hz) is strongest in superficial layers (layers 2/3), and LFP power in the alpha/beta band (8-30 Hz) is strongest in deep layers (layers 5/6). The gamma-band is strongly linked to bottom-up sensory processing and neuronal spiking carrying stimulus information, while the alpha/beta-band is linked to top-down processing. Deep layer alpha/beta Granger causes that in superficial layers, and is negatively coupled to gamma. These oscillations give rise to separate channels for neuronal communication: feedforward for the gamma-band, and feedback for the alpha/beta band. Attention, working memory, and prediction processing all involve modulation of gamma and alpha/beta synchronization, both within and across areas of the frontal/parietal/visual network. These rhythmic interactions breakdown during anesthesia-induced unconsciousness. Based on these observations, we hypothesize that the interplay between alpha/beta and gamma synchronization is a canonical mechanism to enable cognition and consciousness.
8/22/2023, 1:30 AM
Keith Duggar
Active Inference and the Actor Model
Active Inference and the Actor Model are two deeply connected understandings of the world, both born ahead of the time. They provide foundational frameworks for defining the dynamics of complex systems with a focus on autonomous agents that interact in an ecology of nested systems. In this content we explore some of their key connections including: the role of agents, concurrency, information processing, reactive systems, and emergent behavior. We will see that Active Inference and the Actor Model are a paradigm shift away from a deterministic, centralized, step-by-step thinking to a decentralized, networked, concurrent perspective of both computation and cognition.
8/22/2023, 2:00 AM
Sanjeev Namjoshi
Developing next-gen Active Inference tools: Broadening accessibility, educational resources, and the software ecosystem
Over the last decade, research in the Active Inference field has resulted in an enormous diversification and development of new ideas spanning multiple subdisciplines and domains. However the field is still relatively nascent despite its potential to revolutionize many industries and areas of research and applied sciences. Due to the current interest in both generative AI and reinforcement learning, the field is uniquely positioned to capture the attention of machine learning researchers and spur a proliferation of growth in both industry and academia similar to the deep learning revolution that began at the start of the 2010s. In this talk we will explore various next-generation tools that will provide educational resources and interactive media to increase accessibility and understanding of the complex concepts that have been developed in the field. These tools include a work-in-progress textbook on Active Inference, Bayesian Mechanics, and the Free Energy Principle aiming to be self-contained and spanning the full scope of the research literature. Accompanying this textbook are under-development Jupyter notebooks, simulations, video lectures, and interactive software intended to teach and educate the concepts of Active Inference and Bayesian Mechanics in order to bring these ideas to a wider audience and induce a paradigm shift leading to the next phase of both applied and theoretical AI research.
8/22/2023, 2:30 AM
Inês Hipólito
Resurrecting Gaia: Harnessing the Free Energy Principle to preserve life as we know it
We apply FEP and an active inference framework to the Gaia Hypothesis with the aim of providing novel insights into how to catalyse meaningful ecological improvement.
8/22/2023, 3:00 AM
Aswin Paul
Sophisticated Inference in pymdp
A short hands on session using pymdp to specify generative models and simulate sophisticated inference experiments from and similar to Karl Friston, Lancelot Da Costa, Danijar Hafner, Casper Hesp, Thomas Parr; Sophisticated Inference. Neural Comput 2021; 33 (3): 713–763.
8/22/2023, 4:30 AM
Takuya Isomura
Canonical neural networks for shared intelligence — an active inference modelling
Empirical applications of the free-energy principle at the cellular and synaptic levels are not straightforward because they entail a commitment to a particular process theory. We addressed this issue by developing a reverse engineering technique that allows precise linking of quantities in neuronal networks to those in Bayesian inference. According to the complete class theorem, any system that minimises a cost function can be read as Bayesian inference. In light of this notion, we showed that any canonical neural network—whose activity and plasticity minimise a common cost function—can be cast as performing (variational) Bayesian inference. By combining reverse engineering with an in vitro causal inference paradigm that we previously established, we experimentally validated the free-energy principle by assessing its ability to predict the quantitative self-organisation of in vitro neural networks. In this talk, we will also discuss a possibility to extend reverse engineering to modelling canonical neural networks for social and shared intelligence using the example of songbirds that communicate with multiple conspecifics.
8/22/2023, 5:00 AM
Shanna Dobson
Dark Imaginarium: Shared Intelligence as an Infinity-Curiosity Type
We investigate the idea of sleep as the protostate, and posit the idea of dark consciousness where dark is a 2-fold hybrid. We model dark consciousness as a 2-topos in p-adic time, and outline perfectoid and diamond-like versions. We then introduce and illustrate implications of Dark Imaginarium, which is a higher order Curiosity Artificial Intelligence, an Infinity-Curiosity Type, that thinks in infinity categories.
See: Dobson, Shanna, Dark Imaginarium: Infinity-Curiosity & Dark Consciousnesss in P-adic Time, , 2023.
8/22/2023, 5:30 AM
Nynke Boiten
Current advances in machine theory of mind and theory of mind sophistication
This session is about Active Inference approaches to Theory of Mind (ToM). ToM is the ability to reason about others thoughts and intentions (mentalization) and to flexibly predict the behavior of others, thereby facilitating effective communication and joint action. Recent advances in robotics, artificial intelligence, neuroscience, and cognitive science elucidate some of the computational mechanisms underlying ToM as an adaptive cognitive ability. However, ToM sophistication, the ability to form recursive beliefs about others internal states, is often ignored. This talk will review current active inference proposals of ToM. An active inference ToM-model of the matching pennies game, a simple strategy game, will be proposed and some ideas for implementing ToM sophistication will be discussed.
8/22/2023, 2:30 PM
JF Cloutier
Growing a collective of theorizers; first steps
Can a robot, by actively inferring its environment, grow and sustain a society of mind, a collective intelligence of cognition actors? I am embarked on a long term project to explore this question. My tools are symbolic machine learning and Kant's philosophy. A first step was to implement predictive processing as unsupervised discovery of symbolic causal theories. The next step will be to form a collective of theorizers that collaborate to actively maintain homeostasis.
8/22/2023, 3:30 PM
Conor Heins
Multi-agent active inference and multi-scale alignment: current developments and challenges
Active inference continues to grow in popularity as more and more disciplines, many outside of its original computational and cognitive neuroscience origins, take note of the tools and advantages it offers in the design and study of adaptive agents. Active inference is often not only cited as a powerful framework for constructing single agents, but also more ambitiously as a multi-scale, emergent account of natural complex systems, where agents compose, and are composed of, hierarchically-composed, interacting active inference sub-systems that all minimize free energy functionals in a neat cascade of nested inference processes. In this presentation I'll overview past, current, and anticipated research projects in the active inference community that have tried to test or demonstrate the validity of this claim. In general, it appears that active inference does not guarantee anything multi-scale by itself, and more research is required to understand the precise conditions under which systems of active inference agents will automatically also form a larger active inference agent (and so on). In doing so, I will also propose ways in which the active inference community can incorporate research from other disciplines (e.g. in economics, swarm robotics, and collective animal behavior) that have attempted to tackle this issue using their own research methodology, terminology and frameworks.
8/22/2023, 4:00 PM
Bert de Vries, Dmitry Bagaev, & Bart van Erp
Toward User-friendly Design of Synthetic Active Inference Agents
We will discuss technical issues on what is needed to develop the “tensorflow”-toolbox for active inference (AIF) agents. In particular, we focus on a toolbox that would enable democratization of design of AIF agents.
8/22/2023, 5:00 PM
Active Inference Institute
Active Inference Institute & Active Inference Ecosystem
We will overview and discuss the past, present, and future of the Active Inference Institute and Active Inference Ecosystem. If you are here, you’re in it!
See our recent collaborative publication “The Active Inference Institute and Active Inference Ecosystem”:
8/22/2023, 5:30 PM
Rafael Kaufmann
Gaia: An ActInf-powered network for planetary-scale sensemaking
I introduce Gaia, a decentralized active inference system that helps us make sense of, and decisions for, the real-world entities that constitute our planet. I demo Gaia’s existing capabilities, focusing on the science and practice of agroecology, briefly highlight the technology behind it, and discuss future applications.
8/22/2023, 6:00 PM
Avel Guénin-Carlut
Embedded Normativity
We introduce and motivate the concept of embedded normativity to account for the externalization of social norms in the material environment through human social activity. We ground this notion in the Active Inference framework, and more specifically through the derived Skilled Intentionality framework of ecological perception and action. This framework considers that skilled agent experience the world as a landscape of affordances, or opportunities for action. This landscape is inherently normative, as its experience is tied to the agent's anticipations over its own behaviour (and therefore, indirectly, to its motivations). We emphasize that given this framework, normativity does not exist inside or outside the agent's boundaries, but is brought about by its engagement with the world. We discuss the dynamics of internalization and externalization by which agents come to project normativity onto elements of their environment, and experience this normativity as a simple attraction toward favoured states. Given this account, we revisit earlier descriptions of the shared material and sociocultural niche enable the broadcasting and integration of norms. Finally, we discuss how embedded normativity can be brought into existence by the perception of humans, and relate our discussion to the ontological stance of participatory realism.
8/22/2023, 6:30 PM
Pablo Fernandez-Maquieira
The Adventure of Curiosity
"Exploring Active Inference through Art and Role-Playing Games" is a talk that shares my personal journey of discovering active inference via NFT art and its application in developing a role-playing game. The session will also discuss how active inference can enhance the functioning of autonomous organizations.
8/22/2023, 7:00 PM
Mahault Albarracin
Shared protentions in active inference
Understanding how coordination happens between agents across scales
8/22/2023, 7:30 PM
Karl Friston, Anna Lembke, Rafael Kaufmann, Bert de Vries, Guillaume Dumas, Curt Jaimungal
Roundtable: “Enacting Ecosystems of Shared Intelligence”
Ecosystems of intelligence Facilitated by: Curt Jaimungal With panelists: Karl Friston, Anna Lembke, Rafael Kaufmann, Bert de Vries, and Guillaume Dumas.
8/22/2023, 9:00 PM
Informal conversation in Institute
Continue the conversation in the .
We see this informal conversation session as an opportunity for Symposium participants to connect with each other, find collaborations, and learn about ongoing activities in the Active Inference Ecosystem. These conversations will be continuing to explore the themes from the presentations, and also have a special focus on welcoming new participants.

Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
) instead.