Our framework for advancing the accessibility, rigour, and applicability of Active Inference.
A world where active inference, a framework of adaptive behaviour grounded in first principles, is known, taught, and applied across science, engineering, and everyday life.
To translate the science of how natural systems adapt and self-organize, known as active inference, into practical tools, resources, and educational experiences that empower people, communities, and technology to enhance decision making and reduce uncertainty.
Our work is organised around three interconnected goals, each supported by specific programmatic initiatives.
To reduce the barriers to entry for new researchers and practitioners by developing standardised educational resources, open-source tools, and common standards.
We will support and coordinate the development of the field's core computational infrastructure. This includes stewarding the Active Inference Ontology to provide a standardised vocabulary and investing in robust, well-documented modelling toolkits (e.g., RxInfer.jl) to improve research reproducibility and accelerate model development.
The Institute will develop and disseminate a structured educational curriculum. This includes hosting the Annual Applied Active Inference Symposium, organising focused Textbook Groups and workshops, and producing modular online courses (such as related to Physics and Social Sciences). The aim is to create clearly defined learning pathways for participants at all levels, from novice to expert.
We will support projects that focus explicitly on translating Active Inference principles and methods for specific scientific domains (e.g., social sciences, clinical psychology, economics). This involves creating domain-specific glossaries, tutorials, and model libraries to facilitate effective transdisciplinary research.
To move the field's dominant mode of inquiry from demonstration and theoretical elaboration towards rigorous, comparative, and falsifiable empirical testing.
This programme will directly fund, facilitate, and promote pre-registered research that tests unique, quantitative predictions of Active Inference models. A primary focus will be on comparative studies that benchmark AIF-based models against established alternative frameworks (e.g., standard Reinforcement Learning, predictive coding models) using standardised datasets and transparent evaluation metrics (e.g., Bayesian model selection, cross-validation).
We will lead the development of open-source computational and experimental benchmarks. These "crucible" tasks will be designed to probe the boundary conditions of Active Inference, identifying the specific domains and parameter regimes where its assumptions hold and where they fail. The goal is to establish a clear, evidence-based understanding of the theory's predictive scope and limitations.
The Institute will actively host and promote structured debates, workshops, and collaborative projects with leading researchers from alternative theoretical traditions. This "Sceptics in Residence" initiative will ensure that the Active Inference framework is continuously challenged and refined through external critique.
To bridge the gap between preliminary, simulation-based applications and robust, deployed solutions that address concrete scientific and engineering problems.
The Institute will function as an intermediary, connecting academic research groups with industry and non-profit partners who have well-defined problems. We will provide project management and technical oversight to facilitate the co-development of proofs-of-concept and prototypes.
We will launch and coordinate multi-year, mission-oriented research programmes targeting high-impact societal challenges. Initial programme areas are proposed as follows:
AI Safety and Alignment – Investigating how the Active Inference formulation of agency and world-modelling can contribute to the development of robustly safe and beneficial AI systems.
Computational Psychiatry – Developing and validating generative models of psychopathology (e.g., PTSD, psychosis) to serve as in silico platforms for testing therapeutic interventions.
Complex Ecological and Systems Modelling – Applying Active Inference to model and manage complex, multi-scale systems relevant to climate science and systems biology.
The Active Inference Institute is committed to fostering an open, collaborative, and scientifically rigorous research ecosystem. We invite researchers, engineers, academic institutions, and industry partners to engage with our programmes and contribute to the principled advancement of this field.