我们做什么
Active Inference Institute 通过识别、建立、搭建支撑与提供支持,推动以下工作的可持续实施:教育与研究服务;研究院及 Active Inference 生态系统内参与、沟通、咨询和治理的机会与条件;确立社群成果开源、合理使用与有效传播规范的出版及许可规程;以及包括沟通、教育计划和协作项目在内的生态系统支持服务。
我们承载研究院的计划与项目,提升生态系统项目的可见度,并管理和维护 Active Inference 领域的公共信息资源。
此页面由机器翻译成中文。 查看英文原版。
寻求机构导向的访客
一个支持与Active Inference相关的教育、研究、培训和应用的参与式开放科学机构。
在本页
Active Inference Institute 是一个由志愿者领导的 501(c)(3) 非营利组织,致力于提高 Active Inference 的可访问性、严谨性和适用性。我们欢迎来自不同背景、时区和熟悉程度的人们。
Active Inference Institute 通过识别、建立、搭建支撑与提供支持,推动以下工作的可持续实施:教育与研究服务;研究院及 Active Inference 生态系统内参与、沟通、咨询和治理的机会与条件;确立社群成果开源、合理使用与有效传播规范的出版及许可规程;以及包括沟通、教育计划和协作项目在内的生态系统支持服务。
我们承载研究院的计划与项目,提升生态系统项目的可见度,并管理和维护 Active Inference 领域的公共信息资源。
研究院起源于 2020 年一次围绕对 Active Inference 的共同兴趣召开的联合创始人团队会议。“Active Inference Lab”(ActInfLab)从这些讨论中形成,并于 2021 年开始运作,成为一个由志愿者主导、致力于提高 Active Inference 可及性、严谨性和适用性的社区。
2022 年年中,ActInfLab 实现了发展上的跃迁,成为 The Active Inference Institute,在美国特拉华州注册为非营利组织,并通过设立董事会、管理人员和科学咨询委员会确立正式治理结构。2024 年,在 Fried Frank 律师事务所的公益法律服务支持下,研究院正式获得 501(c)(3) 非营利组织资格认定。
完整的逐年记录请参见“历史”页面,其中包括 Textbook Group、Applied Active Inference Symposium 系列,以及“Institute and Ecosystem”动态更新文档。
我们的愿景是:active inference 这一基于第一性原理的适应性行为框架,在科学、工程和日常生活中广为人知、得到传授并付诸应用。我们的使命是,将关于自然系统如何适应和自组织的科学——即 active inference——转化为实用工具、资源和教育体验,帮助个人、社区和技术改善决策并减少不确定性。
我们的价值观与原则包括 Active Inference 与探索、诚信、研究院及生态系统尺度的生成建模、预见性行为、持续发展和参与式投入。我们践行开放科学:使研究可复现、便于检索浏览,在开放许可下发布成果,并妥善维护共享基础设施。
背景、学科和熟悉程度的多样性是社区的特色,而不是限制。
研究院由董事会治理,管理人员负责领导日常运营,科学咨询委员会承担科学咨询职能。两个组织部门负责开展研究院的各项计划:负责教育的 EduActive 部门和负责研究的 ReInference 部门,两个部门均由协调人领导。
项目、志愿者、学习者和实习生在这一结构内开展工作,并与更广泛的 Active Inference 生态系统相联系。
研究所项目是参与的结构化路径:志愿者、实习、导师指导、研究奖学金、合作伙伴关系、慈善和资助。这些项目将兴趣转化为贡献,并将个人参与与研究所的研究、教育和生态系统目标联系起来。
研究所根据开放许可发布其工作,并维护公共存储库、视频档案、新闻通讯、Active Inference 本体和 Active Inference Journal。我们参与并支持更广泛的 Active Inference 生态系统——一个由研究人员、实践者和学习者组成的不断壮大的网络,他们跨越计算神经科学、认知科学、生态学、经济学和工程学等学科开展工作。
机构叙事
公共使命、愿景、价值观、历史、策略和重点领域的文字,针对该研究所。
Directions for the Institute describe ongoing areas of activity and development at the Institute scale.
The following table lists current developmental and connections with
| Direction | Method | Deliverables | Impact / Implication | Primary Focus Areas addressed |
|---|---|---|---|---|
| Research Advancement | Support core Active Inference research; Explore theoretical implications in Philosophy; Examine group cognition functionality | Research papers; Theoretical frameworks; Computational models | Deepened understanding of Active Inference; New insights at the intersection of multiple fields; Improved models of collective cognition | Research Advancement and Cross-disciplinary Expansion |
| Software Development | Improve RxInfer.jl visualization capabilities; Enhance PyMDP usability; Develop and curate examples ofDomains of Application | Updated Open Source software tools; User-friendly interfaces; Application case studies | More accessible and powerful Active Inference modeling; Increased adoption by researchers and practitioners; Practical demonstrations of Active Inference in action | Software Development and Practical Applications |
| Educational Outreach | Develop curricula for different languages and contexts; Provide courses and workshops; Increase Communications efforts | Comprehensive curriculum; Industry-focused courses; Educational materials for various skill levels | Wider accessibility of Active Inference concepts; Increased industry engagement; Growth of skilled Active Inference practitioners | Educational Outreach and Resource Development |
| Cross-disciplinary Expansion | Seek grants for cross-disciplinary AI research; Pursue features in popular science media; Focus outreach to social sciences | Grants and proposals; Media articles; Collaborative research projects | Broader adoption of Active Inference across disciplines; Increased public awareness; New applications in social sciences | Research Advancement and Cross-disciplinary Expansion |
| Community Growth | Facilitate intern-mentor connections; Encourage SAB member interactions; Foster edge interactions within community | Mentorship program; Enhanced community engagement; Collaborative projects | Stronger, more connected Active Inference community; Knowledge transfer between experts and newcomers; Innovative cross-pollination of ideas | Community Growth and Engagement |
| Public Engagement | Translate concepts for broader public; Address societal challenges through Active Inference; Provide foundations for trust and ethics in AI | Accessible content; Applied solutions to real-world problems; Ethical guidelines for AI development | Increased public understanding of Active Inference; Real-world impact on societal issues; Responsible AI development informed by Active Inference principles | Community Growth and Engagement, Public Engagement and Ethical Considerations |
| Practical Application | Develop policy appraisal methodologies; Consider ethical and cognitive security aspects; Research capabilities in various domains | Policy frameworks; Ethical guidelines; Domain-specific applications | Informed decision-making in policy; Enhanced cognitive security measures; Demonstration of Active Inference's versatility across fields | Software Development and Practical Applications |
Below, we revisit the Focus Areas for the Institute and outline some Directions for the Institute.
Essentially all Institute Projects use the shared space as a document system.
Clicking through links and documentation of Institute Projects you will find many examples of links within and across documents — this Institute & Ecosystem was written collaboratively in the shared space, and then exported for snapshot (whereas in 2023 version 1 we used a Google Document linear manuscript co-editing style).
the shared space is the primary platform for knowledge and project management at The Institute, Ecosystem, community, and individual scale. It organizes all information and content related to each project (or sub-project). the shared space is version-controlled and access-restricted, ensuring that all of our data is protected against accidental deletion and inappropriate user access. We use the shared space for storing and organizing important documents, such as policies, procedures, project plans, and meeting notes.
We follow best practices for the shared space, including: (1) creating dedicated the shared space “documents”, or work areas, for different departments or projects to ensure easy access and organization of relevant information, (2) implementing a clear folder and file structure within the shared space to maintain document organization and version control, (3) archiving unnecessary and irrelevant pages, files, and folders, and (4) granting appropriate access permissions to users, allowing them to view, edit, or comment on documents as required.
With adequate future support, the shared space will be upgraded to an Enterprise License and consultants will assist in development of templates and low-code applications for streamlining support, records and knowledge management, and project management functions. Further, an Enterprise License will allow for a variety of new mechanisms for user-access control and permissioning, and for tracking of work activity and community engagement with hosted content.
Institute participants, Officers, Volunteers, and other roles communicate with one another and with members of the community as follows:
The Institute communicates with potential partners, sponsors, and relevant constituencies through channels including:
The goal of our organizational communications plan is to provide the foundation for sustainable and accessible funding, and to work toward making Active Inference a household term, used as widely as “Machine Learning”, reflecting its demonstrable utility and impact in implementation. An ideal next step toward this goal is the professionalization of Active Inference core competencies and techniques and related competency and qualification standards.
Join the Discord:
The Active Inference Institute (AII) maintains a Discord server as its primary communication hub where all meetings, discussions, and collaborative activities take place. This digital shared spaces serves as the central nexus for the institute's diverse community of researchers, practitioners, and enthusiasts interested in active inference.
The Discord server facilitates:
As with the Institute overall, the Discord server welcomes participants from:
The server can be accessed through the link. It serves as the primary venue for all institute meetings and collaborative activities, making it an essential platform for anyone interested in engaging with the Active Inference community.
Below are some Focus Areas for the Institute, and how those are addressed by Directions for the Institute.
The Focus Areas were developed from feedback from participants, and presented here as a part of the overall milestones/snapshot.
| Focus Area | Area Description (why is it challenging, what are the risks? | Related Directions & Steps |
|---|---|---|
| Research Advancement and Cross-disciplinary Expansion | Bridging diverse disciplines and translating Active Inference concepts across fields is complex. Without this, we risk siloed knowledge, missed opportunities for innovation, and limited real-world impact of Active Inference principles. | Research Advancement, Cross-disciplinary Expansion |
| Educational Outreach and Resource Development | Active Inference involves abstract concepts and mathematical formalisms, making it difficult for newcomers to engage. Failure to address this could result in a limited pool of practitioners and researchers, slowing the field's growth and application. | Educational Outreach |
| Software Development and Practical Applications | Developing user-friendly, robust software tools for Active Inference is technically challenging. Without accessible tools, we risk limiting practical implementations and real-world testing of Active Inference models. | Software Development, Practical Application |
| Community Growth and Engagement | Maintaining a cohesive, productive community across diverse backgrounds and interests is complex. Failing to do so could lead to fragmentation, reduced collaboration, and slower progress in advancing Active Inference. | Community Growth, Public Engagement |
| Public Engagement and Ethical Considerations | Translating complex Active Inference concepts for broader public understanding while addressing ethical implications is challenging. Without this, we risk public misunderstanding, potential misuse of the framework, and missed opportunities for societal impact. | Public Engagement |
The Institute hosts and disseminates information using the shared space, YouTube, Discord, Github, Newsletter and other platforms as needed. This stack of platforms streamlines specific levels of access to shared resources, and enhances overall productivity within the organization. We aim to ensure that participants are aware of the platforms being used and understand their purposes and functionalities. We regularly evaluate, communicate, and reinforce best practices for information storage, access, and organization. We implement security measures, such as strong passwords, 2-factor authentication, and appropriate access permission in order to protect sensitive information. We back up important data regularly to prevent loss due to technical issues or accidental deletion. We conduct periodic reviews and audits of the information storage systems to identify areas for improvement and optimization. The specific use of each platform is described below.
YouTube is the primary platform for storing audiovisual content created for and by The Institute. Our designated YouTube Channel holds distinct playlists for courses, live streams, symposia, and other content that we host. We share and embed links within internal and external communication channels to provide easy access to relevant content. The content on YouTube is also backed up in a personal cloud storage service as well as in offline hard drives.
Discord is our primary platform for engaging with the Active Inference Ecosystem and broader community. We use Discord for real-time communication, informal discussions, and team collaboration. Dedicated channels are used within Discord to categorize discussions based on topics or projects. Participants are encouraged to share relevant files, documents, or links within Discord channels, fostering easy access to shared resources. We regularly monitor and moderate Discord channels to maintain professionalism, and eagerly look to improve our protocols and guidelines here and elsewhere.
Since the initial activities of the Institute (History of The Institute), we have written a monthly Newsletter.
See the archives
The Institute intends to evaluate quality, performance, and growth within community development at three scales, listed below, based on best practices within the Open Source community and adapted for our use-cases which include software, videos, and other products.
Evaluation at the level of individuals, with consideration for a plurality of individual priors (i.e., diversity in perspective, experience, culture, language, preferences, discipline, and level of expertise) and a focus on accessibility and onboarding. Objectives include quality of participant and user experience, plurality of educational mediums and formats (i.e., accessibility), networking and collaboration opportunities, and professional development. Pending grant or donor funding, The Institute will work with user experience, communications, and requirements engineering professionals to improve current and establish new feedback mechanisms and implement best practices for aforementioned evaluations. The following tools serve as a basis for evaluation:
Evaluation at the level of The Institute will consider various areas such as sustainability of personal and collective efforts, support reliability, and user experience quality, and Institute quality control and improvement. Objectives include increasing collaboration opportunities, ensuring consistency and rapid handling of inconsistency in documentation, and supporting and facilitating projects. Specific metrics of quality, performance, and growth at The Institute scale may include:
Evaluation at the level of the Ecosystem and community scale with consideration for impact and relationship management, and a focus on impact. Objectives include minimizing turnover rate in educational courses, increasing the number of participants, and maintaining and adding partnerships. Metrics of quality, performance, and growth at the community scale may include:
October
The History of The Institute begins in the co-founder team meeting in 2020 around a common interest in Active Inference. This resulted in productive collaboration and the publication “Active Inference & Behavior Engineering for Teams” in September 2020 (Vyatkin et al. 2020). The group was then known as “Team Comm”. Check out our first livestream, ActInf Livestream #001.1 ~ “Narrative as active inference", on July 28, 2020.
Following the 2020 publication, discussions turned towards exploring approaches that could catalyze the accessibility, rigor, and applicability of Active Inference, and how to merge the developing framework with the Systems Approach and Open Source. Out of these discussions an “Active Inference Lab” (or ActInfLab) was formed and began operations in 2021.
Over the first year of our operations, dozens of individuals from around the world engaged with ActInfLab through various projects such as educational Production, Open Source publishing, collaborative research projects, focused learning groups, The Active Inference Journal, and initial developments of the Active Inference Ontology.
Since the first quarter of operations in 2021, the ActInfLab hosted Quarterly Roundtable livestreams for communicating quarterly expectations and results to the community, a tradition that we continue to this day.
Beginning in 2022, a cohort-based Scientific Advisory Board (SAB) was established to connect the ActInfLab to cutting-edge theoretical work as well as various domain-specific applications. As interest in both the ActInfLab’s activities and Active Inference itself began to grow, ActInfLab soon emerged as a key facilitating organization in what was then a primarily academic community working on the underlying theory and potential implications for Active Inference.
The first Active Inference textbook comes out in 2022 (Parr, Pezzulo, Friston 2022), and the Institute begins hosting a Textbook Group (ongoing through 7 cohorts in 2024). The Textbook Group is an important ecosystem service, as there are few academic/institutional locations where learners can be supported through the curriculum of the textbook and beyond. Additionally, the Institute has curated and categorized learning materials that learners create while participating in the group, including questions and discourse.
The Institute begins the Internship program to scaffold and support the learning journey of learners. Interns come from different backgrounds — including high school, college, and graduate students on academic tracks, as well as professionals and others outside of academia. Interns, with their mentors, develop a personalized education and research curriculum which lasts months-years.
In mid-2022, ActInfLab made the developmental leap to become The Active Inference Institute, a non-profit organization registered in Delaware, USA with the intention of making its facilitatory role in the community impactful and sustainable. As part of the requirements for a non-profit, we also laid out the Institute Structure, comprised of the Organizational Units: Administrative, EduActive (Education), and ReInference (Research).
At the end of 2022, the Board of Directors has its first meeting. The Board continues to meet on a quarterly basis.
Institute Projects and Institute Programs continue, including the first two full course offerings: Physics course and Social Science course. These courses span months, and include office hours with the lecturer and teaching assistants.
In addition to continuing livestream Production on YouTube (GuestStream, ModelStream, PaperStream, etc), the Institute hosts the popular Active Inference Insights podcast.
During the year, we begin researching and applying for private and government Grants.
Organizationally, the Institute receives official recognition as a 501(c)(3) non-profit organization, supporting our Philanthropy efforts. We were able to achieve this milestone with the pro bono support of the Fried Frank law firm.
The largest cohort to date of the Scientific Advisory Board makes many diverse contributions across projects.
The Fellows program begins to highlight and scaffold the work of Ecosystem member. As of November 2024, there are 5 Research Fellows have joined. Fellows represent members of the Ecosystem who have contributed substantially to the ecosystem through publications and presentations.
To meet the needs of trainees and Interns for one-on-one guidance with projects, we introduced the Mentorship program. Members of the Scientific Advisory Board and select other individuals, volunteer to mentor and connect with individual trainees.
Following the Quantum Active Inference Prepare-Measure cycle described by Chris Fields in the 2023 Physics course, we implemented a “Prepare and Measure” system for Institute Projects and Ecosystem Projects. Prepare and Measure allows people to set goals and report back when they have reached them. Project ~ Measurement can be provided by anyone about different Domains of Application for Active Inference. In contrast Project ~ Preparation describes what someone is preparing to do, whether they are just letting us know,
These always-open reporting systems are used to gauge the ongoing projects and work done by community members, and provide visibility to these updates in the newsletter.
Work during this year remains all-volunteer. Philanthropy support begins to come in, supporting some operational software costs. We applied for several Grants (such as FarmWorks and related to AI safety with the RxInfer.jl Learning Group).
The Authors collaborated on this Institute & Ecosystem leading up to the 4th Applied Active Inference Symposium on November 13th, 2024.
See 2025 for all information from 2025!
By adhering to this code of conduct, members of our academic and research driven institution contribute to a culture of excellence, integrity, and collaboration, fostering a positive and inclusive environment for the pursuit of knowledge and innovation.
To support the accessibility, rigor, and applicability of Active Inference.
Act. Infer. Serve.
The formal mission statement of the Institute only scratches the surface of the goals and aspirations of its members and the many parties in its broad ecosystem.
This is screenshot/text from our Form 1023 (this is from the IRS 501(c)(3) status application), submitted in 2023.
The formal mission of the Institute, seen in the screenshot to the left, is:
> Active Inference Institute, Inc. (the Institute) is dedicated to developing, supporting, and promoting open science and integrative frameworks such as active inference. In furtherance of its mission, the Institute will conduct the following activities: (1) education, (2) research, (3) grantmaking, and (4) administration.
The Active Inference Institute serves as a scaffold for stabilizing and connecting myriad fields around a central tradition and approach of Active Inference.
The Institute aims to make the Active Inference framework and the Ecosystem we serve more accessible, applicable, rigorous, and integrated.
We facilitate educational, theoretical, and applied engagement with Active Inference, promoting awareness of the field within the lay, academic, public-sector, and professional communities.
We envision a future in which the term “Active Inference” is used as widely as “Machine Learning”, as a result of its demonstrated utility and impact in a variety of Domains of Application.
We are committed to fostering a culture of excellence, collaboration, and innovation. Our values and principles serve as the guiding principles that shape our Strategy and define our organization's character.
Active Inference Institute (AII) is on the path of open-endedness.
Our Strategy considers learning and applying Active Inference for changes in the niche over multiple nested scales. Through time we increase the degree of hierarchical organizational complexity to overcome competing interactions and frustrated states.
We engage in policy selection across multiple scales, reducing our uncertainty about realizing our expectations and preferences. We learn, finding epistemic value along the way, while pragmatically ensuring Institute persistence and development.
The three scales that Active Inference Institute modifies and interact with:
关键表面
Began as ActInfLab在2021,成为德克萨斯州非营利组织在2022,并于2024获得501(c)(3)认可。
所有背景、时区和熟悉程度的人都在研究所的项目和活动中受到欢迎。
本研究所将 Active Inference 理论与应用项目、研讨会、工具和生态系统支持相结合。
指导每个项目、计划和社区倡议的口号。
相关资源
Audience: Newcomer
Primary public community channel for conversation, coordination, and newcomer orientation.
Audience: Newcomer
Recurring public update surface for activity orientation and current work.
Audience: Researcher
Public ecosystem shortlink for Institute context, projects, activities, and conceptual maps.
Audience: Newcomer
Public videos and podcasts shortlink for browsing recordings by format and topic.
Audience: Developer
Public GitHub organization for Institute repositories and open-source work.
Audience: Researcher
Research and publication-oriented repository for the Active Inference Journal.
Audience: Newcomer
Public short-form update channel.
Audience: Newcomer
Public social profile for community updates.
Audience: Newcomer
Public social page for Institute visibility.
官方页面
Audience: Newcomer
Current public Institute landing page with mission, vision, nonprofit status, community metrics, and Get Started pathway.
Audience: Newcomer
Official public homepage for the Active Inference Institute.
Audience: Newcomer
Official history page for the Institute.
Audience: Newcomer
Public newsletter destination.
Audience: Learner
Public Textbook Group shortlink for cohort and learning-group coordination.
Audience: Partner
Public symposium shortlink for Applied Active Inference Symposium information.
Audience: Researcher
Public Active Inference Ontology shortlink for shared conceptual infrastructure.
Audience: Newcomer
Public Strategy shortlink for institutional orientation and planning context.
仓库
Audience: Newcomer
Active Inference Institute organizational website.
Audience: Newcomer
Public institute_website repository in the ActiveInferenceInstitute GitHub namespace.