Overview

Active Blockference is an open source package (https://github.com/ActiveInferenceInstitute/ActiveBlockference) that is implementing Active Inference simulations in cadCAD (https://cadcad.org/), with a focus on blockchain and Decentralized Science (DeSci) ecosystems. We are working to develop Active Blockference into a modern framework for cognitive systems engineering in blockchain & cyberphysical ecosystems.

The Active Inference Institute (AII, https://www.activeinference.org/) is a non-profit open-science institute curating and developing applications related to the Active Inference framework. The goal of AII is to produce cutting-edge education and research to enable real-world applications of Active Inference. AII also seeks to scaffold the Active Inference community: increasing the competency of participants & raising broader awareness of these topics.

As the DeSci ecosystem rapidly expands, it will become increasingly important have historical, real-time, and anticipatory awareness for researchers and organizations. This kind of situational awareness will be vital for maintaining public goods such as Science, Education, and social Communities. In this grant we are aiming to operationalize our Active Entity Ontology for Science (AEOS, https://coda.io/@active-inference-institute/active-entity-ontology-for-science-aeos), through the use of the Active Blockference open source package, to enable this kind of modeling.

Active Inference is an approach to modeling cognitive entities which is being applied to many fields ranging from artificial intelligence to communication, trust, and collective intelligence, and regenerative finance. We are building Active Blockference to be a simulation and design package that uses cadCAD to implement Active Inference (ActInf). Briefly summarizing, ActInf agents act to minimize their uncertainty about the world though sensemaking and action, reducing the divergence between their predictions and observations.

Resources to learn more: ~300 Livestreams covering many Active Inference papers and topics https://coda.io/@active-inference-institute/livestreams Active Blockference presenting on the cadCAD community call from August 3, 2022 https://youtu.be/aRPhKnxXLtQ?t=493 Active Blockference presenting on the SCRF community call from August 25, 2022 https://youtu.be/7Pr1CpfhmAk?t=375

Where is the project and team currently at? April 2023 Updates -- We applied for a grant that reflects our current directions: "Systems Modeling and Cognitive Audits for Hypercert Ecosystems" https://zenodo.org/record/7626769 -- We wrote a new paper and framework "Generalized Notation Notation (GNN)" that should augment the ability for Active Blockference models to be implemented https://coda.io/@active-inference-institute/generalized-notation-notation -- Currently, several working Active Blockference simulation notebooks are available, and our all-volunteer project team meets twice weekly.

We have been in development on this project specifically for around 1.5 years, and the Institute has been in operation since the beginning of 2021. We are seeking resources to increase the pace of development, scaffold partnerships, and acquire computational resources.

This project is scaffolded and stewarded by the Active Inference Institute (https://www.activeinference.org/). We are an international team of distributed researchers. We are working closely with the cadCAD development team, and using their platform BlockScienceLabs. Additionally we are connected/collaborating with several relevant groups in Web3 (such as Kernel Community), the DeSci space such as OpSci. If you want to collaborate or get involved, please let us know!

We have been interested to apply Active Inference to the setting of Web3, DeFi, and Decentralized Science (DeSci) for some time now, resulting in our paper from March 2022: “An Active Inference Ontology for Decentralized Science: from Situated Sensemaking to the Epistemic Commons“ https://zenodo.org/record/6872311 . In this paper we provided an overview and contextualization of Decentralized Science, and introduced the AEOS framework which this grant will develop upon.

Currently, Active Blockference simulates the cognitive processes and goal-directed behavior of a single-agent. Our goal is to expand this simulation into a multi-agent model, in order to explore the cognition and behavior that drive successful DeSci and DeFi projects. As a multi-agent simulation, Active Blockference would facilitate the rigorous analysis of consensus protocols and Web3 communities prior to deployment. It can serve as a sandbox where we explore the cognitive, micro-economic, behavioral, and decision-making processes of DeFi. Hence, we aim to develop the tooling, use cases (case studies), and educational resources that enable a cognitive audit of DeFi protocols in Web3. The addition of this cognitive layer into the DeFi development stack will facilitate trust in DeFi protocols within the Ethereum community, thus maximizing the value of these projects.

What are the next steps for Active Blockference?

1. Specify the entity types that are involved in DeSci using AEOS
2. Create Active Blockference entity models in cadCAD
3. Compose simulation environments based upon specific use cases
4. Do parameter sweeps in cadCAD across system and cognitive parameters
5. ???????????????????????????
6. Co-create an epistemic commons that works for everybody

This grant we will have 3 main aims and areas of output.

1. Develop the Active Blockference package to address the blockchain setting with a focus on the social and technical aspects of DeSci ecosystems.
2. Use Active Blockference to understand behavior within ETH/Web3 ecosystems, including DeSci and DeFi. Here we will complement existing cadCAD simulations and case studies
3. Improve educational materials, documentation, and public communications (e.g. blog posts) to onboard new people to the Active Blockference community. This will included Low/No Code application running Active Blockference

What would success look like for Active Blockference?

Providing a library/package anyone can build on top of with software affordances (like Git), to enable cognitive modeling & cognitive auditing of epistemic ecosystems in DeSci and beyond.

How does this project benefit the greater Ethereum ecosystem?

The ETH ecosystem is challenged by projects that fail to realize their expectations, due to insufficient consideration of the cognition & behavior of people using them. This can look like the failure of communities/protocols to persist for various reasons, here we are most interested in the behavioral aspects of these phenomena. The Active Blockference package and framework will facilitate rigorous cognitive audits of Economic protocols, epistemic ecosystems (as in DeSci), as well as Web3 communities, during design time and run-time (realtime tools for adaptive organizations). Implementing Active Blockference within the ETH ecosystem will add value, and scaffold community cohesion in several ways:

1. We have advanced structural and macroeconomic models for Ethereum in cadCAD. However, there are no entity models for ETH, limiting the usage of behavioral and microeconomic models. These types of entity models could complement structural models, for example by enabling better understanding of the behavioral aspects of cryptoeconomic systems, fostering healthy interactions, planning with foresight, and applying of corrective interventions
2. While ETH mainnet transactions are visible for all, explainability is not. Active Blockference will contribute towards explainability via interpretable visualizations, natural language descriptions, and useful simulations.
3. ETH elaborations: ETH infrastructure is undergoing rapid development and deployment, Layer 2’s, ZK rollups, and other innovations position the ecosystem in a constant state of change that can benefit from the ActInf approach presents inherent principles and sensitivities working with cognitive agents acting under uncertainty.
4. Cognitive Security in ETH: Cognitive auditing — Beyond today’s smart contract auditing (inferring situated, motivated outcomes).
5. The implementation of an Active Entity Ontology in cadCAD will provide the ETH community with a working model of active and informational agents as well as their affordances (e.g. a smart contract is an informational agent with predefined affordances).
6. The accessibility of on- and off-chain systems will be facilitated via open source Active Blockference models of cognitive-behavioral systems, permitting validation of ETH protocols using no/low-code simulations.
7. An open source software package that connects ETH on- and off-chain activity and packages already in use (such as cadCAD) to broader currents related to Cognitive Science. This connection could enable algorithmic project “cognitive audits” to maximize the value of projects deployed on Ethereum. For example currently, smart contracts are statically audited for security (e.g. unintended interactions within/among contracts). With Active Blockference it would be possible simulate a testnet blockchain environment, where smart contracts usage is evaluated as interacting with dynamic cognitive agents. This could result in fewer projects that are viable from a “smart contract security” perspective but unviable from a behavioral perspective.

What is new in your approach and why do you think it will be successful?

We are un-boxing DeSci collaborations through ontology-driven applications of cognitive science and organizational design. Researchers would know what ropes to climb, in order to contribute and succeed in terms of implicit goals.

Active Inference centers the role of intrinsic motivation (motivators such as curiosity and learning, as opposed to extrinisic motivators such as tokens), a crucial factor for DeSci.

Our work will be a solution which can templatized and inherited upon to adapt to different team settings. We are building tools for many kinds of organizations in the ETH ecosystem, not only scaffolding our own work.

Using Active Blockference, we are developing a new open source approach for modeling, evaluating, and designing DeFi / DeSci / Web3 community protocols. The framework of Active Inference will facilitate the cognitive modeling of Web3 system dynamics (individual and collective behavior, funding, treasuries, protocols, etc.) as well as the general Ethereum consensus mechanism.

THANK YOU FOR THE CONSIDERATION!