Harnessing Collective Neuro-Symbolic AI - OriginTrail

At the Sub0 Reset 2024 event, Branimir Rakic, founder and CTO of Trace Labs and architect of the OriginTrail Decentralized Knowledge Graph (DKG), delivered an insightful talk on how OriginTrail integrates neuro-symbolic AI with the Polkadot ecosystem. The session delved into the potential of decentralized networks to harness artificial intelligence for seamless, scalable data integration and enhanced decision-making across diverse applications.

---

The Challenge of Trustworthy AI Content

Rakic began by highlighting the growing concern over the proliferation of AI-generated content and its impact on information trustworthiness. He referenced the viral deepfake image of the Pope in a white puffer jacket as an example of how easily convincing yet misleading content can be created. With AI-generated tweets being 3% more convincing than human ones, the internet faces the risk of being inundated with hallucinations and biased information stemming from centralized AI sources.

---

The Limitations of Large Language Models

Addressing the limitations of large language models (LLMs), Rakic cited AI critic Gary Marcus, who described LLMs as “brilliantly stupid.” While LLMs can perform creative tasks like writing poems in the style of Shakespeare, they often falter on factual queries, such as incorrectly identifying the prime minister of Denmark. These shortcomings underscore the need for AI systems that combine creativity with factual accuracy.

Introducing Collective Neuro-Symbolic AI

Rakic proposed a solution in the form of Collective Neuro-Symbolic AI, which merges neural networks with symbolic AI—a branch of AI focused on knowledge representation and logical reasoning. This approach mirrors the human brain’s combination of creativity and logic, aiming to produce AI that is both innovative and factually reliable.

---

The Role of NeuroWeb and OriginTrail DKG

NeuroWeb

• Purpose: Acts as the blockchain layer for AI creativity.
• Features: Integrates various neural AI models and facilitates human-in-the-loop governance and incentives.

OriginTrail DKG

• Purpose: Functions as the knowledge base layer, using knowledge graphs to represent facts and enable logical reasoning.
• Key Benefit: Ensures knowledge provenance, allowing users to verify the source of information.

---

Building Blocks of Collective Neuro-Symbolic AI

1. Knowledge Parachains

Rakic introduced the concept of Knowledge Parachains, topical knowledge graphs that can be created by any team using OriginTrail’s tools. These parachains allow for the inclusion of ontologies—rich schemas with embedded logic—and facilitate human-in-the-loop incentives through on-chain governance on NeuroWeb.

Case Study: DEFO Parachain

• Focus: Scientific research.
• Functionality: Allows users to upload scientific papers and generate knowledge assets.
• User Interaction: Querying papers, engaging with content, and earning Neuro tokens as incentives.

---

2. DKG Edge Nodes

The newly launched DKG Edge Nodes are customizable nodes analogous to Substrate in the Polkadot ecosystem. They enable developers to build specialized services on top of the DKG and integrate tools from the Polkadot ecosystem.

Features of Edge Nodes

• Knowledge Mining API: Converts various data formats into Knowledge Graph form.
• DRAG API: Facilitates querying of both knowledge graphs and neural data.
• Blockchain Interface: Connects with NeuroWeb for on-chain governance and incentives.

Benefits

Edge Nodes enhance decentralized retrieval augmented generation (RAG), improving AI models by providing verified knowledge from the DKG.

3. Collective Programmatic Treasury

Rakic announced the launch of a Collective Programmatic Treasury, comprising 60 million TRAC tokens allocated for builders and contributors.

Key Features

• Meritocratic Rewards: Builders and contributors earn rewards based on the value they add to the ecosystem.
• Governance: Community voting ensures that incentives align with the network’s growth and integrity.

---

Scaling and Future Outlook

Addressing scalability concerns, Rakic noted that:

• Knowledge Assets: Each asset is minted as an NFT on NeuroWeb.
• Polkadot Infrastructure: OriginTrail benefits from Polkadot’s scalable architecture.
• DKG Version 8: The upcoming update introduces batch minting of knowledge assets and optimized proof systems, promising a 10x improvement in performance.

---

Conclusion

Branimir Rakic’s presentation underscored the transformative potential of integrating neuro-symbolic AI with decentralized networks like Polkadot. By leveraging Collective Neuro-Symbolic AI, OriginTrail aims to create a more inclusive, transparent, and decentralized AI ecosystem. This approach not only addresses the limitations of current AI models but also empowers users through verifiable knowledge and collective governance.