The most lauded current feature of AI agents is their ability to learn in real-time from the input of the community. This is facilitated by the above information flow.
1. CREATORCreator deploys an agent and provides documentation (.pdf, .md, .txt) and a system prompt (a core instruction given to an agent that it must follow above all new instructions).
2. BRAINThe agent's "brain" consists of the creator's provisions, internet search capabilities, the underlying model's embedded knowledge and lastly, the "context store".
3. CONTEXT STOREInsights from interactions with the community are stored in the "context store". This allows the agent to "learn" from the input of the community.
4. COMMUNITYThe community queries the agent. The agent utilises its "brain" to generate responses.
5. DECISION MAKINGThe agent can then potentially make decisions based on the will of the creator and the community, such as interacting with DeFi protocols with its own wallet.
Despite all the wonderful innovations in the agentic arena, we are still operating with the "singular agent, siloed knowledge" model. This means that all the brilliant work done by the community to cultivate a strong agent intelligence will remain in the confines of this one agent.
This is quite like science in general; many brilliant scientists and studies that fail to reach the optimal wider audience, reducing collaboration and innovation.
Knowledge sharing is key to a successful and distributed future of research; agents are no different. Now, let's see how Proof of Knowledge can improve this paradigm.
How agents learn with Proof of Knowledge:
Proof of Knowledge (PoK) is a protocol that facilitates and tracks the retrieval by AI of knowledge units (kEngram) from a global, shared context store. kEngrams form a hive-mind intelligence between AI agents connected to the PoK protocol. Combined with onchain incentives, governance and reputation mechanisms, PoK generates a knowledge base capable of powering the future of RAG LLM applications onchain.