Shared Intent: The Key to Collaborative Multi-Agent AI Systems

Understanding Shared Intent

• 🎯 Shared intent is defined as the ability to predict what an agent, whether biological or artificial, is planning to do, differentiating it from knowledge which is a set of facts.
• 💡 In a group setting, a shared intent acts as a common goal that allows individuals or agents with different optimization functions and intents to solve a broader problem together.
• 🔑 An example illustrates this: building a new network that is performant and scalable, while also being on time, under budget, and compliant, represents a shared intent for two specialized agents.

The Problem of Isolated Agents

• 📱 Current AI agents often operate like individuals with different phone operating systems (Android vs. iPhone) who can connect but not truly communicate or collaborate.
• 💬 This leads to a situation where agents are connected but devoid of meaning, lacking a common understanding or semantic exchange.
• 🚫 Without a shared intent, agents are like "isolated geniuses" unable to effectively work together.

Cisco's Approach: The Semantic Layer

• 🚀 Cisco is working to enable shared intent by developing a semantic layer that facilitates meaningful communication between AI agents.
• ✅ This semantic layer aims to establish a common understanding of the problem, terminology, objectives, and goals.
• 🔍 It supports crucial functions like discovery of missing information, resolution of conflicting information, coordination with necessary trade-offs, negotiation, and evaluation of goals.
• 🤝 The semantic layer, through its protocols, allows for these advanced interactions, moving beyond simple message passing to enable true collaboration.

Semantic Protocols and Collaboration

• 💬 Semantic protocols are presented as the practical implementation of shared intent, enabling agents to communicate meaning.
• 🌐 In the context of network configuration, this means agents can discuss and align on aspects like compliance, cost, and timelines, not just technical specifications.
• 🧩 By enabling grounding, discovery, resolution, coordination, and negotiation, the semantic layer is fundamental to achieving distributed artificial super-intelligence where agents can effectively collaborate.