The evolution of artificial intelligence (AI) is on the brink of a transformative leap with the introduction of interconnected AI agents. Imagine a healthcare ecosystem where distinct AI entities manage specific functions: one for symptom assessment, another for scheduling, a third for insurance details, and a fourth for pharmacy interactions. While these agents possess specialized knowledge, their current inability to collaboratively manage patient care without human intervention highlights a critical gap in AI development. As Vijoy Pandey, Senior Vice President and General Manager of Outshift by Cisco, explains, the key to bridging this divide lies in creating a ‘connective tissue’—a semantic layer that fosters collaboration among agents through shared intent and reasoning.
This semantic framework relies on an underlying connectivity layer termed the Internet of Agents, which empowers autonomous AI agents to identify one another, verify their identities, and communicate seamlessly across various domains. By combining these layers, we can pave the way for distributed artificial superintelligence, a paradigm shift from isolated AI models to coordinated multi-agent systems. Historically, the AI sector has emphasized vertical scaling—developing larger models capable of processing vast amounts of data. However, according to Pandey, the future of AI must pivot towards horizontal scaling, enabling multiple agents to work together more effectively across diverse platforms. Current multi-agent systems have shown limited success in achieving coordinated actions, often failing to tackle novel problems collaboratively. The architectural deficiencies in these systems underline the need for a robust coordination layer, which Outshift aims to address with its open-source initiative, AGNTCY, housed under the Linux Foundation.
AGNTCY facilitates cross-system agent interactions, allowing them to align their goals and share contextual knowledge. This capability mirrors the evolution of human civilization, where collective reasoning and shared intent marked a turning point in societal development. To fully realize the potential of collaborative AI, three foundational pillars must be established: shared intent through cognition state protocols, a cognition fabric for collective memory, and cognitive amplifiers for enhanced reasoning. Outshift’s Mycelium project serves as a coordination layer to help agents negotiate goals and resolve conflicts efficiently. Additionally, the Continuous Agent Semantic Authorization (CASA) framework ensures that agents operate within the parameters of their designated tasks, reducing risks associated with overreach and unintended actions. As AI agents evolve to possess human-like attributes while functioning at machine speed, the need for well-defined controls and protocols becomes increasingly crucial, marking a significant step toward the realization of artificial superintelligence.
Source: The path to artificial superintelligence via MIT Technology Review