Abstract
In this paper, we envision a novel multidisciplinary approach to macro AI infrastructure by proposing an AI-to-AI Language, Multi User-Agent Network, Cognitive Functions of Thought (CFoT), and Unsupervised Learning with Cognitive Feedback. By equipping Language Models(LMs) with a purpose built AI-to-AI language, we empower AI agents to ‘Talk’ in a syntax that is token efficient and optimizes task-related communication and function calling. Building upon single-party multi-agent frameworks like Autogen, we propose a Multi User-Agent Network that 'Links' agents utilizing decentralized local open-source AI (OSS AI) models for distributed inference, enhancing efficiency, scalability, and trust. We propose advancing AI’s ability to ‘think’ by evolving Chain of Thoughts (CoT) and Tree of Thoughts (ToT) into Cognitive Functions of Thought (CFoT). The proposed LLM model improves upon Mistral’s Mixture of Experts(MoE) into a mixture of eight Jungian Cognitive Functions (MoCF). By employing these cognitive functions, we can apply inverse functions to enable unsupervised learning during idle computational periods, akin to the unconscious ‘Dream’ state of humans. The envisioned outcome is a collaborative AI network contributing to a collective knowledge pool, significantly advancing towards Artificial Collective Intelligence (ACI). This progression promises a new model for a safe, equitable, efficient, and aligned coexistence between humans and AI.