Точка Синхронізації

Researchers specializing in artificial intelligence published a paper on arXiv on July 24, 2024, detailing how a functional AI architecture called 'Agentic Flow' accidentally mirrored four fundamental theories of human cognition during its development in the United States. The team discovered that their practical attempts to solve the operational limitations of Large Language Models (LLMs) led to a structural convergence with the psychological and neurological frameworks of Daniel Kahneman, Karl Friston, Marvin Minsky, and Andy Clark. This unintentional alignment suggests that the path toward more efficient and reliable AI may naturally follow the established blueprints of biological and cognitive intelligence. The 'Agentic Flow' architecture is composed of five specialized, interlocking modules: Retrieval, Cognition, Control, Action, and Memory. By organizing these systems into a structured loop, the developers sought to overcome the hallucinations and reasoning gaps common in standard LLMs. The study highlights how the 'Cognition' and 'Control' modules mirror Kahneman’s Dual-System theory, which distinguishes between fast, intuitive thinking and slow, deliberate reasoning. Simultaneously, the system’s feedback loops simulate Friston’s predictive processing, where the brain—or in this case, the agent—constantly updates its internal models based on environmental sensory input. Beyond internal processing, the architecture also reflects Minsky’s 'Society of Mind' by treating intelligence as a collection of smaller, specialized sub-agents working in concert. Finally, the inclusion of robust Retrieval and Memory modules aligns with Clark’s 'Extended Mind' theory, which posits that cognition is not confined to the individual but is distributed across external tools and environments. This convergence indicates that as AI engineering becomes more sophisticated, it is fundamentally validating long-standing cognitive science theories through practical implementation, potentially bridging the gap between artificial and human-like reasoning.