EGO: A self-referential AI architecture that mimics biological cognition for AGI

A new paper on arXiv introduces EGO (Environment Generative Operator), a cognitive architecture that rethinks AGI not as a larger language model but as a self-referential, autopoietic system inspired by biological cognition. Authors Alberto Mangiante, Paolo Totaro, and Domenico Ninno argue that current Large Language Models are structurally incapable of true general intelligence because they lack the ability to maintain internal organization and self-reference. EGO instead relies on a formal "E-language" that allows the system to recursively define and modify its own structure, similar to how a living cell maintains itself through autopoiesis—a concept from biologists Humberto Maturana and Francisco Varela.

The architecture is designed to be autonomous and embodied, meaning it interacts with its environment and adapts without external reprogramming. By grounding intelligence in self-organization and recursive self-reference, EGO could overcome key limitations of LLMs, such as their inability to truly understand context or exhibit persistent agency. The paper, updated in July 2026, positions EGO as a bridge between artificial intelligence and biological theories of cognition. While still theoretical, it offers a radical departure from scaling-based approaches, suggesting a path toward AGI rooted in principles of living systems rather than brute-force computation.