The question of why LLMs will never achieve AGI is rapidly moving from a niche academic debate to a central topic in the tech world. While Large Language Models are incredibly powerful, their foundational design has inherent limitations. This week, the Financial Times amplified this critical conversation, featuring VERSES Chief Scientist, Professor Karl Friston, in an article that pinpoints exactly why the current path is hitting a wall and what the real solution looks like.
While LLMs are powerful tools for pattern matching and text generation, they are ultimately sophisticated mimics, not true agents. They lack the one crucial ingredient for genuine intelligence: agency.
The Key Reason Why LLMs Will Never Achieve AGI: A Lack of Agency
As Professor Karl Friston explained to the Financial Times, today’s generative models will never achieve AGI because they “have no agency under the hood.” This is the core of the problem and the simplest answer to why LLMs will never achieve AGI. An LLM has no goals, no beliefs, and no internal model of the world it can act upon. It is a passive system that predicts the next word in a sequence based on statistical patterns from its training data.
This is why, as the FT article notes, these models break down in novel situations. They can analyze historical traffic patterns, but they can’t intelligently predict and adapt when a completely new factor—like a Taylor Swift concert—is introduced. They are brittle because their “knowledge” is not grounded in a genuine understanding of cause and effect.
Active Inference: The Biological Blueprint for True AGI
So, if LLMs aren’t the answer, what is? The solution lies in the principles that govern all biological intelligence, from a single cell to a human brain. This framework is called Active Inference.
As Professor Friston states, true intelligence must be “active and embodied and situated physically in a world that you can act upon.” An Active Inference agent isn’t just processing data; it is driven by a fundamental imperative to understand its environment. It builds a model of its world, makes predictions about it, and then actively takes actions to test those predictions and update its model. It learns by doing. It has agency.
From Theory to a Real-World AGI
This isn’t just a theoretical distinction; it has profound practical implications. As VERSES CEO, Gabriel René, told the Financial Times, “AI breaks when it hits the real world because the real world keeps changing.” The future of AI isn’t a single, massive model in a data center. It’s a network of distributed, adaptive agents that can learn, specialize, and cooperate, forming a true collective intelligence.
This is the path to AGI that VERSES is building. The vision is to create truly intelligent, adaptive agents using the principles of Active Inference. These are the systems that will power the next generation of technology, from autonomous logistics and smart cities to the Spatial Web. And we are architecting them to be deployed on decentralized computational networks like HyperCycle, creating a resilient and intelligent digital ecosystem.
Intelligence is not about compressing historical knowledge. As Gabriel René says, “Intelligence is about adaptation.” By focusing on Active Inference, we are moving beyond the limitations of LLMs to build the first true AGI.
The conversation about the future of intelligence is evolving rapidly. Sign up for our newsletter to stay ahead of the curve and understand the shift from generative models to truly adaptive, Active Inference AI.
