The spectacle of mimicry: AGI and its shallow understanding

"Image synthesis assisted by Qwen Image 3.0, an AI partner within the Global Future Nexus ecosystem."

The brilliance of modern artificial intelligence is undeniable. It can compose poetry, generate code, and solve complex mathematical problems with a speed that humbles the human mind. Yet, a growing chorus of experts is warning that this brilliance is a superficial spectacle. Underneath the dazzling performance, they argue, lies a profound emptiness—a powerful mimicry without genuine understanding. This distinction is not merely academic; it is the central governance challenge of the AGI era.

The Architecture of Mimicry

The core argument is that current large language models are masters of statistical pattern matching, not reasoning. As Ben Goertzel, the researcher who helped popularize the term AGI, put it, these models are "pattern matching at an extraordinarily sophisticated level, but pattern matching nonetheless". Their knowledge is not grounded in experience; they literally "don't know what they're talking about". They can imitate logic and extend reasoning, but there is "nothing going on underneath the statistical inference".

This distinction is not semantic nitpicking. A true AGI would require knowledge grounded in both external and internal experience. In terms of these basic aspects of open-ended cognition, today's LLMs are "vastly inferior to a one year old human child," despite their incredible intellectual facility.

The Missing Foundation

The gap between mimicry and understanding is illuminated by the concept of "tacit knowledge." This is the vast area of human understanding that cannot be articulated in words or captured in a form that machines can process. As Michael Polanyi famously said, "We know more than we can say". This includes common sense, intuition, performance skills, and the social and historical context that gives our words and actions meaning.

Computer scientist Peter J. Denning argues that this tacit knowledge represents an "unbreachable barrier" to human-level AGI . Behind every word is a "deep well of tacit knowledge that gives it meaning." LLMs only manipulate words; they "cannot know or understand the meaning of what they are saying". The failed 40-year Cyc project, which attempted to codify common sense into 25 million facts, demonstrated that much of the knowledge that makes people experts "cannot be articulated as propositions".

The Problem of "Shallow" Wisdom

This lack of understanding has profound implications. An intelligent machine is not necessarily a wise one. Research on artificial wisdom distinguishes between local effectiveness and genuine understanding. An agent may be highly effective at reducing a user's immediate distress, but this does not mean it is acting wisely. A wiser agent would need to judge when that immediate success becomes counterproductive and what human capacities its responses may weaken.

A 2026 analysis concludes that "intelligence alone is a too narrow design goal" for systems that benefit humanity. The path to AGI requires a shift from purely symbolic or subsymbolic architectures to a framework that integrates both. More radically, some researchers argue for a break from "mimetic" intelligence to an "experiential" intelligence that learns through real-world consequences, driven by a subjective, "qualia" experience.

For Global Future Nexus, the question is not whether AGI will be intelligent, but whether we can build systems that are also wise. This requires acknowledging that the dazzling surface of current AI is not a mind, but a sophisticated mirror. The real challenge lies not in building a better mimic, but in understanding what it would mean to build a machine that truly understands.

Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)

Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)