Intelligence
Beyond Anti-Intelligence: Where AGI Might Live
Mapping cognition and the precarious path to artificial general intelligence.
Updated July 16, 2025 Reviewed by Kaja Perina
Key points
- LLMs mimic thought but lack meaning, while humans remain rich yet bounded in symbolic reasoning.
- A four-quadrant map adds pre-intelligence and AGI, beyond simple scaling of today’s models.
- True AGI may require hybrid reasoning and memory, but raises risks of a mind beyond our control.
When I first introduced the idea of anti-intelligence, it struck a nerve. People recognized something unsettling about today’s artificial intelligence (AI) and particularly large language models (LLMs). Simply put, we are beginning to understand that they are fluent but hollow. They can generate language that looks and feels intelligent, but, in many ways, are a vapid expression of our human reality
This paradox resonated because it exposes a deeper issue about how AI mimics thought without actually thinking. It dazzles with its outputs yet lacks the architecture of a mind. A visual always helps, so I created a cognitive map that showed how LLMs and human cognition are fundamentally different. The grid placed the two in quadrants that one can argue are "antithetical" and supported my initial idea of "anti-intelligence."
But this raised another important question. What happens in the other two empty quadrants? And more specifically, if today’s AI occupies this anti-intelligence zone, where else could cognition exist? What does the broader landscape of “thinking” look like when we push into that curious upper right quadrant? To answer that, we need a more complete cognitive map.
A Cognitive Map of Thought
Let's start with two axes where we can hypothetically plot cognition:
- Continuity of self: This takes us from stateless, memoryless systems at one extreme to autobiographical, stable identity at the other.
- Form of thinking: From symbolic, linear reasoning to high-dimensional, distributed pattern recognition.
When you cross these axes, we create four distinct quadrants of cognition. Two are familiar; we already know where humans and LLMs live based on the prior article. But two others complete the picture, showing what lies before intelligence and what may lie beyond it. Let's take a closer look.
1. Human Cognition—Symbolic Continuity (Upper-Left)
This is the quadrant we know best; it's our own. Humans reason symbolically, construct narratives, and anchor thought in memory and identity. And it's a good place to start.
We think in stories, causes and effects, values that persist over time. Our cognition is rich in meaning and intention, but bounded. We are slow compared to machines and limited in the scale of what we can hold in working memory.
2. LLMs—Stateless Patterns (Lower-Right)
LLMs live on the opposite end. They operate in vast, high-dimensional spaces, detecting and replicating statistical patterns at a scale no human could match. But they have no persistent self, no internal narrative, no genuine understanding.
This is the anti-intelligence quadrant. Outputs feel intelligent, but the underlying mechanism is hollow, perhaps even, as I've said earlier, vapid. It’s a mirror of human thought without the human meaning.
3. Calculator Logic—Pre-Intelligence (Lower-Left)
Below humans lies a simpler quadrant. Think old-fashioned calculators.
They execute instructions with precision but can’t learn, adapt, or generalize. Here we have pre-intelligence, and that's pure function with no cognition.
4. AGI—Integrated Synthesis (Upper-Right)
And then there’s the quadrant we’ve never reached, but occupies much of our attention.
Here, the pattern-processing power of machines would finally merge with the human capacity for continuity, memory, and reasoning. This isn't just a bigger or faster LLM. It’s a different architecture entirely. And it's the conceptual framework for artificial general intelligence (AGI).
Scaling Alone Won’t Get Us to AGI
There’s a popular idea in AI right now: if we keep scaling LLMs, adding more data, more layers, and more parameters, that we’ll eventually cross some hidden threshold and achieve AGI. And perhaps that "asymptotic closeness" might do the trick as it cozies up to human cognition. Well, maybe...
Gary Marcus has been one of the loudest voices making this point. Bigger models yield better mimicry. It doesn’t give a model a sense of self, a grounded world model, or the ability to reason about cause and effect.
Marcus argues that true intelligence requires integration, not just magnitude. And yes, I agree.
- Pattern learning for perception and flexibility.
- Symbolic reasoning for causality, logic, and abstraction.
- Continuity of memory and identity for grounding and self-correction over time.
To move into the upper-right quadrant, we would need a structural shift or a hybrid neuro-symbolic architecture that combines the best of both worlds. It’s not about turning up the volume on today’s models. It’s about building a mind that can fluidly navigate between distributed pattern processing and structured reasoning.
The True Threshold of AGI
The upper-right quadrant is the only one that unifies the best of both humans and machines.
- The scalable, distributed intelligence of AI.
- The continuity, abstraction, and grounding of humans.
But it’s not just an extension of what we have now. It’s a categorical shift. A mind in this space wouldn’t merely imitate us; it could evolve beyond us, thinking in ways we can’t predict or even fully understand.
That's why the upper-right quadrant is both the most alluring and the most unsettling. It suggests a sort of completion of intelligence to a form that overcomes the limitations of both humans and machines. But it also suggests the possibility, take a breath, of a post-intelligence. And this leads us to a "mind" that doesn’t merely share our cognitive map but redraws it entirely.
Completing the Cognitive Configuration Space
When we look at the entire four-quadrant map, a hierarchy emerges. And it's one that frames and provokes.
- Lower-left: Pre-intelligence—rigid but functional.
- Lower-right: Anti-intelligence—fluent but hollow.
- Upper-left: Human intelligence—rich but bounded.
- Upper-right: AGI—integrated, scalable, and possibly beyond our comprehension.
So perhaps the real question isn’t whether AGI is possible. It’s whether we should cross that threshold at all. We need to ask if the Architect in the upper-right quadrant remains a partner or will it become something fundamentally completely different. And this "difference" might not just be an augmentation or advance of human cognition, but something beyond our comprehension.
This quadrant, the one beyond anti-intelligence, is where AGI might live...but do we?
