AI Doesn't Fit in a Box, and Neither Should Brains

Photo by Mr Xerty on Unsplash...

The other day I wrote a social media post explaining the differences between various terms people throw around about AI and brains.

I don’t like to do this often because I find the conversation dysfunctional, as this article explains, but many genuinely curious people get sidetracked by these philosophical discussions. The result is often a discounting of AI abilities, which I think is unsafe on several levels. We at least have to take seriously what we’re dealing with.

What really bothered people most was calling AI “highly intelligent.” I got the usual tirade about how what I don’t understand is how AI companies do <x> and how they are manipulating language. They try to say I am a hypester. They point out something LLMs and their extensions can’t do—a list that keeps shrinking. Fermat’s Last Theorem was thrown at me as one example. Seriously? Are we now comparing AI to the brightest in human history? I’m glad they’re not rating my intelligence. Though I suspect the bar would shift if another human were the comparison.

I address the topic, painfully, because it gets at an overarching theme in all these discussions. A cognitive approach that dominates modern society, amplified by traditional education, and highly damaging to both.

Categorical thinking.

Concepts Don’t Have Boundaries

Categorical thinking places distinct boundaries. Intelligent or not. “AI good” or “AI bad.” Liberal or conservative. These boundaries differ between people. In some categorizations, discrete identities are carved out, defining in-group and out-group in ways that can cause real harm.

And yet those are only symptoms. The cause is thinking a concept like “intelligence” even has natural boundaries.

Facts have clear boundaries. It’s one of the reasons school likes teaching them. Concepts are different. They have degrees and distributions, not sharp boundaries. A factual focus means you don’t have to try to align the brains of millions of teachers about the nature of fuzzy, intractable concepts.

But even the words we use for concrete things are often concepts. I use the concept of “chair” when discussing AI’s conceptual nature in AI Wisdom Volume 1: Meta-Principles of Thinking and Learning.

Let’s try to define what counts as a chair and what doesn’t. Four legs and a seat? That excludes bar stools. Something to sit on? That includes rocks and floors. Does it need a backrest? What about bean bags? Exercise balls? Tree stumps?

“Chair” isn’t a fact with crisp boundaries. It’s a concept—a fuzzy distribution of examples with varying degrees of “chairness.” Some things are definitely chairs. Some are chair-like. Some are edge cases where reasonable people disagree.

Imaginary objects categorized by “chairness.” Each dot is an individual chair example. The categorizations are arbitrary for purposes of illustration. Real concepts have many more axis “dimensions” to consider, but the mentality of concepts as clouds of “ish” is the key. From AI Wisdom Volume 1.

This is how almost all concepts work. They have extent, degrees, and fuzzy boundaries that blur into neighboring concepts. There’s no bright line where “hill” becomes “mountain” or “warm” becomes “hot.” We navigate these spectrums intuitively every day without demanding precise definitions.

We need to choose a word when communicating. That’s a categorization but a necessary one. It’s an imperfection of language in describing more continuous phenomena. We need to pick some imperfections. We understand words themselves as having shades. “Bad” and “unfortunate” overlap only sometimes, and learning those subtleties is a big part of cognitive maturation.

Yet when it comes to “intelligence,” people suddenly insist on binary answers. Is AI intelligent or not? The question is malformed—like asking whether a bean bag is “really” a chair. You can debate it forever without resolution, because you’re demanding a categorical answer to a dimensional question.

Intelligence exists on a continuum that extends further than most people realize. Plants respond to stimuli and optimize resource allocation. Insects learn and adapt their behavior. Mammals strategize and solve novel problems. Humans build abstract theories and write symphonies. Where exactly does “intelligence” begin? Wherever you draw the line, it’s arbitrary. No definition can handle it. It’s a dimensional concept being described in terms of similarly squishy words. Useful for communication, but not for how we should think about the world.

The Philosophical Rabbit Hole

I often call philosophical musings about AI “rabbit holes.” Not worthless if based on a clear understanding of both AI and a realistic brain, but that’s not common.

Having been close to the AI world for several decades, I’ve seen this many times before on a smaller scale. These discussions happen with those new to thinking about AI. I’ve been in study groups that debated AI definitions for weeks, to little effect. Those who’ve worked with these systems for years don’t waste energy on the terminology. They accept that intelligence is defined by what a system can do and what it can learn. They think functionally, and most functionality is continuous rather than binary. Even if the output is binary, the characterization of performance is statistical.

Experienced AI operators know they’re usually nudging the qualitative, vague, and indescribable aspects of AI “thinking” toward their goals. They feel the “alignment issue” as AI imperfectly representing concept distributions. “Original thinking” doesn’t always align with my notion. Perhaps we agree on those examples (points on the “original thinking” scatter plot) in the center of the concept. But on the edges, we may not. The AI or us may not have seen the unusual cases. They might have a culturally shifted concept, as with color perception.

So when I see debates about “intelligence” or “knowing” or whatever I mostly think “unimportant.” Everyone goes round and round arguing definitions some medieval scholar wrote, giving apparent rationale that just substitutes one vague term with another. It’s “intelligent” because it “understands,” or “reasons.” Each objection just adds fog. Yesterday someone told me AI isn’t intelligent because it mimics humans. Um…ok…do we discount young children for whom mimicking is a primary learning approach as unintelligent? We’re not clarifying anything. We’re simplifying where simplicity doesn’t exist.

These objections reveal more about how critics think than about AI’s limitations. They’re protecting a category—”intelligent”—that they need to remain exclusively human. Every time AI crosses a threshold, the threshold moves. Chess was intelligence until Deep Blue. Go was intelligence until AlphaGo. The definition becomes “whatever humans can do that machines currently cannot.” (A joke in AI circles for decades.) That’s not analysis. It’s defense.

The specific objections fall apart under scrutiny, but that’s almost beside the point. What matters is the pattern of thinking behind them.

The Dimensional Thinking Imperative

That pattern is categorical thinking. It’s often characterized by binary answers requiring simple explanations. Complexity is wished away. We live in a sea of black-and-white thinking, but real-world challenges rarely have simple answers.

This is especially damaging in an AI era. AI natively works in conceptual space with notions that have degrees, extents, caveats, and uncertainties. That’s actually why it struggles with discrete facts. Categorical thinking from humans operating AI will actually neuter AI’s abilities. You get better results when you think dimensionally about what you want, when you understand that AI capabilities exist on spectrums rather than in boxes.

Yet the practical stakes are higher still. All this philosophical hand-wringing serves to diminish AI’s abilities while inflating human abilities. It’s a comfort blanket. If you’re genuinely worried about AI’s impact—on jobs, on education, on how we work—that comfort blanket is exactly the wrong response.

Employers don’t care about philosophical categories. They care about output. Can the system do the work? Is the quality acceptable? Is it faster or cheaper? For many practical tasks, AI outperforms the average human worker right now—not in some hypothetical future. Arguing that AI isn’t “really” intelligent doesn’t change that reality. It just leaves people unprepared for it.

Remember why school likes teaching facts? Clear boundaries. Easy to test. No need to align millions of teachers on fuzzy, intractable concepts. The same logic extends to everything else traditional education emphasizes. Repetition. Single correct answers. Strict subject boundaries. Emphasis on detail over relationships.

Studies show students become less likely to explore multiple approaches as they progress through school. Schools train dimensional thinking out of them.

The AI intelligence debate is a symptom of this deeper problem. Teaching them to evaluate capabilities on a spectrum—what AI does well, what it does poorly, how those interact with human abilities—is far more valuable than teaching them philosophical objections that feel satisfying but prepare them for nothing.

If I were leading an educational institution, I’d focus on a small number of goals relevant to every class, teacher, and student. Dimensional thinking would be one of them. The ability to see concepts as overlapping, interacting, and evolving. The comfort with ambiguity that lets you work productively without bright lines.

Because I really don’t give a hoot if AI is really “intelligent.” Some days I’m not sure I am.

©2026 Dasey Consulting LLC