Here’s a question that sounds simple.
When you see a red apple, where does the redness come from?
The obvious answer, from the apple. The apple reflects certain wavelengths, your eye detects them, your brain processes the signal, you perceive red. Information flows in from the world, gets processed, produces experience.
Except that’s not what happens.
A person with normal color vision, a person with color blindness, and a mantis shrimp with sixteen types of photoreceptors are all looking at the same apple. They’re receiving completely different experiences from the same physical event. The redness isn’t in the apple. It isn’t even in the wavelength. It’s produced by the structure of the organism doing the perceiving.
Humberto Maturana, the Chilean neurobiologist spent years studying color perception in frogs. His conclusion? The nervous system doesn’t process information from the world. It produces a world consistent with its own structure.
This is a radical move and it’s worth sitting with.
The traditional model says: world sends signal, organism receives and processes signal, organism builds internal representation of world. The organism is essentially a receiver and processor. Information flows in.
Maturana’s model says: the world triggers structural changes in the organism, but what those changes produce is determined entirely by the organism’s own organization. The nervous system is operationally closed. It doesn’t reach out and grab pieces of reality. It maintains its own coherence while being perturbed by an environment.
What you perceive is not a picture of the world. It’s a stable pattern your nervous system produces in response to perturbation. And that pattern is shaped by everything you are, your biology, your history, your previous perturbations, your structure at this moment.
This is why two people can witness the same event and construct completely different accounts of what happened. Not because one is lying or careless. Because they are different structural systems producing different coherences from the same trigger.
Winograd and Flores import this directly into their critique of AI. If cognition isn’t information processing, if it’s structural coherence-maintenance in a biological system with history and embodiment, then building intelligence by optimizing information processing was solving the wrong problem from the start.
You can’t get to knowing by getting better at processing. They’re not on the same spectrum. They’re different phenomena.
This is where the conversation about AI gets genuinely interesting rather than just critical.
LLMs process at a scale and sophistication that dwarfs anything before them. But the processing scale is still processing. If Maturana is right, and Flores believed he was, then what’s missing isn’t more parameters or better training data.
What’s missing is structural history. Embodiment. The fact of having been a thing in the world that encountered resistance, maintained coherence through breakdown, developed a nervous system shaped by actual stakes.
An AI has no stakes. Nothing it generates has ever cost it anything. No representation it produces emerges from the need to maintain coherence under genuine perturbation.
This doesn’t make AI useless, a calculator has no stakes either and it’s extraordinarily useful. But it does mean the “knowing” you’re waiting for, the thing that would let someone debug code they didn’t write and don’t understand, that isn’t a processing problem waiting to be solved with more compute.
It’s a different kind of problem entirely.
And we may not have the framework yet to even ask it correctly.
@ Lee Greene