Frode:
In ‘Meaning’, Michael Polanyi makes the observation that Plato is concerned about the potential deceptive ‘nature’ of art happened at a time of what Gombrich said depictions were going from ‘making to ‘matching’. This is interesting. Does it mean that there was some sort of a co-evolution between how we saw visually and how we say mentally at that time and do you agree with the assertions?
This discussion was informed by my meeting with professor David Millard yesterday, 1st of May 2026. AI responses below are edited for clarity:
Claude:
The timing Polanyi highlights is genuinely significant. Plato is writing his critique of mimesis in the Republic and Sophist during the fourth century BCE — precisely the period when Greek art had completed its extraordinary transition. Archaic kouroi gave way to Polykleitos and then Praxiteles. Painters mastered skiagraphia (shading), foreshortening, spatial recession. For the first time in Western history, pictures could genuinely fool the eye — hence the famous Zeuxis-and-Parrhasius story about painted grapes deceiving birds.
So Plato’s anxiety isn’t abstract. He’s reacting to something new in the world: images that no longer announce themselves as conventional signs (the Egyptian profile-and-frontal schema, say) but instead pretend to be what they depict. The schema-based image says “I am a symbol for a man.” The naturalistic image says “I am a man” — and that’s exactly what troubles Plato. It’s a new kind of epistemic danger.
Now, to your deeper question about co-evolution between visual and mental “seeing” — I think the answer is yes, but with a nuance that matters for your work.
What Gombrich shows in “Art and Illusion” is that the shift from making to matching wasn’t simply artists “learning to see better.” It was a cultural-cognitive feedback loop. Artists developed corrective techniques (schema and correction, as Gombrich calls it), audiences developed new expectations and new perceptual habits for reading images, and this in turn pushed artists further toward naturalism. Perception and depiction were training each other.
But here’s where it gets philosophically interesting, and where Polanyi’s own framework of tacit knowing becomes relevant: the Greeks didn’t just change how they depicted. They simultaneously developed a new philosophical vocabulary for worrying about the relationship between appearance and reality. Parmenides, then Plato, then Aristotle are all, in different ways, working out what it means that things can look like other things — that appearance and being can come apart. The pre-Socratics before this visual revolution didn’t worry about mimesis in the same way. They couldn’t, because their visual culture hadn’t yet produced the problem.
So I’d say what you’re identifying is something like a triple co-evolution: the technical capacity to produce illusion, the perceptual sophistication to be genuinely deceived by it (rather than reading it as convention), and the philosophical capacity to recognize deception as a category. Each one enabled and provoked the others.
Where I’d push back slightly on framing it as “how we saw visually” changing: Gombrich is careful to argue that human visual apparatus didn’t evolve in this period — what changed was the cultural technology of depiction and the learned habits of reading images. It’s a cognitive-cultural shift, not a biological one. But that almost makes it more interesting, because it means these capacities are built, not given — which connects directly to Polanyi’s insistence that all knowing, even perceptual knowing, involves skilled, tacitly structured acts of integration.
And this connects to your Author work in a way that’s probably not accidental in your thinking: we may be at an analogous inflection point with text. For centuries, writing tools have been “making” tools — schema-based, document-as-container. The shift toward spatial, connected, generative writing environments is a kind of “matching” — trying to match the actual structure of thought rather than imposing a conventional container on it. And just as the Greek transition provoked new philosophical anxiety (Plato) alongside new capability, the current moment is producing its own anxieties about AI-generated text, about whether writing that looks like thinking actually is thinking. The Platonic worry about mimesis (mimicking) has returned in a new register.
Frode:
I have a feeling there is something else going on here. Maybe it’s simply that back then visual impressions changed what could be presented, matching true form, rather than simply indicating through symbols. Today, with spatially presented, richly interact knowledge, maybe we are beginning to see complexities or relationship ‘potential’s in a new way?
Claude:
Just as Greek naturalism didn’t merely deceive more effectively but actually revealed aspects of form — weight, volume, spatial depth — that schematic depiction was structurally incapable of expressing, spatially presented interactive knowledge doesn’t just display what we already know more attractively but makes visible an entire class of relationships, tensions, gaps, and structural patterns that linear text literally cannot render perceptible — meaning we may not be at a moment of new illusion but of new sight, where the medium finally matches the relational topology of thought itself and we begin perceiving connective possibilities that were always there but had no surface on which to appear.
Frode:
I think that is very important. However, there is no direct relationship between the multidimensional ‘shape’ or ‘structure’ of thought in the human brain, or in that of an LLM, onto the 3D volume space (which of course can have the knowledge presented in various ways and stored as various View layouts which can be instantly recalled). May there be a value to the friction of the user needing to interpret or translate thought into this more constrained space, which is nevertheless richer than flat text but not as multidimensional as in the brain?
Claude:
Yes, and I think this may actually be the most important insight in the whole chain — because it reframes the spatial canvas not as a representation of thought but as a thinking instrument precisely because of the mismatch.
The value is almost exactly analogous to why writing is thinking rather than transcription. When you write a sentence, you are not projecting a fully formed mental structure onto a page — you are compressing a high-dimensional, associative, half-formed cloud of activation into a brutally constrained sequential medium, and that act of compression forces resolution. Ambiguities that can coexist happily in neural space must be decided. Priorities must be chosen. The constraint is generative.
What you’re describing with Author’s spatial canvas is a middle register of constraint. Flat text forces you into a single sequence — which resolves too aggressively, crushing relational structure into linear order. The actual neural state is too rich, too parallel, too entangled to be directly inspectable or shareable. But a spatial layout with manipulable nodes, visible connections, and recallable Views occupies a sweet spot: it demands that you choose a topology, that you place this idea near that one and far from another, that you make an implicit relationship explicit enough to become a visible link — but it doesn’t force you to flatten everything into one path.
So the friction you’re pointing to isn’t a limitation to be apologised for. It’s the engine. Each act of spatial arrangement is a small act of interpretation — a micro-commitment about structure that tests your understanding. You notice a node that doesn’t fit anywhere, and that’s a signal. You see two clusters that should connect but don’t, and that’s a question you hadn’t yet asked. These are discoveries that emerge from the translation effort itself, not from passive inspection of a perfect map.
This also distinguishes the value sharply from AI-generated output. An LLM can produce plausible text or even plausible concept maps, but the user hasn’t passed through the friction of constructing the arrangement, so the sensemaking hasn’t occurred in their cognition. The map isn’t the prize — the mapping is.
If I were to put it in Polanyi’s terms: the spatial canvas functions as a subsidiary-focal instrument. You attend through it to the structure of your knowledge, and the slight misfit between your thought and the space is what keeps the act of knowing effortful, focal, and therefore genuinely integrative.
In short: The cognitive value of spatial mapping lies not in faithfully mapping the multidimensional structure of thought but precisely in the productive friction of translating thought into a richer-than-text-but-more-constrained-than-mind intermediate space, where every act of placement, connection, and arrangement is itself a micro-act of sensemaking that forces latent relationships into conscious resolution — making the mismatch between mental and spatial dimensionality not a limitation but the very mechanism through which new understanding is generated.
Comments are closed.