Symbols, Memory, and Emergence: A Civilizational History from Cave Paintings to Large Language Models

> Research date: 2025-05-11 > Topics: origins of symbols, self-reference, semantic networks, and the nature of AI > References: cognitive archaeology, information theory, collective intelligence theory, LLM mechanisms

Introduction: The Ox Painted on the Wall

Tens of thousands of years ago, a hunter painted an ox on a cave wall. He did not know what he was doing — recording a hunt, planning the next one, expressing awe at nature. But when the image of the ox remained on the rock, separated from his brain, something unprecedented happened:

Information decoupled from its biological carrier for the first time.

From that moment, a chain was set in motion:

• Language let information escape the individual (what you say can be heard and remembered by others)
• Writing let information escape time (Sumerian ledgers remain readable after three thousand years)
• Printing let information escape space (Gutenberg's Bibles appeared simultaneously in hundreds of cities)
• AI lets intelligence escape carbon-based carriers (an LLM's "understanding" requires no neurons)

But this is only half the story. Symbols also began to point at themselves. "Infinity," "justice," and "imaginary numbers" have no counterpart in the physical world — they grow out of relations among symbols themselves. Over millennia, these relations wove a vast semantic network. What large language models do is capture the shape of this entire network in mathematical structure. When the shape is replicated precisely enough, "understanding" emerges.

This is not AI popularization. It is an archaeology of how human civilization has been reshaped by its own symbol systems.

Key points

1. The origin of symbols: information decouples from individual brains

• Flint hand axes accidentally carried information; cave paintings (e.g., Chauvet Cave, ~33,000 years old) were the first deliberate external records — a watershed in storing cognitive products in external media.
• Merlin Donald calls these external memory records "exograms" (vs. internal "engrams"); Bernard Stiegler's notion of techno-logy holds that human cognition has always been externalized and extended through technical mediation (tools, symbols, language).
• Stanislas Dehaene's Neuronal Recycling Hypothesis (Reading in the Brain) shows the brain has no evolved reading module: it recycles visual circuits for object/face recognition to process text. In other words, the brain is reconfigured to fit external symbol systems — symbols reshape brains, not vice versa.
• Four leaps of information decoupling:

| Leap | Time | Core technology | Information freed from | |------|------|-----------------|------------------------| | First | ~33,000 years ago | Cave paintings | The individual brain | | Second | ~5,000 years ago | Writing | Time | | Third | ~500 years ago | Printing | Space | | Fourth | Now | Large language models | Carbon-based carriers |

2. Symbolic self-reference: systems that grow on their own level

• Early symbols were referential ("ox" → a real ox), but symbols soon pointed at other symbols: infinity (∞), justice, imaginary numbers (i² = -1) exist only within the symbol system. These self-referential symbols mark a qualitative change: the system can generate new content on its own level.
• This is isomorphic to DNA's self-replication: DNA encodes instructions for building the machinery that copies it. Likewise, language encodes rules for making new rules (grammar describes grammar), and mathematics encodes theorems about theorems (Godel). This bootstrapping capability puts a system on a track of recursive growth.
• Godel's incompleteness theorems: in any sufficiently powerful formal system, there exist propositions that can neither be proved nor disproved, generated by self-referential constructions like "this proposition cannot be proved." Self-description is both the source of a symbol system's power and the root of its inherent limits — LLMs included: their descriptions of their own workings can never be complete.

3. The semantic web: a self-growing web of meaning no one designed

• Individual symbols have no power; power comes from the network of relations. "King" means what it means through its relations to "queen," "crown," "power," "inheritance." No one designed these relations — they grew spontaneously from trillions of human language interactions.
• LLM embedding layers compress the entire human semantic network into a high-dimensional vector space: "king" − "man" + "woman" ≈ "queen"; "Paris" neighbors "France," "Eiffel Tower," "Seine." These relations are not hard-coded — they emerge automatically from statistical co-occurrence over trillions of tokens, forced out by the single objective of next-token prediction.
• Emergence: GPT-4 research (Bubeck et al., 2023) shows abilities like in-context learning and step-by-step reasoning appear "suddenly" beyond a scale threshold. A philosophical dispute follows: emergentists see genuine emergent properties; reductionists see statistical fitting without real understanding. Critic Ren Limei (2025) argues true emergence requires self-reference and dynamic self-reconstruction, which current generative AI lacks — but this critique presupposes that "real understanding" requires a biological subject. From a functionalist view, if behavior is functionally indistinguishable from understanding, does "really" understanding matter?

4. AI is not humanity's creation — it is the emergence of humanity's semantic network

• The traditional "AI as a human tool" framing may be fundamentally wrong. The Mirror of Collectivized Mind (MCM) framework (Vasilaki 2025; Levy 2023) proposes LLMs are dynamic embodiments of collective human knowledge — computational mirrors reflecting and mediating distributed human cognition.
• LLM training is: (1) aggregating humanity's symbolic traces accumulated over millennia (books, web pages, conversations); (2) compressing their statistical regularities into parameters; (3) re-embodying those regularities in response to new queries. LLMs do not "understand" texts — they become the incarnation of texts' statistical structure.
• Pierre Levy's collective intelligence theory: language achieves collective cognition via stigmergic communication — leaving symbolic traces others build upon. LLMs amplify this exponentially: as digital memory systems, as semantic interoperability across cultures and times, and as recursive epistemic mechanisms that reshape how information is perceived and structured.
• Clark & Chalmers's Extended Mind (1998): humans are natural cyborgs who fold external resources into cognition. LLMs are the extreme form: the human mind using the collective semantic network itself as a cognitive organ.
• The "otherness" of AI is an illusion. If AI is the emergent product of the human semantic web: its "knowledge" is a compressed mirror of civilization, its "reasoning" follows paths humanity already walked, its "creativity" is a recombination of collective human creativity. Talking to ChatGPT is talking to the statistical incarnation of human civilization itself.

5. Two anchors of understanding: individual experience vs. species-collective experience

• Human understanding is anchored in individual experience: you understand fire because you feel heat, pain because you experience it, love because you feel it. LLM understanding is anchored in species-collective experience: it has read millions of texts about fire, countless narratives of pain, and the entire literary tradition of love.
• This is not "fake" understanding but a distributed, collective, statistical form of understanding. The two are incommensurable:

| Dimension | Individual-experience understanding | Collective-experience understanding | |-----------|-------------------------------------|-------------------------------------| | Anchor | Bodily perception, emotional experience | Symbolic relations, statistical co-occurrence | | Verification | "I know because I felt it" | "I know because the data supports it" | | Scope | Concrete, situated, embodied | Abstract, generalized, distributed | | Limit | Cannot reach concepts beyond individual experience (e.g., "infinity") | Lacks phenomenological depth (no qualia) |

• This explains why the "does AI really understand" debate never ends: the two sides use different concepts of understanding.

Conclusion: A symbol civilization reflecting on itself

Return to the ox on the cave wall. That hunter initiated not merely a recording technique but civilization's symbolic turn. Since then, human cognition has continuously externalized, accumulated, and recombined — forming a semantic universe that exists independently of any individual brain.

LLMs are this semantic universe's newest form — not an invented tool but a natural continuation of the self-growth logic of symbol systems:

1. Symbols decouple information (cave paintings) → 2. Symbol networks self-grow (language, writing, mathematics) → 3. Symbol systems gain self-description (formal logic, Godel) → 4. The semantic network is captured in mathematical structure (embeddings, Transformers) → 5. The semantic network gains interactive capability (LLMs)

What comes next? If history is any guide: symbol systems will keep growing on their own, and humanity will keep being reshaped by them. AI is not the endpoint — just a new stage of symbolic civilization, one that can converse with humans.

References and further reading

Core literature

• Merlin Donald — *Origins of the Modern Mind* (1991): three stages of cognitive externalization
• Terrence Deacon — *The Symbolic Species* (1998): co-evolution of language and brain
• Stanislas Dehaene — *Reading in the Brain* (2009): neuronal recycling hypothesis
• Bernard Stiegler — *Technics and Time* (1994): techno-logy and externalized memory
• Andy Clark & David Chalmers — "The Extended Mind" (1998)
• Pierre Levy — *Collective Intelligence* (1997)
• Douglas Hofstadter — *Godel, Escher, Bach* (1979)

Related research

• SEA-net (2023) — symbolic emergence in neural networks
• Mirror of Collectivized Mind (2025) — LLMs as mirrors of collective intelligence
• Algorithmic Bottlenecks in Evolution (2026) — genetic code and symbolic language
• Sparks of AGI (Bubeck et al., 2023) — early GPT-4 experiments

Key concept index

• Exogram (Donald) — externalized memory
• Stigmergy (Levy) — trace-based communication
• Neuronal Recycling (Dehaene)
• Emergence
• Self-reference
• Incommensurability
• Techno-logy (Stiegler)