Magic Spells for Intelligence

From Icelandic Sagas to Reasoning in LLMs

If you're thinking without writing, you only think you're thinking.

- Leslie Lamport

Writing is such a foundational aspect of modern life that it’s surprisingly hard to imagine how pre-literate societies functioned. Yet that’s how humans spent most of our time on earth - without writing.

Anatomically modern Homo Sapiens have been around for 300,000 years, and while we likely had spoken language for 50,000 of those years, the earliest written script we know of is only 6,000 years old. We’ve had the ability to record, inspect, and externally manipulate our thoughts for only 2% of our existence as a species. But 2% is all it took for us to leap forward from tribal societies to globe-spanning empires, to rocket humans to the moon and manipulate our own genes.

“Until writing was invented, men lived in acoustic space: boundless, directionless, horizonless, in the dark of the mind, in the world of emotion, by primordial intuition, by terror…”

- Marshall McLuhan

What was the human mind like before it knew writing/reading? Was it as primitive and groping as McLuhan would have us believe? The answer, it turns out, reveals something profound about how intelligence works, and offers a surprising window into understanding modern AI systems.

The Mind Before Writing

To understand pre-literate cognition, the Icelandic Sagas are an ideal starting point. Unlike older epics like Homer’s, whose conversion from oral to written form happened so long ago that we can’t know how much changed in the process, the Icelandic sagas were textualized relatively recently - in the 13th century - giving us a pristine window into the transition between oral tradition and written literature.

Njal's Saga, for instance, passed down orally for centuries before being written down, is a stunning and psychologically rich tale of friendship, justice, and petty, tragic revenge. A bitter feud develops between the wives of two close friends - one a legal scholar, the other a Norse warrior - and despite their attempts to settle things through law and financial compensation, a cycle of revenge killings eventually consumes both families.

While this would create riveting drama on HBO or Netflix even today, a close reading of such sagas reveals something crucial about pre-literate cognition: oral cultures had clever techniques for composing and recalling long stories without writing. They used rhythm and repeated phrases, mnemonics, exaggerated archetypal characters, and, most interestingly, proverbs and sayings to do their cognitive heavy lifting.

In Njal’s Saga for instance, legal proceedings that follow each revenge-killing rely on proverbs to reason through cases: “Brother shall compensate for brother,” “One shall always pay half-compensation for work of the hands.” etc.

As anthropologist Walter Ong observed in his seminal work, “Orality and Literacy”, in oral cultures the law itself is enshrined in such formulaic sayings: proverbs aren’t mere decorations but constitute the law itself.

These “wisdom formulas,” as Ong calls them, weren’t just memory aids. They were the building blocks for complex reasoning. He argues that in oral cultures, proverbs “form the substance of thought itself. Thought in any extended form is impossible without them, for it consists in them.”

The insight that complex thinking can emerge from combining simple, memorable formulas turns out to be surprisingly relevant to understanding how Large Language Models work today.

How Writing Changed Our Consciousness

"Since in a primary oral culture... knowledge that is not repeated aloud soon vanishes, oral societies must invest great energy in saying over and over again what has been learned arduously over the ages." - Walter Ong

Writing transformed human societies in two ways.

The first and obvious one: it created stable external records for everything from accounting to philosophy, enabling high-fidelity communication across vast distances and time periods. Ideas could outlive their authors and progress could compound over the ages.

But the second transformation was more subtle and profound: writing reshaped our interior landscape. In oral cultures, Ong found, thinking tends to be situational and concrete rather than abstract and analytical. It was somatically embodied, intuitively felt, emotionally driven. Ideas were additive (”and... and...”) rather than subordinative. Thought was often agonistic (verbal dueling) and necessarily redundant (repetition aids memory).

Writing made our thinking more abstract and analytical. It created separation between knower and known, increasing objectivity. It helped us develop a new kind of self-aware, reflective consciousness that was lacking in oral cultures.

How Writing Generates New Kinds of Ideas

"Writing is not just a way to convey ideas, but also a way to have them."

- Paul Graham

Our minds, remarkable as they are, struggle with complex chains of reasoning purely internally. Most of our thinking remains frustratingly vague until we externalize it. The act of putting pen to paper gives shape to the inchoate mass of thought-feelings in our heads, revealing gaps in our reasoning, exposing the brittleness of our logic, the shallowness of our certainty. Writing isn’t just about recording ideas - it’s about examining them critically and generating new ones.

By externalizing the symbols of our language and committing them to physical space, we achieve new ways to examine, manipulate, and reflect on our thoughts. We become detached observers of our own thinking. No longer limited by fuzzy short-term memory and automatic System 1 processing, we can zoom in and out, see both forest and trees. We also conscript our ancient visual brain - hundreds of millions of years optimized to spot hidden patterns - giving us many more a-ha moments.

Paul Graham observes that “a good writer will almost always discover new ideas in the very act of writing.” There is, he notes, no substitute for this kind of discovery. Talking about your idea helps, but writing it down changes it and creates new ideas.

This transformation from oral to literate thinking offers a lens for understanding what’s happening in AI, particularly in the leap from basic large language models to reasoning systems.

Proverbs as Wisdom Formulas -or- Magic Spells for Intelligence

"Red in the morning, the sailor's warning; red in the night, the sailor's delight." "Divide and conquer." "To err is human, to forgive is divine." [...] Fixed, often rhythmically balanced, expressions of this sort and of other sorts can be found occasionally in print, indeed can be 'looked up' in books of sayings, but in oral cultures they are not occasional. They are incessant. They form the substance of thought itself. Thought in any extended form is impossible without them, for it consists in them." - Walter Ong

If writing was so crucial for complex reasoning, how did pre-literate societies accomplish as much as they did without it? How did they create lengthy sagas and develop sophisticated navigational systems and clever oral mathematics?

As we saw, one answer lies in proverbs and sayings - Ong’s “wisdom formulas.” These proverbs weren’t just memory aids or colorful expressions. They were compressed intelligence: building blocks for reasoning, shortcuts to frozen insight, giving people quick access to the output of higher reasoning without deriving everything from first principles. Much of thinking in oral cultures was stringing together these formulas, letting them do the heavy lifting for discourse and action.

Language as Vectors into Meaning

Isn’t it amazing how knowing a language gives you the power to understand every single possible future sentence in that language? Consider the phrase "golden mountain”. Even though no such thing exists, you can immediately picture it: gleaming in sunlight, its slopes shining with luster. By invoking this odd pairing of words, we’ve created a new kind of mental concept, an idea that implicitly has attributes of both gold and mountains.

Language isn’t just syntactic coherence, it’s a key to a certain kind of intelligence. Each valid combination of words can unlock new understanding.

How? Because meaning is connective tissue. Words and concepts don’t exist in isolation—they have semantic tentacles attaching them to other words and meanings. This web of hidden connections forms what we might call a “latent space” of meaning. I’ve written a lot more about this here.

When we say “gravity,” we’re not just naming a force; we’re accessing a web of connected concepts about mass, acceleration, apples falling on heads. When Einstein described gravity as “curvature of spacetime,” he was using language to summon a conceptual framework that fast-forwarded our understanding.

From this perspective, “a stitch in time saves nine“ isn’t simplistic advice - it’s a compressed formula for understanding causality and preventive action.

So, in a computational sense, these “Wisdom Formulas”, the proverbs that people of Njal’s time relied on to make decisions are thought-heuristics, miniature programs, algorithms that can be strung together to access intelligence from our collective past. This has parallels to how large language models (LLMs) work.

LLMs as Giant Repositories of Latent Wisdom Formulas

"Information is not stored anywhere in particular. Rather, it is stored everywhere. Information is better thought of as 'evoked' than 'found.'"

---David Rumelhart and Donald Norman (about early neural networks)

When a large language model predicts the next token in a sequence, it’s navigating through a vast space of meaning-relationships - the same latent space of connections we just described. Each token has semantic tentacles that could attach it to other tokens, and the stickiness of each tentacle changes depending on context. The LLM chooses the path that makes most sense given everything that came before.

How does it know which path to take? Through billions of patterns learned from data. When you train a neural network on the entire corpus of human writing - trillions of words capturing nearly every recorded thought - it absorbs much of the underlying deep structure of meaning. The lines connecting conceptual dots. The web of associations, conditionals, and causal relationships: “Ferraris are often red,” “if an apple is red, it’s typically ripe,” “striking a match leads to fire.”

By extracting these patterns, LLMs learn millions of miniature reasoning formulas. These are not literal proverbs, but complex mathematical relationships guiding the model's predictions. Each "wisdom formula" acts as a gravitational force pulling the vector of meaning in its direction. In this sense, LLMs function remarkably like oral minds, drawing on compressed wisdom to navigate complex situations.

Why Basic LLMs Resemble Pre-literate Minds

One key limitation of non-reasoning LLMs (like GPT-4, as opposed to reasoning models like O3) is that they’re trained to generate answers efficiently without extended deliberation. When there’s a ready-made set of patterns that apply, they produce good results quickly. But when tasks require complex reasoning without pre-existing “wisdom formulas,” they struggle. Once an LLM starts down the wrong path, it has limited ability to step back and explore alternatives. The vector of meaning is relentlessly pulled toward whatever formula seems most appropriate.

In this way, basic LLMs are like oral, pre-literate minds. They rely heavily on interpolating between compressed wisdom formulas and memorized heuristics. They lack the affordance of lengthy reflection on externalized thoughts to retrace their steps and develop more complex reasoning chains. Just as oral cultures struggled with abstract reasoning that couldn’t be encoded in memorable proverbs, basic LLMs struggle when their learned patterns don’t directly apply.

Chain of Thought: Writing's Gift to AI

The breakthrough of reasoning AI systems can be understood through our oral-to-literate analogy. These systems work in two complementary ways.

First, they’re trained on more examples of human reasoning traces - the “rough drafts” of thinking that show the process of working through problems, not just polished answers. Most internet text (the usual training data) contains finished writing but less of the iterative exploration, the finding of flaws, the retracing of steps.

Second, and more importantly, these models are rewarded for taking time to “think through” problems step by step before arriving at answers. This is equivalent to giving LLMs the affordance of writing, encouraging them to externalize their reasoning in a way that allows for reflection and course-correction.

By traveling down longer paths, the chances of grasping a better semantic tentacle for each next token increase. As opportunities to course-correct multiply, so does the possibility of genuine a-ha moments, or even the creation of never-before-seen wisdom formulas. The process of “writing out” tokens through chain of thought, then reflecting on that process through self-attention, allows for the same leap in intelligence that writing gave to pre-literate societies.

The reasoning LLM is therefore like the literate mind, having freshly discovered the affordance of writing.

New Magic for a Post-LLM World

It won’t be long before there’s a burgeoning field of LLM psychology concerning itself with understanding AI behavior and its philosophical implications. Already Karpathy calls LLMs “people spirits”—stochastic simulations created from the universe of all our thoughts.

Despite the pitfalls of anthropomorphizing LLMs (or mechanizing human intelligence), as we’ve done throughout this essay, there’s valuable insight in such analogies. Comparing non-reasoning LLMs to oral minds and reasoning LLMs to literate minds helps us grasp both the limitations and possibilities of different kinds of intelligence.

Without writing, we were cognitively constrained in ways not unlike non-reasoning LLMs. We overcame those constraints first by relying on tricks like proverbs, then by externalizing our thoughts in writing. The same tricks seem to work for LLMs too.

But we didn’t stop evolving after we learned to write. On top of writing, we developed mathematics, scientific notation, programming languages, collaborative knowledge systems—and now AI as thought partners. Each dramatically expanded our cognitive reach. What new forms of “writing” might we invent for artificial minds? We already see external tool usage—calculators, web searches, code execution. But there are many more ways to extend their intelligence, in both human-like and entirely non-human ways.

The proverb-wielding storytellers of Njal’s time could never have imagined the cognitive landscape that writing would make possible. Similarly, we may be standing at the threshold of forms of intelligence that our current frameworks can barely comprehend. The magic spells we’re teaching machines today may be just the first words in a much longer saga.

PS: If you liked what you read, hit the little heart button or restack this post so that Substack helps other people find this post