AI Tokens Are 10,000 Years Old: From Mesopotamian Clay Tokens to Language Models

The word “token” has become one of the defining terms of the AI era. We pay for them, track them, and argue about their cost. But here’s the thing — the concept of a token is not a 21st-century invention. It’s nearly 10,000 years old.

What Is an AI Token and Why Does It Matter

With the rise of AI, the word “token” appears everywhere. It’s the unit we pay for when using OpenAI, Anthropic, Google, and every other LLM provider. Tokens keep getting more expensive. The more tokens a model can process, the more powerful it becomes.

So what exactly is a token in the world of AI? It’s the smallest unit of text that a language model uses to break down a request. It can be a word, part of a word, or even a single character. On average, one token equals about 0.75 words in English.

Modern models are trained on trillions of tokens, and both input and output are measured in token counts. This is the new unit of intellectual labor — the foundation of pricing and billing across the entire AI industry.

It all seems like the most cutting-edge concept imaginable. But the idea behind tokens is actually about 10,000 years old.

The Ancient History of Tokens: Mesopotamian Clay and the First Accounting System

Around 8000 BCE — five thousand years before the first cities and written language — the people of Mesopotamia, in what is now Iraq, began using clay tokens.

Small objects, roughly the size of a finger joint, fired from clay in various shapes. Each shape represented a specific commodity: a cone meant a small measure of grain, a sphere a large measure, a disk represented a sheep or goat, a cylinder stood for a cow, a tetrahedron marked one day of human labor, an ovoid indicated a jar of oil.

These weren’t just numbers — they were numbers with meaning. For thousands of years, tokens sat in temple storehouses, traveled with trading caravans, and passed from hand to hand.

How AI Tokens History Mirrors the Shift From Physical to Abstract

Then efficiency won. Around 3500 BCE, people grew tired of hauling these physical tokens around and decided to simply record numbers on clay tablets instead.

That’s how mathematics moved from the physical world into abstract notation.

When you think about it, AI tokens today and tokens from 10,000 years ago are remarkably similar. Both are the minimal unit of meaningful accounting — a measure that defines “capability” and “power.”

What will AI tokens look like in another thousand years?