The Painting You Couldn't Have Seen

I've been thinking a lot about evolution lately. Not as biology, but as a search algorithm. I'm trying to understand what parallels it has with how we train neural networks. That's led me down a lot of rabbit holes about genomes, cognition, and timelines.

The other night I went down one of those rabbit holes. Started with continental drift, ended up at cave paintings. And I noticed something I can't believe nobody has pointed out.

If you painted an animal you never saw, someone described it to you. That's not symbolic thinking. That's language.

There's a cave in Russia where researchers found a two-humped camel painted on the wall, dated 14,500 to 37,700 years old. Two-humped camels didn't exist anywhere near that cave. Their range was thousands of kilometers away. The researchers noted that "artists, or information, traveled long distances" and moved on.

They didn't follow through. If the artist never saw the animal, someone described it in enough detail to paint it. That requires real, descriptive language.

Date that painting, and you have a lower bound for when language existed.

Nobody has proposed this as a methodology for dating language. I searched. Extensively.

The method is simple. Take dated cave art with animal depictions. Cross-reference against paleofauna range maps. Find cases where the animal couldn't have been local and the artist couldn't plausibly have traveled. What's left is evidence of verbal description. The oldest case is the earliest hard evidence of language.

The cave art databases exist. The paleofauna maps exist. Nobody has cross-referenced them.

The researchers study paintings. The linguists study skulls. Nobody is standing in the middle asking the obvious question.

Why do I care? Because I want to know what made human capability compound. Humans existed 300,000 years and did nothing. Then explosions, each faster than the last. Was language the unlock? Was writing? Was it something else? The timelines matter. If language is 100,000 years old but the explosion started 12,000 years ago, language alone wasn't enough. If they line up, language is the answer.

I think about this because I'm trying to understand what makes intelligence compound in general. Not just human intelligence. Machine intelligence too.

The point isn't that the data is clean. The point is that nobody is asking the question.