How to build a better archaeologist. AI needed a hobby, and this could be it.

Between roughly fifteen hundred and twenty-two hundred years ago, in the Peruvian desert, the Nazca lived and worshiped. Also, they graffitied the hell out of that desert.

By selectively scraping away rock to reveal a distinct geologically different color stone, they made art that stood the test of time for centuries, but it’s not the easiest to find and identify so long after they were created.

Since their discovery a century ago, human archaeologists have painstakingly mapped and trekked and pored over aerial photography. Enter AI.

Look, there’s a lot of stuff AI is bad at, but analyzing massive piles of data and finding patterns is actually one where it’s actually a huge help. Using AI tools, archaeologists have discovered hundreds of new petroglyphs, ranging in size from geometric shapes up to 30 miles long to more detailed depictions of plants and animals as large as 1200 feet.

Don’t ask me why Peruvian petroglyphs are measured in miles and feet.

Maybe the human archaeologists just needed more time. But there’s also the possibility that the Nazca people just weren’t such great artists. These images really are more a testament to skill at cartography and surveying over artistic talent. I can kind of imagine an ancient Nazca committee getting together for a presentation where the lead petroglyph guy shows a sketch and tells everyone “like this, but way bigger,” and gets a green light to spend probably years worth of labor chiseling it into being. Worse, I can imagine the sketches looking a LOT more like the intended subjects, and the head carver swearing up and down that from the heavens, it most certainly DOES look exactly like he planned. It’s not like they had drones or helicopters or satellites to prove him wrong.

But AI was able to put together images from less and its preconceptions of what constitute “art” have already been shown to be fairly loose and flexible.

So… a win for AI. Maybe this is a sign that its skill really are better suited to protein folding, gene mapping, and now petroglyph identification, and it should work on more of that sort of thing.