It took nearly a century to figure out just 430 of these Nazca geoglyphs, but now AI nearly doubled the number overnight, adding 303 new geoglyphs to our knowledge.
AI might've also revealed why the Nazca lines were constructed!🧵
For background, the Nazca lines are a set of exceptionally well-preserved geoglyphs and walking routes that exist in the agriculturally-unsuitable Nazca Pampa region.
The traditionally-known lines seem to depict things that make sense. For example, here's a spider:
Line construction is a practice from the region that's at least 2,000 years old and it results in lots of very interpretable pictures, like this monkey:
The lines have been classified into many types.
It's believed that the different types are distributed in geographically distinct areas, created in different times, and most obviously, that they have different meanings.
The new lines discovered by AI are a bit harder for humans to understand or pick out, but when the AI points them out, it becomes apparent we were overlooking them, perhaps because they're so weird and foreign to us
They're also older and smaller than known ones. Take a look:
These newly-discovered Nazca lines depict very figurative rather than literal shapes, but they also depict ritual, and maybe even lawgiving or war.
These lines are weird precisely because they had to be distinct to suit their purpose.
I'll come back to this.
The new lines were found by the AI like so.
The AI highlighted certain areas as being particularly 'line-like' (A), and then the researchers visually inspected the photographs the AI had been provided (B).
After the AI pointed it out, it was often clear something was there.
This exercise was repeated over and over again throughout the whole region between Lima and Nazca, leading to a lot of new targets.
The team then set out on foot and by drone to see if the lines were real, in-person.
Real lines are clearly dug into the ground and often contain artefacts like pottery shards and such.
The AI found them and the humans confirmed them.
A very large number of these new discoveries went overlooked for so long because they were reliefs.
Reliefs are less distinct in the landscape, but are still persistent, like the better-known lines. They also depict different things than the lines do:
When you look at the reliefs, you'll see evidence of ancient domestication and lawgiving or warring.
The lines, however, more often just depict an animal. Why might they be so different? It's not like the complexity differed all that much.
Here's the meat:
What makes these lines so interesting is that the AI makes it clear that they had distinct purposes.
The relief-type lines that show people and such? Well, those mark trail-heads. They're less distinct because you're intended to be close to them!
If you wanted to navigate across the region, you could walk until you saw a certain relief, and then you'd know what trail to take!
Critically, this network was informal. The state, however, made a formal network leading to the Cahuachi Pyramids, which were a ceremonial center.
The lines rather than reliefs are younger because they're associated with state formation/organization.
The region's state co-opted the informal, cultural practice of making reliefs for navigation in order to set up the line network to get to and be seen from the temple complex!
And there we have it! We might now know the purpose of and who funded the Nazca lines. (Thanks, AI!)
To review, there's an old, informal line network used for navigation. You walk up a trail, you see a relief. The number of these known just blew up due to AI.
The old informal line network where you see a relief then you go down the right trail was formalized by the region's state to create massive lines, sharp breaks that are more obvious from a distance
These lead to and support the temple complex, facilitating worship/organization
And guess what? With the power of AI, we're just getting started.
The authors of the study said that there are more than 250 additional geoglyphs flagged by the AI, which they didn't have the time to examine in person.
What else will we learn about this ancient civilization?
OK, this is just absurd now. The Biden-Harris administration has attacked three emergency services agencies in a week because apparently these exams are too hard for Black people.
Want to see how hard this exam is? Let's look at some questions.
Lots of people are saying this is frivolous because firefighters should have to pass some test, but let's go ahead and look at some test questions so you can see why this test is just way too hard.
This first one's a doozy:
Sorry if you're already struggling, but please know: You're not allowed to use a calculator for this one.
"SAT scores just reflect zip codes" is probably dead.
A new study used a sample of 760,000 military children whose families were randomly assigned to different counties/zips and found living in a +1 SD county or zip code for twenty years upped scores by just 0.05 or 0.19 d:
That 1 SD gap between Blacks and Whites? It's not explained by Blacks living in about 0.6 SD worse neighborhoods. Maybe about 10% is.
That's probably to much though, because the instrumental variable analysis suggested the sign of the effect on SAT scores was negative!
The authors had this to say on the negative estimated effect on SAT totals:
Looking closely, all of the causal estimates of place effects on SAT scores were at best marginally significant, unlike the effects on attained income, college attendance, and earnings.
The College Board just released this year's SAT scores!
I thought I'd go ahead and put everything in familiar terms and make some plots.
This thread will have a lot of pictures. First up: How did everyone do?
All of the typical race differences are there. Blacks did roughly 15 IQ points worse, Hispanics did about 10 points worse, Asians did similarly better, etc.
If we scale all that by the sizes of the populations who took the tests, we get this:
Another way to look at this data is to stack everyone into a single population, like so: