There is a long history of artist and artisans training which entails copying, even tracing. This is probably somewhat related to the use of rote exercises and memorization in education as well. This all fell out of favor in the last 50 or so years, but I do think there are benefits.
Lets focus on copying in craftsmanship, rather than say doing a ton of boring long division problems. If I have to copy a bust of my teacher, then by copying I learn a bunch of skills related to making a bust. These are technical, such as pouring a mold, or chisling fine features, I also learn about a particular style, how it feels to create something in a particular way.
Many years ago I wrote a program that was a text editing program that created time lapse snapshots of the state of a work as a user copied it. If I copy out large swathes of Hawthorn, will I learn something, is there something in the muscle memory of writing that maybe I can capture in recording the act of writing? If not, how can I capture this.
I am thinking a lot about the work of Peli Grietzer (dissertion here) and the idea that the act of an autoencoder learning is somewhat related to the pedagogical technique of copying, or memesis in general. Here the autoencoder is engaged the act of memesis in learning. But what are the skilled that the autoencoder actually learns when the autoencoder creates a facsimile. You can say a human has a number of sub skills that are acquired in learning on large skill, but this is not the case with the autoencoder, or maybe it is – we do not know since this is unintelligible to us.