A lot of art can't be reverse Engineered

#done

A lot of art can’t be reverse-engineered

The construction of a most paintings isn’t something you can reverse-engineer because they weren’t ever forward-engineered. [Rembrandt], I will assume, was not working from a detailed [prompt]. The problem space of how to put together a Rembrandt is so poorly-defined and physical, that I think making a new one would even slow down a superintelligence. Maybe a truly intelligent machine could make a passable Rembrandt one day, but it couldn’t do it quickly. Paint has requirements that can’t be intelligenced away. Many important materials and methods remain both undocumented and unavailable. If we want someone to make a Rembrandt-like new painting, a single, well-trained human painter is likely to do better than a machine – or teams of people using machines – for a long time.

Note that the problem is not just in making paintings as good as Rembrandts, or even as old. It can be just as hard to assemble a bad or otherwise unforgettable painting.

This one of the most brutal lessons of being an art student! You can work hard and produce garbage. I suspect this is a lesson that can show up in a very real way even if you only ever make images by typing prompts.

Most paintings are themselves the most detailed available records of a specific set of decisions made with a specific set of materials at a specific moment in time. No comparable records exist outside such paintings, and much of the record is lost within them.