I replied to a post on a fb group where someone was asserting that “Just an fyi that Midjorney and other AI art programs learn by stealing real art from artists like me. “

I responded with something that got a bit long winded, but I think it has some valid points:

It’s not “stealing” – perhaps it’s closer to a kind of plagiarism, but that might be a stretch.

The trick here is, Copyright was established to provide a means for an artist to exclusively profit from their work, and ensuring that works of art are preserved, in exchange for limitations based on public interest considerations that are covered under fair use.

Copyright protects the original expression of an idea in the form of a creative work, but not the idea itself.

Non-exclusive factors that permit fair use are:

  • the purpose and character of one’s use;
  • the nature of the copyrighted work;
  • what amount and proportion of the whole work was taken;
  • the effect of the use upon the potential market for or value of the copyrighted work

The established precedents guide us on where those lines are and machine-learning is definitely forging some new ground here. Typical clear fair use exclusions are: dealing uses are research and study; review and critique (literal and parody); news reportage and the giving of professional advice

If a person learns to draw by reading and replicating Garfield comics, so the style is inherently Garfield-esque, would drawing comics about a different cat be ‘stealing’? Likely not. How does automating that so a machine does it alter the equation?

If ‘study’ infers learning, and learning == machine learning… that’s definitely troublesome.

But on the other hand, copyright does grant the copyright holder “exclusive rights to create derivative works”, which is why you can’t create a Mickey Mouse poster from scratch and sell that (independent of Trademarks, which in that case are also involved)

The question is “what is a derivative work”?

I suspect that the courts would be loath to make that too broad, as the impact on society would be chilling, and a slippery slope for who owes what for how much of an infringement.

Advances in technology have constantly pushed the boundaries of Copyright into new areas, and we’re crossing another threshold here. Some notable advancements:

  • Printing Press made it so that a book could be printed many times with little effort, making copies a lot more feasible to do
  • Photography made it easy to capture a work of art that was simply impossible before.
  • Motion picture films made it trivial to capture a live performance so that someone could replay that without the original actors
  • When computers came about, the notion had to be extended to software, where duplication wasn’t even a physical thing (just odd arrangements of bits on a disk)
  • VHS, DVD, etc made those motion pictures easy to replicate
  • In the smartphone era, suddenly everyone has a camera and a studio in their pocket to create, edit and extend anything they see…
  • And now we have machine learning, which creates a whole new breed of transformational works.

Where does this lead us? Only sweet merciful Zeus knows.