Training Data

Somewhat topical post today: the subject is AI Art. Specifically, one complaint/controversy which seems to hang over it. Apparently one of the complaints about AI art as a concept is that the AI was “trained” using a huge amount of art created by human artists, and those artists weren’t compensated or even consulted before this happened.

If you’re not sure what I mean, think of it like this: if you Google “paintings of flowers” you’ll get a bunch of, well, paintings of flowers. AI art-bots are trained using that data to know what a “painting of a flower” looks like in the aggregate and then is able to produce a painting of a flower that draws on those elements even though it isn’t an actual copy of any of those paintings.

And this is my objection to the objection. When my daughter was learning how to paint flowers, she also Googled a bunch of paintings of flowers, and that’s how she learned to do the exact same thing – make paintings of flowers that weren’t copies of any specific piece, but that contained many of their elements.

She didn’t email those artists to ask their permission. She didn’t pay them. She just looked at their art.

That’s how everyone learns everything.

I learned how to write by reading. I didn’t pay those authors for teaching me. We all stand on the shoulders of giants, and all art is partially inspired by art that comes before.

I suspect that people who object to AI art aren’t really objecting to that facet – that’s just a convenient proxy for their greater complaint, which is the age-old complaint of technology taking away our jobs that has pervaded throughout all time. People said this stuff about Photoshop… and about cameras, for that matter. But just like with all those things (and all the ones that came before and will come hence), the complaint is without merit. Technology advances, and it certainly changes art – but there will always be artists. Some will use those tools and some won’t, and that choice is also part of art. Some people play a guitar and some people use an entire sound studio full of effects and auto-tune and synthesized sounds, but the end result of both methods is still music.

And that music, however you make it, trains the next generation of musicians. And they don’t ask your permission.