Aardvark Daily

Please visit the sponsor!

A huge copyright storm is brewing

I wrote a column about Stable Diffusion a month or so ago in which I marvelled at just how clever this machine-learning/AI system appeared to be.

Well thanks to things such as the DALL-E service, anyone can effectively harness the power of this system to create an artwork from a few descriptive words.

The problem is that this system is now blurring the dividing line between being "inspired by" and simply "using" other copyrighted works. The result could be a huge copyright bunfight.

Those artists whose work clearly has a role in the creation of these new computer-generated works may lay claim, under copyright law, to some kind of compensation if the newly created work is sold for monetary or other gain.

Oh dear.

Right now a number of companies are scurrying to position themselves in a way that avoids legal complications related to the potential infringement of a myriad of copyrights, perhaps even in a single AI-generated image.

The photo library company Shutterstock has made an announcement today in which it claims that it will compensate artists whose work has been used to train the DALL-E system.

However, the devil may be in the detail and there is no clear indication as to exactly how those artists will be compensated or what the formula for calculating such compensation might be.

Perhaps it'll be a small credit for each and every image that contributors have uploaded to Shutterstock or maybe it'll only be paid to those artists whose works are actually being used in any specific AI-generated picture. Either way, the payout is likely to be very small and will perhaps only exist to prove that the work has effectively been licensed for this purpose -- so as to avoid legal action in future.

The real problem arises from the fact that in order to achieve its extraordinary abilities, DAL-E has been trained using a huge number of images seemingly scraped from the web and there appears to be no way to compensate the creators of those original works for any appearance they may have in derived works.

I predict some real bunfights over copyright once these AI-generated images start appearing in expensive advertising campaigns or are sold as valuable art works on their own.

Perhaps the best analog of this situation is the case of "sampling" in the music industry.

If a musician decides to use a tiny snippet of another musician's work as a part of their own creative work, that usually involves obtaining a specific license, even if the pitch, tempo or other characteristics of the sample are altered in the process.

It strikes me that what DALL-E and other AI systems are doing is exactly this... taking a piece of one creative work and sampling it for use in works it creates itself. Just as with music there should be compensation to the original artist however, when the training data is simply billions of images scraped from the web, how do you do that?

What's more, how does anyone assert their copyright if they think that an AI-generated image has infringed their copyright? It is possible that several different artists or photographers may think their work has been used in an AI-derived work. Is there any way to tell for sure?

With some commentators speculating that AI-created imagery may soon eliminate the need for artists and photographers I am picking that there will be an awful lot of disenfranchised "creative types" seeking compensation through the courts.

It looks as if ShutterStock thinks the same way and is trying to preempt any such move. The big unknown is just how many of its creators will be happy with the paltry token sum it will likely offer them.

Carpe Diem folks!

Please visit the sponsor!

Have your say in the Aardvark Forums.

PERMALINK to this column

Rank This Aardvark Page

Change Font