I have signed away my rights thousands of times and you know you have too. I need to use that digital tool right away due to a deadline. I never read the T&Cs anyway so, yes, I agree and just give me access already.
A Line in the Sand for AI Training and Fair Use?
The AI race swirling around data pushes companies to cross a pesky line in the sand—training AI systems on copyrighted material. Fair Use is not too clear about where to draw that line and, due to companies’ need for more data, the question of where the line stands is heating up.
Original creators, who have copyrighted their work, are grappling with what it means for their copyrighted creations to be fodder for AI training. Companies need more data to train their AI systems and easier and frictionless access to that data. Having exhausted public domain sources, they look to training with copyrighted sources.
Who Owns What
Digital tools we use to create, such as the LinkedIn tool I am using right now to write this article, will use our input/output to train AIs. If you are an artist who makes a living from your creations, surely you have questions. Take, for example, Adobe’s recent flareup with users due to their updated Terms of Service stating:
“We may access your content through both manual and automated methods, such as for content review.”
ChatGPT 4o Experiment with Copyrighted Works
On the other hand, when I asked ChatGPT 4o, for a rendering of an “AI system training on copyrighted paintings,” it gave me an image featuring the Mona Lisa.
I’m no lawyer, nor art historian, but I do know da Vinci’s masterpiece is public domain and not copyrighted. 4o’s output was INcorrect, but it did not cross that copyright law line in the sand in serving me up the DALL·E. However, I wish 4o had responded to my first prompt by explaining it could not grant my request since reproducing copyrighted paintings infringes copyright law. Instead, it decided to serve me up a DALL·E depicting public domain masterpieces without any explanation.
Out of curiosity, I asked 4o to then “use an Andy Warhol instead of the Mona Lisa” since Warhols are copyrighted. I was reassured:
“I was unable to generate the requested image because it involved using copyrighted material, which goes against our content policy.”
The irony is Warhol infringed on others’ copyrights in his artwork.
AI Training in the Realm of the Copyrighted
The question remains if 4o and other AIs get—or will be—trained on such copyrighted works.
With the two main players in the game—the companies who offer transformational digital creation and sharing tools and the creator-customers who avail themselves of these tools, are they now opponents versus partners?
Leave a Reply