From a daily email I received that summarizes the previous day’s AI news, called “The Rundown AI:”
‘Sony Music, the world’s second-largest music group, is sending warning letters to over 700 AI developers, restricting unauthorized use of its artists’ content for training, developing, or commercializing AI systems.’
This is the latest in a string of maneuvers by companies that threaten AI model trainers with legal action if they use their data to train the AI models. This is an area of the law that is not clear to me (I’m not an attorney) and apparently not clear to a lot of attorneys. It’s uncharted territory that has developed from the recent introduction of this new type of artificial intelligence model (Large Language Models, or “LLMs”). While specifically copying someone’s work is clearly not legal, this is not what’s happening.
When an AI model is fed data, it is learning. The models typically simulate a big brain in many ways, complete with neural networks. To be clear, there are some major differences. But it is a brain-type structure nonetheless. So when they are fed data, they “learn” about it in similar ways to how a human learns. They literally use neural networks in the process.
With that in mind, say I listen to the radio or Spotify. Subsequently, if I write a song, I don’t have to pay the artists who created the music I listened to and from whom I drew inspiration in order to publish my song, assuming it’s a completely original work that doesn’t mimic or copy anything from an existing work.
Likewise, if I’m a college professor who has learned the intricacies of a particular field via much study of published research, books, and other inputs, I can produce an original work without paying a royalty for each of those external influences that went into my knowledge base. I would argue that an AI learning from the same group of resources should also not have to pay a royalty.
It may seem silly for me to categorize an AI with human beings in this regard. But the process by which it learns is basically the same. It synthesizes it all in order to create original works of its own.
So why should data made available to the public for free, arguably to learn, have to be paid for by companies producing AIs? Note, while not exactly the same, search engines profit off the results of crawling publicly available websites. They don’t pay for that information. Even though the appearance of a symbiotic relationship exists between the company and all the websites, there is no direct compensation for the use of the data.
Note that in both cases, a transformative use of the data is undertaken. And they both provide public good, with search engines providing a service that is helpful to the public and websites, and AIs providing information which can drive innovation and provide significant public benefits.
For instance, AI has the potential to assist in finding cures for diseases such as cancer. By analyzing vast amounts of medical research, patient data, and other relevant information, AI can identify patterns and insights that might elude human researchers. This capability can accelerate the discovery of new treatments and improve patient outcomes, providing a profound societal benefit. As such, a powerful argument can be made for the use of the fair-use doctrine in relation to AI learning.
The fair-use doctrine provides for the legal limited use of copyrighted material in cases where a benefit to society at large can be ascertained. This can run the gamut from quoting small portions of published materials for reviewers’ use to providing limited information that may be useful in creating a theory to cure a type of cancer.
In summary, even though I produce creative goods as an amateur photographer and blogger, I believe AIs should not be hampered from learning freely based on any law of which I’m aware. I know it may produce material that competes with material produced by other creators. But I don’t see validity in the argument that it “steals” the works of others in order to directly profit from them. Rather, it’s in general producing original works like any other artist with whom it competes.
Thoughts?
One response to “AI: should Creators be Compensated for learning From Their Material?”
Full disclosure: I use AI daily in multiple ways. And many of the pictures/drawings in my posts are generated by AI. Not in my photo galleries, but just in my posts. So I have a vested personal interest (but not financial) in its succeeding. Nonetheless, I’ve thought about this topic a lot. And, at this point, I believe in what I said above. Until someone can convince me otherwise.