The State of Copyrights and Generative AI
By Betsy Burton
The State of Copyrights and Generative AI
I spent the last several days doing some research and attending some webinars on the issue of copyrights as they apply to AI-generated text, images, sounds, videos, etc.
Let me be clear that I’m not a copyright lawyer but rather, I am like a lot of people who are trying to figure out how copyright applies to generative AI.
State Of Copyrights
US protections: The US copyright office is the place where copyrights are applied and determined. As such the US copyright office has recently released guidance on applying copyright to content used within generative AI. This is the best source for information about US copyright laws.
In terms of applying protections over the internet and AI content, the Copyright Office provides guidance, ways to apply for copyright, and reviews of case-by-case situations.
International protections: In addition, there is an international agreement, The Berne Convention.
The Berne Convention was created in 1886 and has been signed by 181 countries. The basic premise of The Berne Convention is that a country agrees to recognize the protection of work from artists and authors from other countries in the agreement to “at least” as well as those of its own nation.
The catch is that while the Berne Convention provides some minimum standard for protection, only registered works are protected with statutory damages and attorney fees.
This means it requires a lot of diplomatic and legal work for content creators to try to enforce The Berne Convention. And only registered works are protected with any fines.
AI Output Protections
Dealing with the output of content from generative AI is much easier than input. This is because AI systems are not human. Copyright only extends to human-created works that are not in the public domain.
For example, the case of the monkey selfie is not copyrighted because the monkey is not human.
The same applies to AI-generated content. Unique AI-generated content is not copyrighted because the AI system is not human. But this is where the output issue gets more complex.
If a generative AI system outputs unique content that is then changed by a human, is it copyright? Possibly, it depends on how much the content is changed by a human.
Let’s say I take the image output from generative AI and use it as the basis for a whole new image – then this could/might be copyright. And in this case, it is the changes that are copyrighted – not the whole image.
AI Input Protections
So, what about the content used to train an AI system overtime? This is, from what I can tell, pretty uncharted territory.
Pros For Training AI with Copyright Content
One side would argue that training AI is like training a human. That everything we read, hear, see, smell, and feel are inputs to our learnings as a human.
Using this as a model, the argument would be that we can’t say all books written by humans about vampires are copyright violations because the Twilight book series is copyrighted. Taking this line of argument further, we can’t say that AI systems can’t be limited by copyright and be precluded from generating new content.
This side argues that teaching AI systems freely will increase innovation and opportunities; the deregulation argument.
Cons For Training AI with Copyright Content
The other side of the argument is that we need to protect copyrighted content from being used to train AI systems because the AI system will generate content that then competes in the market with the original content. This case is being argued in the suit Getty Images is taking with Stability AI.
This side of the argument states that allowing AI systems to be trained on copyright material without licensing and paying for that right will kill the content creation market. There will be no reason for artists, writers, cinematographers, or musicians to create their content because it can all be used unfettered by AI systems; the creative rights argument.
The Monkey In the Argument
The reason why these issues become really complex can be tied to the monkey selfie settled case.
If we teach a mynah bird to sing exactly like Beyonce for commercial purposes, that is protected by Beyonce’s copyright (melody and lyrics). But if our bird listens to a Beyonce song and creates a new song based on what it hears and sings it for commercial purposes, is that a copyright violation? Likely not.
If an AI system collects content based on what it is taught/input and learned/input over time and it creates wholly new content based on its learning, is that a copyright violation?
How do we handle the copyright issues when a system learns like a human? For AI everything is can be an input – listening, reading, feeling, seeing are all inputs for learning.
Or… should we treat AI systems like tools that are used to mimic, deepfake, mash-up, and create new content based on an artist’s sound, voice, image, words, and melody – thus subject to copyright.
Music companies are negotiating with generative AI providers to license artists’ melodies and voices for use in AI-generated songs. While this may work for major artists with lots of lawyers and representatives, it still leaves the issue open for everyone else.
It feels like it is quickly becoming too late. Generative AI systems are being created and input with content at an alarming speed. The generative AI Pandora’s box has already been opened really wide.
Again, I am not a lawyer or copyright specialist, and, frankly, I don’t think they even have this all figured out – especially internationally. This issue is far from settled and will become even more complex as AI generation capabilities become capable, responsive, and richer.
How Will AI Impact the Workplace?
Does the evolution of artificial intelligence (AI) into the workplace mean a job desert or a gold rush? The answer, in my view, is neither of these extremes. But it will absolutely change the workplace landscape and we must work on understanding and planning for these changes.
During this webinar, we will be exploring the potential impact artificial intelligence will have on different jobs and on the workforce, in general. In addition, we will be introducing Aragon Research’s new AI Technology Arc.
- What jobs will be the most impacted by AI technologies?
- How will AI change the workforce?
- What technologies do leaders need to track over the next 3 years?