Protecting Creators in the Age of Generative AI
Advancements in artificial intelligence technology have taken the creative world by storm, enabling anyone at any skill level to use their own words to create content the way they dreamed it – from images to videos and 3D renderings. Creative professionals and new creators alike are already using generative AI in new and powerful ways. But this technology has also raised valid concerns about how to tell real from fake in an AI-generated world, the viability of generated content in commercial settings, and importantly, how creators can benefit from their skills and maintain credit and control over their work.
When it comes to copyright in the age of AI, there are questions both around the data being used to train an AI model, as well as the content AI models generate. To successfully address this issue, we need careful consideration and collaboration from the public and private sector to ensure that ultimately – whether a digital artist, a traditional painter, a singer-songwriter – it is the creator who wins.
Providing a pathway to digital content ownership
Creators need tools to show content ownership if they want to seek copyright in the digital world. With content going viral in a matter of minutes, it is often difficult to pinpoint where content originated online. In 2019, we anticipated this problem and sought to restore transparency to digital assets.
We co-founded the Content Authenticity Initiative (CAI) to enable artists to get credit for their work wherever it goes. The CAI’s Content Credentials enable creators to attach important attribution data to their content, like their name, date and what tools were used to create it. That information travels with the content wherever it goes so that by the time someone sees it, they know who created it, where it came from, and what happened to it along the way.
Content Credentials help creators prove that they were responsible for a given piece of content and gives them a mechanism to prove content ownership. Once Content Credentials are enabled, creators have more autonomy and traceable ownership of their work online.
Understanding data training with generative AI
As excitement builds around generative AI, questions around content ownership have only grown. Not only do people question where content originated, they also question how their content could be used if they share it online.
Some creators are happy to see their work used to advance this new technology. Others want their content out of AI training datasets. Our first model of Adobe Firefly, our image-generating AI service, is trained on Adobe Stock images (which we have license to train on per our terms of use), openly licensed content and public domain content where copyright has expired. That content is then carefully curated and preprocessed to mitigate against harmful or biased content. Because Firefly is trained on licensed images, we can also mitigate user concerns about copyright or brand infringement.
We’re also leveraging the Content Credentials technology to help our community prevent their work from being used in other generative AI companies’ datasets. Creators can attach “Do Not Train” credentials that travel with their content wherever it is used, published, or stored. With industry adoption, these credentials will prevent companies everywhere from using works with “Do Not Train” credentials as part of a dataset.
Generative AI output as a starting point in the creative process
The other question around copyright is whether the output of an AI model can earn a copyright. This requires examining existing copyright law in the context of AI.
Global copyright law says that you can’t copyright an idea. What you can copyright is the expression of an idea. So you can’t copyright the word “spaceship” for example. That is an idea. But you can copyright a painting of a spaceship because that’s the expression of the idea. When you apply this principle to image-generating AI, it means that a prompt would not be copyrightable because the prompt represents the idea. When you type in this prompt (or idea), you then receive an output based on the AI’s interpretation of that prompt. In other words, the AI is expressing the idea, not the human. Copyright law is designed to protect the rights of human creators – therefore an AI’s expression of an idea is not copyrightable.
But what about the creator who ideates and brainstorms in a generative AI tool and then goes on to add their own style and flare? Or the artist using AI to quickly change the sky color of their artwork, rather than a pixel-by-pixel manual task? Take our spaceship example. Maybe the artist starts with the AI-generated output but then uses other non-AI tools to change the spaceship’s color, draw more planets, and add in astronauts inside the spaceship. Or maybe they already created an image of a spaceship using non-AI tools and they used AI to simply swap out the sky for a more star-filled horizon It’s important to make sure as they leverage new technologies, these creators have ways to get ownership and credit for the human creativity and expression they bring to their work.
If a creator wants to use generative AI as a starting point and later register their work with the Copyright Office, they will need to disclaim what was generated by the AI. In the event the Copyright Office (or the courts) need to validate this information, Content Credentials give creators a way to show what they have done to an image after AI was used to demonstrate the level of human expression.
Knowing how a piece of digital content was created, whether with human direction, AI assistance, or AI generation, is more important than ever. The Copyright Office has an important leadership role to play in ensuring creators can get ownership and credit for their work and helping the public understand where that work came from and who has the rights to it. Adobe will continue to explore ways to empower creators in this era of new technology.
Join Us and Get Started with These Resources
Together we can deploy AI technologies responsibly and in a way that respects creators and our communities at large. We have grown the CAI coalition to more than 1,000 leading tech and media companies, camera manufacturers, news publishers, creative professionals, researchers, and NGOs working towards widespread implementation of standards that will bring more transparency to AI-generated and other content everywhere.
If you’re interested in joining Adobe’s work in this area, you can find out more information here and subscribe to the CAI’s newsletter to receive the latest updates on tools and ecosystem news. CAI members have access to collaboration opportunities with Adobe’s team on projects, quarterly events, including panels, and forums with the broader CAI community. Members can also utilize the CAI logo and brand assets to indicate their involvement. If you’re not ready to become a member yet, you can still use the CAI’s free open-source tools on your website, app, or service to help protect attirbution for creators and bring more transparency to digital content. Additionally, you can connect, collaborate, and give feedback with the CAI team and developer community on Discord
If you aren’t already a member of the Copyright Alliance, you can join today by completing our Individual Creator Members membership form! Members gain access to monthly newsletters, educational webinars, and so much more — all for free!