A computer creates a generative art image of Botticelli’s Venus.

AI-Generated Art Pushes the Boundaries of Intellectual Property Rights

April 26, 2023 | Purdue Global Law School

Selfie portraits generated by the Prism Labs Lensa AI app have become all the rage on social media platforms. Users simply upload 10 photos into the app’s “magic avatars” feature and the Stable Diffusion neural network generates avatars in the user’s selected art style.

Lensa AI held the No. 1 spot in the App Store’s Top Charts during the first half of December with 12.6 million global installs. Two other artificial intelligence-generated art apps rounded out the Top 3. AI-generated selfie portraits are just the tip of the iceberg. OpenAI’s DALL-E, which is trained using the GPT-3 neural network learning model, allows users to generate detailed images in various styles through a series of prompts. Also, Adobe announced on Dec. 5, 2022, that it would include AI-generated art in its stock image library.

Artists aren’t as enthusiastic about AI-generated images. They say that companies are using their work without permission to “train” neural networks, and some of the resulting images resemble their work in noticeable ways.

AI-generated art may be the new frontier, but it poses several legal questions. Do fair-use standards apply to AI-generated art that includes copyright- or trademark-protected materials? What if the software copies the distinctive style of artwork used to train the algorithm?

How Neural Networks Work

Neural networks mimic the way neurons in the human brain process information, enabling them to identify subtle patterns and correlations. In the brain, each neuron receives information from another neuron, processes it, and passes it along to yet another neuron. Neural networks simulate this process in “layers.”

The input layer receives the raw data, processes it, and sends it to the next layer, and the procedure continues until the data reaches the output layer. Each layer contains a database with all of the algorithms, interpretations of the algorithms, and what the network has “learned.”

Neural networks require large volumes of input data to accurately classify and perform tasks like a human. The Large-scale Artificial Intelligence Open Network (LAION) that powers Stable Diffusion has the URLs of 5.85 billion images posted on the Internet and their associated alt text. It also includes the largest Contrastive Language Image Pre-training (CLIP) model, and a subset of images chosen by a model trained to identify those that are aesthetically pleasing.

Is Neural Network Training ‘Fair Use’?

In its FAQs, LAION states that it does not release any dataset containing images — in fact, the organization discards all the images it uses after processing them. But what if the ingested images are protected by copyright? Some experts contend that it constitutes “fair use,” a legal doctrine codified in Section 107 of the Copyright Act that permits the unlicensed use of copyrighted works in some circumstances. The use of a small amount of a work for education, research, commentary or criticism can be fair use, as well as “transformative” uses that produce a work of a different purpose or character.

Arguably, AI-generated art is transformative. However, some artists say they’ve seen AI-generated art that has the characteristics of their work or that of other artists they know. This has generated intense debate within the artist community. Some artists say work posted online is meant to be saved and shared, while others contend that doesn’t mean they consent to use by a for-profit company.

Applying the Four-Part Test

The level of transformation also depends upon the algorithm. Some apps transform images so radically that the input work is unrecognizable, while others generate output that’s obviously derived from the materials used to train them. Courts would have to consider each use on a case-by-case basis, applying the four-part test specified in the Copyright Act:

  • Is the use not-for-profit or commercial?

  • Is the copyrighted work largely factual or more imaginative?

  • How much of the copyrighted work was used, and to what extent?

  • What impact will the use have on the value or marketability of the copyrighted work?

The last factor is arguably the most important to artists. If people can buy an AI-generated image with the characteristics of an artist’s work for a few dollars, would they be willing to pay much more for an original?

Another question is whether intellectual property laws were violated when the images were collected. For example, some works may be covered by text and data mining (TDM) licenses that specify certain types of use and require the original creator to be credited.

Respect the Artist

TDM licenses give artists a measure of control over use by AI applications. In a recent position paper, the Copyright Alliance advocates for the use of TDM licenses to provide a body of work for AI use while allowing individual artists to decide whether to license their work. This would allow artists to obtain market value for their work rather than subsidizing AI technology.

The Concept Art Association takes a somewhat different stance, arguing that intellectual property and labor laws must be updated to better protect artists. The group also contends that AI technology companies should adhere to ethical rules when obtaining artistic works for neural network training, and use only works that are properly licensed or in the public domain.

Courts have yet to address whether AI-generated art infringes on the source material’s copyright. Meanwhile, AI-generated art continues to proliferate based upon an assumption that it falls under the fair-use exception to copyright law. Artists will need new avenues to protect their creative output while allowing for continued advancements in AI technology.

Stay Up to Date on New Issues in Law

Legal issues are always at the forefront of news. It’s important to stay current with legal developments in California and the rest of the nation with Purdue Global Law School.

Purdue Global Law School offers an online Juris Doctor if you wish to become an attorney licensed in California. If you wish to advance their legal education but do not intend to become a practicing attorney, you may consider an online Executive Juris Doctor.

Single law courses are also available to help you explore a particular area of law without committing to a full degree program. Request more information today.

About The Author

Purdue Global Law School

Established in 1998, Purdue Global Law School (formerly Concord Law School) is Purdue University's fully online law school for working adults.