The Impact of Generative Artificial Intelligence on Copyrights

Generative Artificial Intelligence (“Generative AI”) programs, such Open AI’s DALL-E and ChatGPT, can generate outputs in the form of new images, texts, etc., in response to a user’s inputs. Outputs are accomplished by exposing the input to vast quantities of existing works, such as writings, photos, paintings, etc., to blend the original input with external data to form a new output. Since their emergence, they have presented several novel copyright questions involving issues such as copyrightability, ownership, and infringement. Unfortunately, the courts and Copyright Office are beginning to deal with these questions.

Regarding whether such works are copyrightable, the question focuses on the issue of “authorship,” i.e., who is the author of such Generative AI work? It has long been recognized that a human being must create a copyrighted work. Unfortunately, Generative AI works are not solely created by humans but, instead, by a combination of human input augmented by a computer’s exposure to a vast amount of other material.

Initial decisions involving Generative AI works have held that, without a human author, the work is not copyrightable. That’s changing. Last year, the Copyright Office issued a Registration Guidance that indicated that works “containing” AI-generated material may be copyrighted under some circumstances. The author may only claim a copyright “for their contributions” to such works provided they must identify and disclaim AI-generated parts of the work.

A second question is: who owns the copyright to Generative AI works? Again, the answer turns on the authorship of such works, which, as explained above, is in flux. While these issues are being sorted out, many of those involved in the industry have turned to contract law to answer this question. Frequently, the underlying contracts between the parties involved in creating a Generative AI work will precisely specify who the owner will be, thus bypassing copyright law.

The third question involves the question of copyright infringement, including whether a Generative AI work infringes the copyrights in those works used to “train” the AI. Generative AI systems are “trained” by exposing the input work to massive libraries, much of which is copyrighted. Can a claim be made that the resultant AI-generated work infringes the copyrights in the training data used to create it? Since one element of a claim for copyright infringement is that the author has access to the copyrighted work, their use for training would potentially meet such a requirement when the new work was “substantially similar” to the underlying work. To avoid this problem, however, most companies involved in developing a Generative AI work seek and obtain licenses from the copyright owners of the training works.

The final question presented is when a Generative AI work is found to infringe on a third party’s copyright, who is the infringer? Potentially, both the AI user and the AI company could be liable.

As noted above, Generative AI raises more questions than existing Copyright laws can provide answers to. As such, they will be the subject of contentious litigation over the next few years unless and until existing law is amended to address them or the higher courts weigh in. In the meantime, stay tuned.