Using AI to Create “Creative” Works

I have previously written on this blog about the intersection of entrepreneurship and the creative arts, in particular the copyright considerations that businesses must consider when creating or re-using media. The skyrocketing interest over the past year in artificial intelligence-driven content (aspects of which were covered in recent posts by Alex and Brandi) has led to some exciting but also concerning developments in how we think about creating and protecting media.

 It is now possible to ask a machine learning-powered tool to create an essay or a painting and to receive somewhat convincing results. On the one hand, these developments have been criticized as devaluing the contributions of legitimate human artists. On the other, proponents have argued that these AI tools provide creative services that some people or businesses could not otherwise afford or access.

 A couple of weeks ago, the U.S. Copyright Office stepped in on this question, publishing guidance that emphasizes the requirement that a work be authored by a human to be copyright-protected. This requires applicants for copyright registrations to identify any AI-created parts of their works. (Several of my intellectual property colleagues authored a more detailed summary of this guidance, which can be found here.) However, even the Copyright Office seems to be unsure where it stands, suggesting in its initial guidance that it could be possible for a human to have sufficient input into or control over an AI creation so as to lead to “human authorship.” For example, a human could revise and rearrange the images output by an AI tool in a creative enough way for the final, modified group of images to be copyright-protected.

 This ambiguity is not just a matter of Copyright Office formalities. Whether or not a work is protected by copyright is vital to whether it can be “owned” by any one person and therefore whether that one person can prevent others from using it.

 Say, for example, an entrepreneur uses AI to create an original illustration to be sold on T-shirts and other merchandise. If that illustration is not the product of “human authorship,” does the entrepreneur have grounds to prevent others from taking that illustration and using it on their own, competing merchandise? Or what if a business uses AI to write its advertising materials? Will they have any recourse against a competitor who copies that text wholesale?

 Moreover, where did the image or text created by the AI really come from? These machine learning tools must be trained on a database of examples. Recent research has shown that, rather than creating a unique new work, some AI art generators will sometimes provide just slightly modified versions of images already in their databases. The topic of AI has thus proven controversial among human artists, some of whom argue that these AI tools are “copying” their styles or even specific images.

 The flaws in AI tools that lead them to reproduce existing works, or make only minor changes to such works, also raise another important copyright question: can AI be relied upon to create content that does not infringe the copyright of a third party? By feeding a prompt into an algorithm and using the resulting image or text without question, could a person face liability for copyright infringement?

 This space is developing so rapidly that opinions a year from today will likely be quite different from those today. There is certainly some merit to the argument that these tools, while imperfect, can be a good starting point for emerging businesses and entrepreneurs who might lack the resources, for example, to hire a copywriting staff or a team of graphic designers. But this technology is far from perfect and there remains quite a bit of uncertainty around the interaction of these AI tools with copyright law. Will AI tools become incorporated into our business and creative processes? Will considerations such as the copyright questions being raised today mean that future adoption of AI is limited? Or will the increasing prevalence of AI cause us to place even more of a premium on works created by humans?