Dall-E, Beatoven, Midjourney – these pun-filled, poetic-sounding applications, along with perhaps the most popular ChatGPT, have suddenly become well-known over the past year or so. Generative Artificial Intelligence (GAI) has captured the imagination of individuals and businesses alike, enabling (artificial) creativity on an unprecedented scale. Text, images, music, video, 3D printing, etc., these applications can produce pretty impressive output (although it’s still far from perfect).
GAI is all about creativity, and the first obvious question GAI asks is about intellectual property. On the other hand, there are allegations that GAI’s training procedures infringe existing copyrighted material. GAIs are typically trained on vast amounts of existing data on the internet. This could include articles, news pages, image websites (and, by some reports, entire e-books), all of which may be copyrighted works. A number of lawsuits have already been filed in the United States alleging copyright infringement in the training process by GAI developers. The training process does not necessarily require consent or licenses from all authors, and as a result authors’ autonomy over their work is a concern. GAI’s training includes making and storing copies of pre-existing copyrighted works, but whether this constitutes infringement depends on the fair use doctrine, GAI’s commercialization, and pre-existing may depend on factors such as the degree of infringement of the work, the creativity involved in the output, and other factors. And a threat to the market for original works.
The other aspect in terms of IP is, of course, who is the author of the GAI output and whether the GAI output qualifies as copyrighted work in the first place. The concept of authorship has traditionally revolved around individual authors, their “sweat of their brows,” and their creativity (although ownership may also reside with corporations or other legal entities). Even the duration of copyright is related to the author’s life. Because of this precedent, it is difficult to ascertain whether GAI developers and/or users who provide input should be treated as authors of the work. Because none of them put their creativity or skill into any particular output. AI may not be treated as a legal entity, especially from a copyright perspective, so he may not consider GAI an author. This fragmentation, given the lack of authorship, can lead to the argument that GAI’s work is not intellectual property in the first place.
In addition to the above, the existence of bias and prejudice in AI is well documented. AI has been reported to exhibit racial and ethnic bias in addition to gender bias. Bias like this can have a huge impact, especially on consumer AI. For example, a text-based AI might assume certain traits based on a person’s race, while an image-generating AI might produce output that assumes a certain gender role. In addition to regulatory risk in jurisdictions with strict anti-discrimination laws, such bias could lead to significant reputational risk if a company uses his GAI for customer-facing activities.
This leads to broader issues of accountability, accountability and responsibility. Bias is one source of potential risk, but illegal content can also create liability. For example, if GAI creates hate speech or defamatory content, should the GAI developer, the company deploying GAI, or the user providing the prompt be held responsible? The answer is more complicated when you train the GAI using your own data to respond. GAI training is not always intentional, and understanding or predicting how the GAI will produce the final output is difficult, if not impossible. This “black box” conundrum leads to accountability issues, especially when his GAI is deployed in public use cases. How a GAI developer and her customers assign responsibility for her GAI output is an important point of legal consideration. These legal issues are therefore key to ascertaining both the value of GAI’s business and its services, and the associated risks.
Written by Huzefa Tavawala and Aniruddha Majumudar
