Legal Battle Against AI or Against Copyright Infringement
In recent years, we have faced many issues regarding the impact of artificial intelligence on intellectual property protection. What catches attention amid all the noise is the question: Is it permissible to use someone else’s copyrighted work to train AI?
In the United States, where this issue has become a ‘billion-dollar’ question, the discussion usually centres on ‘fair use,’ and AI companies that have so far been sued over such practices have largely relied on that type of permitted use.
In a response to a lawsuit filed by several firms owned by Disney and Universal against Midjourney, Inc., a provider of AI services capable of generating images and videos, Midjourney defends its unauthorised use of copyrighted work to train AI as “in the public interest” and “transformative use”. (“The limited monopoly granted by copyright must give way to fair use, which safeguards countervailing public interests in the free flow of ideas and information”.)
Following Disney and Universal, Warner Bros. Discovery also filed a complaint against Midjourney for copyright infringement. The complaint is very clear, stating essential aspects of copyright infringement that are not easy to ignore in this case:
- making and keeping illegal copies of the copyrighted works
- reproduction, public display, performance, and distribution of infringing images, videos and derivatives
- use of copyrighted works to promote services, encouraging even more infringement
- commercial nature of use
- failing to employ any technical safeguards that could curb infringing outputs
What makes the complaint even clearer is compared images – side-by-side images of original works and images obtained through Midjourney’s Service.
Warner Bros. Discovery showed that if a user prompts Midjourney’s Service to produce Superman, it will be a spitting image of DC’s famous character.
Even when a generic prompt is used, such as “classic comic book superhero battle” that does not mention a Warner Bros. Discovery copyrighted character, Midjourney generates and displays images featuring Superman, Batman, and Flash.
We still do not know what will be Midjourney’s response to Warner Bros. Discovery’s claims, however in its earlier response to Disney/Universal lawsuit it states that main question is: whether copyrights entitle their owner’s to prohibit the use of their works to train generative AI models, and to prevent the incorporation of elements of those works within the creations of generative AI users, regardless of the contents of those creations and the ultimate ends to which users put them.
It is very hard to see “training of AI” as substantially different from “unauthorised copying “, especially when such copying is on a massive scale of entire works and is used for obtaining commercial value.
It is however, even harder to look at the compared images and not think that “creations” in practice are actually “copies” and that there is nothing “essentially transformative” in outputs obtained from prompts such as “Superman, classic cartoon character, DC comics”.
Without a doubt, AI is rapidly changing our world, and we shall have to adapt and find new solutions for new problems; however, the question here is: is this really a new problem or a very old one. Is this AI training or simply copyright infringement?
