By: Andrew Winegar
A new painting has been added to the Rembrandt collection in Paris’s Musee Jacquemart-Andre. It depicts what seems to be a portrait of a 17th century European man, complete with a frilled collar, a large black hat, and the eerily human, nearly-living eyes that set Rembrandt apart from his peers. However, this latest addition is not a recently-discovered 400-year-old artifact created by the legendary painter himself, instead, it is the first work of a very new type of artist—an Artificial Intelligence program.
Artificially Intelligent Creators
“The Next Rembrandt” is the name given to the painting that now sits along with its namesake painter’s most famous works in Paris, and it was created entirely by an AI program. The Next Rembrandt: ING is a Dutch multinational banking group that partnered with Microsoft, The Rembrandt House Museum, and others to create an algorithm that analyzed over 168,000 fragments from over 350 of Rembrandt’s works and, in part thanks to facial recognition technology, created a completely new and unique work of art that emulates Rembrandt’s style and technique yet remains a wholly new and original piece of art.
A computer program that learned how to paint like Rembrandt is not the only AI disrupter in the world of the arts. Last year, a Japanese-developed AI program wrote a novel that nearly won the nation’s Nikkei Hoshi Shinichi Literary Award. Similarly, Google now has a program that learns how to listen to, analyze, and then compose music all on its own. While these accomplishments are seen as nothing short of fascinating from a scientific perspective, they also have interesting implications from a legal perspective. Specifically, when a program is able to create a painting, write a novel, or compose music, how does the law recognize the copyrights of these works?
The issue involves the creative steps that are needed when creating a work. In the past, most works of art created with the help of computers clearly had copyrights belonging to the programmers who told the computer what to do; all actions were specifically programmed and attributable to human activity. However, today’s programs are informed only by vague parameters set by programmers, and using systems called “artificial neural networks” (“neural nets” for short) to learn and create on their own. Programs are now capable of taking creative steps that in the past have only been attributable to human activity, and the law regarding copyright may not yet have caught up to the these advances in technology.
The Bot Dylan Effect
When an advanced AI creates a work, there are many potential authors implicated based upon how the work came to be. For example, Dr. Obed Ben-Tal of Kingston University has created a program called “Bot Dylan” that is able to analyze Irish folk music and create unique music based upon its analysis. Dr. Ben-Tal has a team of researchers that helped him create the algorithm. When Bot Dylan composes a folk song, there is not a clear claim as to whether the copyright for that song is Dr. Ben-Tal’s, a joint authorship with his entire research team, or something entirely different as the AI itself took all the creative steps.
The US Copyright Code offers protection for a work that is original, fixed in a tangible medium, and has at least a minimal amount of creativity. Bot Dylan’s works are completely original, and Bot produces tangible songs that can be listened to or played by anyone else that comes along after the algorithm creates the songs. However, the question hangs on whether Bot Dylan’s algorithmic approach to songwriting is truly creative; and perhaps even more significantly, if Dr. Ben-Tal and his team can claim they asserted any creativity in a work they really had nothing to do with. While the research team created Bot Dylan, and taught Bot Dylan how to understand Irish Folk songs, Bot Dylan did all the creating after the initial programming. With the creativity issue in flux, it is hard to tell how courts would treat the music composed by Bot, but it is highly likely that all of its songs would end up straight in the public domain due to their inability to be considered copyrightable works.
While the lack of copyright protection may be fine for Dr. Ben-Tal and his team (whose vision in creating Bot Dylan did not end with them touring the world as AI-inspired Irish Folk artists) many businesses could very well use AI to create content that would benefit from copyright protection. Imagine instead that Dr. Ben-Tal and his team were not academics, but instead created Bot Dylan for a corporate purpose. Perhaps the Celtic songs that Bot creates would make great ambient noises in a video game that the company produces. If the works of the AI are not copyrightable, then any music created by Bot Dylan could be reproduced by any competing company that wanted the same sounds in their game, creating the very real possibility of a negative economic impact suffered by the original company using Bot Dylan’s composition.
Domestic and Foreign Perspectives
After the infamous, “Monkey Selfie” case (in which PETA attempted to assert copyright protection for a photograph taken by a macaque), the Compendium of US Copyright Office Practices was updated to reflect that works created by non-human authors are not copyrightable (specifically mentioning “a photograph taken by a monkey” as one of their examples of a non-copyrightable work). In addition, the Compendium notes that “similarly, the Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author.” However, the examples included to reflect this wording make no mention of artificial intelligence and the human influences that lead to programs creating works, leaving a very real gray area in the law when it comes to assessing the copyrightable nature of AI-generated works. Making matters even more complicated is the fact that the US Copyright Act itself does not have a human authorship requirement, primarily due to the fact that the courts have always just assumed that authorship is a “human phenomenon.”
The European Union is perhaps the first major legislative body to spend extensive time evaluating the copyrightable nature of AI-generated works, and has classified such works into four categories: AI co-production, human selection of AI generated works, AI “brute force” works, and AI independently generated and selected works. In co-production cases, the EU has easily found that the human involved in co-production has a copyright in works created with the assistance of AI. In these cases, the use of AI has been likened to the use of a word processor or photo editing program; that is, AI is simply being used as a tool to express human originality. Similarly, the EU believes that works created by AI, but selected by humans, meet the originality bar necessary for copyright protection. Their reasoning lies in the uniqueness of human discernment; a program may be able to create music, but the same program could just as easily create screeching, unpalatable tones and requires a human influence to determine which works are pleasing to the ear and which are not.
The remaining two categories seem to fail to meet the standards for copyright protection. The “brute force” approach involves AI creating every possible combination of variables that it can (most commonly in written works) as eventually all possible combination of letters will be present to create a work. Since brute force algorithms just produce works using every available combination of factors it can, there can be no intentionality attributed to the human programmer; instead, brute force algorithms simply aim to “put together every combination of words without any consideration of which combinations of letters make up good texts.” For this reason, there is no creativity required and therefore no originality recognized by EU copyright law.
Lastly, there is the issue of AI works that are entirely generated and subsequently selected without any human involvement. In these cases, the AI generates the works as well as selects works which it has created which fit the parameters that the program is assigned, all on its own. It can be argued that if the subsequent work is indistinguishable from a human created work, then it “should have the same legal status as one created by a human, regardless of the method of production.” However, the EU concept of originality is actually influenced more heavily by the creative process than the final product. In essence, the act of “intellectual creation” relies heavily on the author’s expressed creativity involved in the process of creating the work more than the product created itself. Following this logic, when an AI creates a work, it is considered to only be running a program and not making creative choices. The final product may seem creative, and may even be indistinguishable from a man-made work; but, because the process did not involve any creative expression, there is no inherent moral right in the process of creation that a copyright exists to protect. Furthermore, because the AI is selecting its own works, the human expression is even further limited. Therefore, as of now the EU’s reasoning would seem to trend away from granting a copyright to works created in this nature.
As AI programs become more and more integrated into society the issues surrounding them will only become more complex. These programs are now producers in their own right, and soon it is entirely possible that mainstream works will be produced with the help of, if not entirely by, AI programs. Eventually the Courts must address the economic consequences and advantages of failing to grant copyrights to AI generated works. Without this important legal protection, some influential content creators may be discouraged from creating knowing that their AI-produced product will not be granted the protection they need in an open market.
However, granting AI-produced works copyright protections would also come along with its own unique issues, such as determining how much ownership a human programmer should be given over independently created works when the programmer’s primary influence was simply setting parameters. If copyrights were given freely to works created by AI, then one programmer or a single programming team could easily amass a huge number of copyrights in a relatively short amount of time, which would seemingly devalue the creative work done by traditional copyright holders. As these issues evolve over time, we will have to wait and see how the courts react to a growing number of non-human content creators. The Next Rembrandt is already here, but the effects of granting this new kind of artist a copyright remain to be seen.
 The Next Rembrandt/ING, J. Walter Thompson Amsterdam, https://www.jwt.com/en/work/thenextrembrandt (accessed November 10, 2017).
 See generally, ING Presents the Next Rembrandt, www.thenextrembrandt.com (accessed November 8, 2017).
 Tim Nudd, Inside ‘The Next Rembrandt’: How JWT Got a Computer to Paint Like the Old Master, AdWeek (June 27, 2016), http://www.adweek.com/brand-marketing/inside-next-rembrandt-how-jwt-got-computer-paint-old-master-172257.
 Chloe Olewitz, A Japanese AI Program Just Wrote a Short Novel, and it Almost Won a Literary Prize, Digital Trends (March 23, 2016), https://www.digitaltrends.com/cool-tech/japanese-ai-writes-novel-passes-first-round-nationanl-literary-prize.
 Devin Coldewey, Google’s WaveNet Uses Neural Nets to Generate Eerily Convincing Speech and Music, Tech Crunch (Sept. 9, 2016), https://techcrunch.com/2016/09/09/googles-wavenet-uses-neural-nets-to-generate-eerily-convincing-speech-and-music.
 Andres Guadamuz, Artificial Intelligence and Copyright, WIPO Magazine (Oct. 2017), http://www.wipo.int/wipo_magazine/en/2017/05/article_0003.html.
 Nicole Martinez, Can an AI Machine Hold Copyright Protection Over its Work?, Art Law Journal (June 1, 2017), https://artlawjournal.com/ai-machine-copyright.
 See generally, 17 USC § 102.
 Martinez, supra note 10.
 Guadamuz, supra note 7.
 Kalin Hristov, Artificial Intelligence and the Copyright Dilemma, IDEA: The IP Law Review, Vol. 57, No. 3, 2017 (Sept. 1, 2016), https://law.unh.edu/sites/default/files/media/hristov_formatted.pdf.
 Copyrightable Authorship: What can be Registered, US Copyright Office, Compendium of US Copyright Practices § 313.2 (3rd ed. 2014).
 Hristov, supra note 19.
 Robert Hart, If an AI Creates a Work of Art, Who Owns the Rights to it?, Quartz Media LLC (Aug. 15, 2017), https://qz.com/1054039/google-deepdream-art-if-an-ai-creates-a-work-of-art-who-owns-the-rights-to-it/.
 EU Copyright Protection of Works Created by Artificial Intelligence Systems, University of Bergen (June 1, 2017), http://bora.uib.no/bitstream/handle/1956/16479/JUS399_V17_183.pdf?sequence=1.
 Id at 20.
 Id at 21.
 Copyright Services, University of Minnesota Libraries (last accessed Nov. 16, 2017), https://www.lib.umn.edu/copyright/purpose.