With so much talk in the mainstream media centered on whether some jobs are becoming obsolete thanks to artificial intelligence, creators may now have to worry whether they, too, might be replaced by an AI machine.

The latest in a string of AI machine inventions aimed at the creative industries is aptly called “Bot Dylan” – a researcher developed program that’s able to generate new folk songs by using over 23,000 pieces of Irish folk music.

This new computerized composer is the brainchild of Dr. Oded Ben-Tal, a senior lecturer in music technology at Kingston University, who decided to use Celtic folk music for its wide breadth of sound. Using transcriptions of 23,000 folk tunes written in a common music notation form called ABC, the machine was able to recognize and master the patterns of a song. So far, Ben-Tal and his team have managed to create over 100,000 new folk songs by taking one ABC symbol and suggesting the next symbol that should follow, creating an entirely new song.

This isn’t the first AI machine of its class to create wholly original creative works – some time ago, we discussed a similar program that was able to recreate a Rembrandt painting by analyzing 350 works by the artist. If an AI machine can create a painting that reflects the caliber of a legendary artist like Rembrandt, than one really has to wonder whether the future of creativity lies in an algorithm – and if we’re considering that, than there’s a whole host of questions that we need to ask ourselves about the future of ownership and copyright protection in creative works.

Can AI Machine-Generated Creations Get Copyright Protection?

Copyright protection in creative works by an AI machine presents an interesting dichotomy for copyright scholars because the question of whether an AI machine’s work product can technically be considered protectable under copyright law is rather vague.

For most works of art, copyright protection is affixed the moment that work is created. That means that the creator of the work enjoys the sole, exclusive right to determine what’s done with that work – distribution, making copies, publicly performing the work – that’s all forbidden without the express permission of the author. However, when it comes to AI machine creations – like folk songs or a painting – then copyright protection isn’t so cut and dried, since there are a few creators involved in bringing this work to life. In our ‘Bot Dylan’ example, would copyright protection be afforded to Dr. Ben-Tal? His team of researchers? What about the creators of the songs the AI machine studied in order to create its own songs? Can the machine itself enjoy copyright protection? Can it’s owner?

copyright protection

AI machines are increasingly making creative works.

The answer is quite muddled, and in order to fully understand who owns the copyright in AI machine-created work, we need to take a close look at the Copyright Act, and determine whether the resulting work is even eligible for copyright protection.

According to the Act, copyright protection is available for 1) an original work of authorship, 2) fixed in a tangible medium 3) that has a minimal amount of creativity. If a work doesn’t have all three of these components, then it is not copyrightable subject matter.

Let’s start with an analysis of the first prong: is an AI machine-created work sufficiently original enough to be copyrightable? Originality is grounded on the idea that the resulting product is new or novel, and not a reproduction. But couldn’t it be argued that an AI machine isn’t creating anything original, just picking up on sound modules to string them along? Not necessarily. Copying a style isn’t the same as copying a work – and while an AI machine is using composites of sounds or paintings to generate original content, the reality is that these works remain original because they’ve never been heard before.

In addition, the researchers admittedly could not predict what the AI machine was going to create once they had mastered the ABC system for folk music. In fact, that’s why the team decided to use Celtic music, since it had so many notes and chords – they essentially gave the AI machine an extremely large canvas with which to work with, and couldn’t predict which notes or symbols they would use.

As a result, the AI machine’s process was sufficiently original enough to satisfy the originality prong of our analysis, since each song is wholly unique.

In addition, the work satisfies the ‘fixed in tangible medium’ component of our analysis because the work has entered the public conscious – the work exists for anyone to hear. Once the Ai bot’s algorithm comes into existence, it is considered to be fixed in a tangible medium, just as the same would be true for a poem written on napkin or a drawing on a sheet of looseleaf paper.

Perhaps the most controversial question is whether or not the AI machine’s work is creative enough to enjoy copyright protection. As humans, it’s difficult to consider that a machine can be intelligent enough to engage in creative thought – and copyright lawmakers have considered the issue in the writing of the law.

503.03(a) Works-not originated by a human author.

In order to be entitled to copyright registration, a work must be the product of human authorship. Works produced by mechanical processes or random selection without any contribution by a human author are not registrable.

So what does that interpretation of copyright law tell us? For one, it unequivocally states that a computer or an animal shouldn’t and can’t hold a copyright, and it makes sense when you think about the function and purpose of copyright law in the first place. Since the holder of a copyright has the sole ability to give permission to others to use and distribute their work, then a computer inherently can’t hold a copyright since it’s unable to express their intention when determining whether or not to do so. Not only can a computer not consent to something like that, it also can’t enforce those rights and prevent anyone from doing so without permission. Similar to the cotroversy over the ‘monkey selfie,’ in which PETA sued for copyright infringement on behalf of a monkey who took a photo of itself, which later wound up being shared across the internet. The monkey selfie lawsuit was deemed rather frivoulous, and it reinforced the notion that animals can’t hold copyright protection in a work. Computers, like monkeys, can’t go to court, after all.

But Someone Has to Own the Copyright in These Songs, Right?

If we’ve determined that these songs – though created by a computer – are original, fixed in a tangible medium, and a creative product – then we have to assign the copyright to someone, right? Not necessarily.

There are some works that enter the public domain the moment they’re created. For example, works or images created by the government are automatically rendered public, non-copyrightable images. In order to determine whether Bot Dylan’s songs would be copyrightable, we have to determine whether the researchers that fed the AI machine the songs would hold copyright protection in the resulting new songs created by the AI machine. But is that a stretch?

ai machine

Unless a creator had an extensive role in the AI machine’s work, they can’t seek copyright protection in the work.

The Supreme Court outlined their ideas of what constitutes ownership of a creative work by a computer in a 1973 court case, Goldstein v. California. There, the Court interpreted the authorship requirement to include “any physical rendering of the fruits of creative intellectual or aesthetic labor.” The Supreme Court reasoned that in most cases, in order for a computer to generate any kind of artistic work, there must be significant input from an author or user, and only then can the mastermind behind the AI machine hold copyright protection in the work.

A year later, Congress created the National Commission on New Technological Uses of Copyrighted Works (CONTU), which studied a variety of new technology issues, including the issue of authorship of computer-generated works. Although CONTU did not expound on the topic extensively, it did conclude that authorship rights should go to the user when the user makes a very substantial contribution to the output. But when the user does very little and most of the output is left up to the AI machine, then it is less likely that the user may own the copyright in the output. 

A great example of an instance in which copyright protection would exist in a creator rather than an AI machine is Photoshop. If you think about it, Photoshop is an algorithm – there’s software behind a designer’s ability to shape an image using Photoshop tools. However, the designer’s input is necessary to the final product – they’re the one deciding what will go where, and directing the algorithm to deploy their creation.

Courts will generally put a lot of emphasis on whether the researcher had any substantial input beyond the simple creation of an algorithm or process for which the computer generates creative work. For example, the Rembrandt example would be one such project in which the software developers could likely exert copyright protection over the artistic work, since there was likely a lot of creative thought behind the project. But could the same be true for the makers of Bot Dylan?

It’s hard to say. You see, the researchers that created the Bot Dylan AI machine didn’t exactly reinvent the wheel – they didn’t create a new algorithm, but rather input over 23,000 song transcripts written in an age-old music notation form in order to ‘teach’ the bot how to play the music.

In fact, Dr. Ben-Tal’s team isn’t the first to utilize ABC music notations to teach computers how to write music. So, even if you could make the argument that the practice for making the music itself is copyright protected, would his team be able to hold copyright protection in that process? It seems unlikely.

Does the copyright have to belong to somebody? If we’re looking to the copyright law outlined above, the answer is no. So, if Dr. Ben-Tal and his team want to exert any type of copyright protection over Bot Dylan’s songs, they’ll have to prove that their input into the songs’ creation was sufficiently creative enough to warrant that type of copyright protection – and if they can’t, these new songs officially belong in the public domain.

Should an AI machine hold copyright protection over a work? Let us know in the comments!

Nicole Martinez
Nicole Martinez

Nicole is a veteran arts and culture journalist. Her work has appeared in Reuters, VICE, Hyperallergic, Univision, and more.