Additionally, if copyright were to arise, who owns it? A computer cannot own property. Many different people are involved in creating an AI work—those who write the program, collect the data, train the algorithm, use the program. Under these circumstances the argument is: it is impossible to decide who should own any copyright in the output created in such a complex way, and that those involved are not “creative".

These are Chicken Little arguments that ignore history.

Just like the use of a camera, an AI machine can be used to create both great and mediocre works. Human artistic skill is needed to use AI to create meaningful AI generated art. Why should this artistic skill not be recognised by copyright law?

How should we think about creativity in the age of AI? As we learn more about creativity and the use of AI, we can then determine what roles in the generative process can be understood as creative, which require incentives or deserve protection, and thus who should own the copyright. 

U.S. Artist Jason Allen, who won first prize in the Colorado State Fair digital art competition with his work Théâtre D’opéra Spatial (2022) is appealing the U.S. Copyright Office’s rejection of his request for copyright protection. The Copyright Office said his piece lacked ‘human authorship.’ Allen argues his work, that he spent many hours creating, uses AI as a legitimate tool of artistic expression.

After this rejection, the U.S. Copyright Office then released new guidelines in March 2023 allowing AI generated works to receive a copyright registration in certain circumstances, considered on a case-by-case basis. There must be a human author, the AI must only be 'an assisting instrument', and the artist or author must provide original mental conception to which the artist or author gave visible form.⁶ The guidelines are vague and attempt to apply old principles to new circumstances. According to the Copyright Office, 'many stakeholders still have questions about these guidelines for works containing AI-generated material and would like more details and more examples of how the Office will approach applications for such works.'⁷

Like AI art, many people are involved in creating a motion picture—the screenwriter, the cinematographer, the actors, the director, the set designer, the composer of the musical score, the financier—and motion pictures can include computer generated special effects. Despite this complexity, Australian copyright law says that copyright in a motion picture is owned by the producer unless there is an agreement to the contrary.

For over 100 years, intellectual property law has been able to determine who is making the intellectual contribution in respect of complex works. For example, patent law has sophisticated rules to determine inventorship. If I design a fast rocketship and sketch out the plan, but you build it with your welding equipment, generally I will be recognised as the inventor. When Christo and Jeanne-Claude surround islands in pink plastic in Surrounded Islands (1983), the copyright is not owned by the teams of people rolling out the plastic but by Christo and Jeanne-Claude who created the design.

The debate should focus on who is creating or adding artistic value, and thus who should own the copyright to AI generated art. Who is the ‘creative producer’ of AI generated art? Let’s not focus on the minutiae of what the law is today, but what the law should be tomorrow, consistent with long standing intellectual property principles.

It is important to start thinking more deeply about the legal implications of AI generated art, so that we can ensure that artists are fairly compensated for their work and that the public has access to this new form of art.

The flip side of ownership is infringement. When training an AI machine, existing copyright protected works are used as input. There are cases before the courts that will soon consider whether the process of training an AI machine using the copyright works of others or the output of such trained AI machine is copyright infringement.⁸

Even without the AI overlay, copyright infringement of artistic works is a hard topic. Copyright protects the expression of ideas but not the idea itself. This idea/express dichotomy is at the heart of copyright to provide the balance between protection and progress. It is often difficult to apply.⁹

It is not copyright infringement to paint a work in the style of Mark Rothko or paint in the abstract expressionism style. It is likely copyright infringement to copy a Rothko’s painting and then modify it. If all of Rothko’s painting are fed into an AI machine for training, and then the AI machine is asked to produce a Rothko-like work, this feels like infringement. However, if all the abstract expressionism paintings in the world are input, and the AI machine is asked to produce an abstract expressionism-like work, this feels more like research or the copying of an idea.

From a strict Australian copyright position, even copying one work as input into an AI machine is likely infringement of that work. This makes the legal training of an AI machine somewhat problematic, unless one only wants to train the machine using old masters where the copyright has expired.

This is where Silicon Valley has changed the rules. Google copied the whole Internet, distributed copyright videos on YouTube, and scanned millions of books in libraries. It just did it. At the time, I took the view that Google was the biggest copyright infringer in the world’s history, and got away with it by dealing with the legal consequences after the fact.¹⁰ The AI industry, based in Silicon Valley, is taking the same approach, and is asserting that training AI using copyright works of others is fair use.¹¹ Unless you own or have licensed a large portfolio of works, the only practical way to train AI is to copy existing works. However, this time, copyright owners are on the front foot bringing lawsuits early in the commercialisation process to make sure they get a piece of the pie.  

Lawmakers will have to consider whether the artists, whose works are part of the massive data sets computers rely on to generate their results, should be credited and compensated. 

I predict that the Silicon Valley AI tycoons will negotiate with copyright owners and collecting societies to pay royalties to use existing works to train AI, unless the tycoons successfully lobby the government to amend copyright law to allow the training of AI without a license. Artists should make sure to be pro-active in this process. Hollywood writers are currently striking, concerned about AI reducing their income, Lawyers are also worried too, for similar reasons.

AI is also enabling new forms of expression. An example is deep fake videos, often in the form of comedies or mockumentaries. A video work can be created that makes it appear that Albanese and Trump are doing the most ridiculous things, and that looks real. This could be art, or political commentary, or both. Or it could be used for improper purposes.

Research is also taking place to tap in and mirror our dreams using AI. By measuring brainwaves when we sleep, AI appears capable in the not-too-distant future of producing a video of our dreams.¹² If the mirror is a tool for self-discovery helping us to learn more about ourselves and our place in the world, then in the age of AI, this new ‘mirror’ may provide the most intriguing view of the human soul.

In short, AI will be used to generate new forms of expression that challenge our understanding of what is real, what it means to be human, and the resulting issues of legal responsibility.

Just as the camera resulted in the development of privacy law in the 1890s, AI may result in the creation of new fields of law to deal with responsibility, truth and ethics in AI. The balance between creative expression and responsibility will be tested.