Behind the scenes of human-machine co-creation

Thanks to the remarkable development of machine learning technology, machines can now “create”. They now have content ownership in their language, can make (relative) decisions, and adapt to their environment to generate new information.

It is often said that such semi-autonomous computational systems can be disastrous and alarming. Some initiatives, such as UNESCO and Myra, warn against potential risks and remind us of relevant issues of ethics, equity and social relevance.

In the art world, generative AI creative tools (ChatGPT, Midjourney, Blender, etc.) fall into a gray area when used in a dizzying array of often copyrighted works. It often causes intellectual property issues. Faced with potential threats to the perception of the artist’s work, the environment ponders and reacts to this issue. WIPO voiced its doubts at an online exhibition on intellectual property and artificial intelligence. The voices of the European Writers’ Council and others are being reflected. In the United States, a class action lawsuit was filed against the creators of the Stable Diffusion tool for copyright infringement. Artists are suing other artists for harming them by incorporating machine learning into their work.

Getting to the heart of the controversy, legal news has shed light on several issues, such as the recent decision by the U.S. Copyright Office regarding machine-generated works. With some nuances, it was determined that images created mechanically using generative AI tools are not the product of human creativity and therefore do not fall under the rights of the author.

Canada, meanwhile, cites certain status quo and reports that more data are needed to properly assess this new phenomenon, as demonstrated in the consultations.

Several legal and research experts have generously answered our questions to better understand the challenges associated with this new technology facing the concept of copyright.

A vast and complex task

The legal issues associated with new machine learning-based artistic practices are certainly complex due to their systemic and political scope.

Husken intellectual property attorney Eliane Elbogen explains that author rights are meant to protect and promote the mind’s creations. This is an exquisite compromise between fair remuneration for artists and accessibility to creative works in the name of competitiveness in the global market in terms of research, public interest and innovation. It embodies balance. A final judicial decision will affect the balance that sustains the economic interests of artists, research and the industry.

One thing is certain, as a leader in machine learning innovation, the industry is in a strong position to ensure that its interests as owners and users of author rights are valued.

Legal Objection 1 – Text mining, a technique unique to machine learning, involves copying data, which can violate the rights of the creator. Should it be considered fair use?

According to the Canadian Intellectual Property Office, it is the copyright owner’s “sole right” to “make or reproduce in any form the work, or any material part thereof,” unless this action is justified under fair use. is. Copyright protection.

Our experts agree that text and data mining is a violation under the current provisions of Canadian copyright law when creating databases based on protected works without the necessary permission for use. doing.

Other jurisdictions such as Japan, the United Kingdom, France, Germany and the European Union anticipated special exceptions to text and data mining. In Canada, the issue is being evaluated to consider the impact of judicial rulings on artists, research and the industry.

“The question of whether this use falls under the doctrine of fair use has not yet been brought to court. We are taking action, which removes the risks inherent in this activity.”- Elian Elbogen

Without such exceptions, industry would have to obtain permission to reproduce copyrighted material for machine learning purposes, as long as the copyrighted material is not in the public domain. And, as Robic partner attorney and trademark attorney Dr. Caroline Jonart pointed out, using out-of-date data can be problematic, as “by deriving models based on content that belongs to the public domain, it’s almost impossible to It is not updated and perpetuates prejudices and prejudices from the past.”

Legal Objection 2 – As a style modeler, the generative AI tool is suitable for your artwork. What if the machine-generated work is too similar to the work that inspired it?

Tools that specialize in pattern extraction, such as The Next Rembrandt, Stable diffusion, and Dall-E, allow you to simulate an author’s style and artistic tendencies.

If artists find their work within the space of these generative AI tools without the necessary permits, the question becomes whether these machine-generated works of art are subject to litigation.

On the one hand, neither the style nor the idea is protected by the rights of the author.

The other, as demonstrated in a paper by Tom LeBlanc, an attorney specializing in digital and copyright law, is that “substantially the talent and judgment of early works of art to be copyrighted.” It is a work that is realistically reproduced. Works that are not in the public domain and made without permission are subject to litigation if they do not qualify for fair use. Furthermore, Caroline Jonnaert points out that using one distinctive element, such as her three-second song, may be enough to qualify as plagiarism.

There is a fine line between the legal act of copying a style that belongs to the public domain and what counts as copyright infringement. Therefore, Canadian jurisdictions have not yet clarified the status of these “stylized” works of art.

Some commercial practices recognize this and aim to prevent overly similar duplication of images, as Open AI did with Dall-E.

Also, the stylized images in these spaces have a special character. They reflect how algorithms build abstract representations of data behavior.Sophiane Audrey, research artist and professor at the University of Quebec-Montreal and author of this book Art in the age of machine learningI will explain. “These artists’ images embody representations distributed in space that allow the generation of not just their images, but other images that have never existed before. It represents a kind of data compression of 2 billion images, not stored directly as a digital file, but represented in a network of artificial neurons, using far less memory space, as well as an infinite amount of new images. .”

How is this new trend different from the past?

The creative act of transforming existing materials is not new. Sofian Audrey emphasizes that the advent of image and soundtrack reproduction techniques in the 20th century encouraged new artistic practices based on various creations such as collage and Dadaism.

However, machine learning technology is revolutionizing remix technology. “Previously, we were remixing content that already existed, but now we are remixing the generation process using algorithms that convey style,” explains Sophiann Audrey.

Never before seen in human history, machines are now capable of automating the creation process with models that can transform themselves, modify and mix. This new capability presents both threats and opportunities, such as the alarming deepfake phenomenon.

As Caroline Jonnaert observes, new collaborations between humans and machines have emerged, allowing creators to create their work. The rules of the creative process by which works are created.

Will the legal status quo kill artists in favor of machines, technology and programmers? Or, conversely, is there a risk of overburdening the research and industry environment? Should the law be amended or enacted in parallel, as other jurisdictions are doing?

This law was established in the 19th century, has a certain universal and timeless character, and has undergone many technological changes without failure. Let’s see how we adapt to this new era.

---