In the Style of…: The Frailty of the Idea-Expression Distinction in the Age of AI

One of the foundational concepts of copyright law is that ideas are not protected, but the expression of ideas is. This distinction is necessary to ensure that copyright law “strike[s] an appropriate balance between giving protection to the skill and judgment exercised by authors in the expression of their ideas [...] and leaving ideas and elements from the public domain free for all to draw upon”. Courts have tried to keep the idea-expression distinction clear despite criticism as to the concept’s usefulness.

Artificial intelligence (AI) has now brought a new layer to this debate. Since the advent of AI text-to-image generators, visual artists have been sounding the alarm about the ability of AI generators’ outputs to replicate the distinctive or recognizable elements of their art.

In other words, some artists allege that AI is stealing their artistic style.

Courts in Canada and the United States have typically conceptualized artistic style as an idea that is not protectable under copyright law (see, for instance, this case from Ontario and this one from Québec, and this overview of older American case law on the issue). Indeed, if someone is able to copyright something as broad as their general artistic style, this could unduly limit what other people can do when creating their own artworks. This, in turn, would run contrary to the public’s interest in new artworks entering a broad public domain.

Yet – is artistic style truly just an idea?

Artistic style comes down to a manifestation of an idea, through visual or textural characteristics, that goes beyond what is being depicted in the artwork, but includes how it is being depicted. Indeed, the how falls in line with what we would traditionally consider to be expression: the author’s choices on how to depict the work’s subject in terms of colour, texture, medium, and tone. How did the artist employ skill and judgement in portraying their ideas?

After all, two artists could both depict the same item using the same medium (ex.: painting) but their artistic styles would shine through in the respective choices they made in representing the idea in question. These are intrinsic, tangible components of an artwork that go beyond what is being depicted in it. The how behind the two artists’ respective applications of brushstrokes to canvas is what will demarcate their works from each other’s and from their peers’—in other words, their style. Yet this crucial component of art remains unprotected by copyright law.

This apparent lack of protection is somewhat justifiable. What constitutes a person’s artistic style can be difficult to define and can change and evolve over time. Excluding style from copyright law’s protection is in line with the need to keep the public domain broad. Indeed, many artists use existing copyrighted works as inspiration for their own works. Many artistic movements have flourished precisely because talented individuals mimicked each other. Entire categories of artworks, such as works of pastiche and parody, would not exist at all were it not for this ability to draw from existing works and their styles.

The problem now is that AI has vastly changed the game. Human artists simply cannot keep up with the efficiency and scope of what AI can accomplish, generating new visual works in seconds. It is therefore especially crucial for artists to demarcate themselves through the distinctiveness of their work.

However, “in the style of…” prompts for AI, also known as the neural style transfer (NST) technique, have made everything that much more complicated. Other possible formulations for the prompt might include “Draw me X that is reminiscent of X artist” or “that looks like...” or “that imitates...” With a few words from a user, AI can generate images that purposefully take elements from a specific person’s artwork (including by identifying commonalities between various works) to produce a final product. AI can therefore identify and replicate the throughline or dominant motifs of that person’s art (in other words, their style) with visible success.

AI cannot currently come up with ideas nor express them on its own; the user submitting the prompt must do that. Beyond the prompt, AI lacks the autonomy to create works of its own volition. Whether this prompt explicitly asks for a specific artistic style or not, all AI can currently do in response is combine and mimic the human-authored work it was trained on. If, under a “holistic” or “qualitative” examination of the original work and the AI-generated image, there is similarity with a substantial part of the human-authored work, it is difficult to see why this would not constitute copyright infringement. This is especially true if the AI is specifically being told to imitate a particular person’s style, and this style successfully came through in the output.

The underlying idea or subject of the work that the user is prompting AI to produce is not relevant. The key consideration should be the replication of the characteristics that define that human author’s expression, especially what makes it unique, recognizable, and original. This collection of characteristics is what draws people to a specific artist’s work and will often constitute the main reason that this artist can benefit financially from their artworks. One of copyright law’s primary purposes is to enable authors to benefit economically from their works for a limited time before the work enters the public domain.

AI-generated images created in response to “in the style of…” prompts have put on full display the problems with the idea-expression distinction. They have also shown there is something to protect in the recognizable artistic characteristics that are the product of a human author’s own originality, and that go beyond specific fixed artworks. Whether this protection should come through copyright law or through a new form of IP remains up for debate. One possibility might be to seek protection through trademark law instead. Certain well-known artists may have amassed sufficient goodwill (repute or recognizability) for their art to be protected in a manner akin to a trademark. Another possibility may be to seek recourse through existing common law torts, such as misappropriation of personality.

Either way, disallowing the use of artistic style in text-to-image AI generator prompts could be in the public interest. People who do not possess traditional artistic skills may view the increased possibility for AI-generated expressive output positively, as a certain democratization of art. However, there is simply no way for human artists to compete with the speed, output, and low cost of AI. Not only could the economic impact on artists be severe, but it could disincentivize the publication of compelling, original, human-authored artwork, and the impact of that would be immeasurable. At the same time, it is also important to ensure that the public domain is not unduly restricted by excessive copyright “red tape” that encroaches upon ideas as opposed to the expression of those ideas.

This opinion was written by a CIPPIC student. The opinion is the author's, and does not necessarily reflect CIPPIC's policy position.