Code just became the newest medium artists are fighting over

Code has become a contested artistic medium: a set of instructions that can also function as material, performance, artwork, and a site of political debate. Artists have long used software to make images, sounds, and interactive experiences; now, disputes over artificial intelligence, copyright, access, and authorship are making the code behind creative work newly visible. That shift matters in a digital economy where the U.S. Copyright Office received 10,753 registration applications involving AI-generated material from January 2023 through December 2024, while emphasizing that copyright protection depends on human authorship. The story connects code art’s history and forms with today’s fights over AI, ownership, and who gets to shape the tools.

Code as an artistic medium: definition and development

Code as an artistic medium means using computational instructions not merely to produce an outcome, but as a consequential part of the artwork’s form, meaning, or experience. In this approach, the artist may compose rules, write or adapt software, select data, or design an interaction. The resulting work can be a fixed image, a changing system, a live performance, or a process that invites audience participation.

This idea predates today’s generative AI debate. In the 1960s, artists and researchers began using computers to make plotted drawings and algorithmic compositions. Frieder Nake’s 1965 computer-generated drawings, for example, used programmed rules to produce images that were structured yet not identical in every detail. The term “software art” later described work that foregrounds software itself, including its logic, interfaces, and cultural assumptions—not only the visual or sonic output.

Generative and algorithmic art

Generative art is made through a system that contributes to the production of the work. The artist might specify a set of rules, parameters, or constraints, then allow software to generate variations. Algorithmic art is a closely related category in which algorithms—the step-by-step procedures used to solve a problem or produce an output—are central to the work’s structure.

The artist’s role may therefore resemble that of a composer or system designer: choosing the rules and deciding which results to keep, alter, or present. The code need not be fully autonomous to be generative. Human selection, editing, and framing can remain integral parts of the artistic process.

Interactive and networked art

Interactive art uses software to respond to a viewer’s actions, such as movement, touch, or choices. Networked art extends that interaction across digital networks, potentially drawing on live data, remote participants, or online platforms. In these forms, the work may change with each encounter, making audience behavior and technical infrastructure part of the experience.

This dependence on systems raises practical questions that traditional static artworks do not always face: Does a work remain the same when its software is updated? What happens when a platform shuts down or a dataset disappears? Preservation can require documenting source code, hardware, dependencies, and the conditions under which a work runs.

Code’s expressive power and its limits

Code gives artists a way to build repeatable procedures, responsive environments, and works that can operate at a scale or speed difficult to achieve by hand. But code is not a neutral material. Its behavior reflects the choices of its authors, the data and systems it relies on, and the institutions that own or control those systems. That combination of creative possibility and embedded power helps explain why artists are contesting not just the outputs of code, but the rules behind them.

Code as score, instruction, and performance

A program can function like a score: it specifies actions that a computer performs, much as musical notation specifies actions for performers. Yet execution introduces a distinction between the written instructions and the event they produce. Different machines, software versions, or inputs may lead to different results, giving some code-based works a performance-like quality.

This makes the source code potentially meaningful in its own right. Artists may expose it, conceal it, invite others to modify it, or use its constraints as a conceptual element. The code’s legibility can also be part of the audience’s experience: a work may ask viewers to consider whether they are looking at an image, a process, or the decisions that generated both.

Bias, labor, and infrastructure

Computational systems inherit assumptions from their design and training materials. A generative model’s output can reflect patterns in its dataset, while platform interfaces and moderation policies shape which work is easy to create, distribute, or discover. Artists who examine code critically may reveal these hidden choices rather than treating software as an impartial tool.

The infrastructure behind digital art also has material costs: servers, energy, specialized hardware, and human labor. Research published in the journal Joule estimated that training GPT-3 consumed 1,287 megawatt-hours of electricity, associated with about 552 tonnes of carbon dioxide equivalent under the study’s assumptions. That figure concerns one model-training process, not every artwork or AI query, but it illustrates why the material footprint of computational creativity is part of the discussion.

Artists’ disputes over AI, authorship, and ownership

The newest high-profile conflict is over generative AI: systems that produce text, images, audio, or other content in response to prompts. Artists, technology companies, museums, and lawmakers are debating whether training on copyrighted work requires permission, how much human input makes an output protectable, and who should receive credit or payment. These disputes turn code into a contested medium because the creative system itself—and the data and labor used to build it—are under scrutiny.

Copyright and human authorship

The U.S. Copyright Office’s 2025 report, “Copyright and Artificial Intelligence, Part 2: Copyrightability,” states that copyright protection requires human authorship. It explains that AI assistance does not automatically disqualify a work, but material generated by a system without sufficient human creative control is not protected simply because a person supplied prompts. Human-authored contributions—such as original selection, arrangement, or modification—may qualify depending on the circumstances.

The Office reported receiving 10,753 applications involving AI-generated material between January 2023 and December 2024. It also noted that the applications ranged from works made substantially with AI to works in which AI was only one element. The number indicates a sharp practical need for guidance, though it is not a count of all AI artworks or a measure of how many claims will ultimately be registered.

Training data and artists’ consent

Training-data disputes concern whether creators’ works may be copied and used to train commercial systems without permission or compensation. Artists argue that their work can become part of a model’s capabilities while remaining invisible in the product and receiving no payment. Technology companies and other users of large datasets have raised legal defenses that include fair use, a question courts assess in context rather than through a single rule that resolves every training practice.

The scale of the issue is substantial, but figures require careful interpretation. The U.S. Copyright Office’s report addresses legal and policy questions rather than establishing a definitive count of works used in every training dataset. Artists and researchers have responded with approaches such as licensing, dataset disclosure, opt-out mechanisms, and technical tools intended to disrupt or identify unauthorized use. None, on its own, settles the underlying legal or economic debate.

Platform power and access to creative tools

The tools artists use are increasingly provided through commercial platforms, cloud services, and proprietary models. Access can lower the technical barrier to making computational work, but it can also create dependence on a company’s pricing, terms, moderation decisions, and continued operation. Artists may have little visibility into a model’s training data or the changes introduced through software updates.

This tension echoes earlier debates about digital art and the ownership of tools. Open-source software can allow inspection, adaptation, and sharing, while proprietary software may offer convenience or capabilities that are difficult to reproduce independently. The choice is not simply between good and bad technology: it involves trade-offs among access, control, privacy, reliability, and the ability to preserve a work over time.

Examples of code as artistic practice

Artists have used code to make formal experiments, question how technology represents people, and expose the systems governing digital life. These examples show that the medium is not limited to AI image generation. It includes works in which programming is a visual language, a participatory structure, or a means of critique.

Refik Anadol and data-driven installations

Refik Anadol’s large-scale installations use data and machine-learning techniques to create moving, immersive environments. In works such as “Unsupervised,” presented at the Museum of Modern Art in New York, computational processes transform collections of images into shifting visual forms. The work makes the system’s transformations part of the spectacle while prompting questions about how cultural collections are interpreted by machine-learning models.

Allison Parrish and language-based computation

Allison Parrish uses computational methods to explore language, poetry, and the unexpected associations produced by text systems. Her practice demonstrates that code-based art can focus on the structure and ambiguity of language rather than on visual effects alone. By making systems that transform or recombine words, such work invites audiences to notice both the creativity of the procedure and the limits of its interpretation.

Early computer art and continuing preservation

Early computer artists such as Frieder Nake showed that programmed procedures could be an artistic method decades before today’s consumer AI tools. Their work also highlights a continuing challenge: digital art can become inaccessible when the hardware, operating systems, or software it depends on are obsolete. Museums and archives increasingly treat documentation, emulation, and technical maintenance as part of preserving the work—not merely as backstage tasks.

Conclusion: treating code as both medium and public issue

Code is an artistic medium because it can serve as material, score, tool, and conceptual structure. Generative and interactive art demonstrate its expressive range; debates over AI copyright, training data, infrastructure, and platform power reveal its social consequences. The Copyright Office’s thousands of AI-related applications show that questions of authorship are no longer abstract, while research on computational energy use reminds us that digital creativity has physical costs.

For artists, audiences, and institutions, the useful next step is to look beyond the output: ask who wrote or trained the system, what sources and labor it depends on, how the work can be preserved, and who benefits from its use. Further reading from the U.S. Copyright Office and the artists and institutions named here can help ground those questions in both policy and practice.

Sources: U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, 2025, https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf; Museum of Modern Art, Refik Anadol: Unsupervised, https://www.moma.org/calendar/exhibitions/5535; Joule, “The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink,” 2022, https://www.cell.com/joule/fulltext/S2542-4351(22)00410-8; Whitney Museum of American Art, Frieder Nake, https://whitney.org/artists/11073; Allison Parrish, official website, https://www.allisonparrish.com/

Related Post