Since late 2022, visitors to New York’s Museum of Modern Art have been mesmerized by Refik Anadol’s 24-foot by 24-foot artificial intelligence–generated artwork Unsupervised. It is a stunning work to contemplate, and all the more so given the realization that artists have just begun to scratch the surface on ways to use AI in their work.
Unsupervised also backdrops some of the complexities raised by calls to regulate AI through government-imposed limits. Simply put, under what circumstances is there a First Amendment right to compute? We will soon need an answer.
The First Amendment provides that “Congress shall make no law … abridging the freedom of speech.” The word “speech” encompasses not only literal speech but also a broad range of expressive activities that can include painting, photography and musical and dramatic works. When regulating expression, the government has more latitude to enact content-neutral laws than laws that target specific content. For instance, while performing live music is expressive, a city noise ordinance that regulates the volume of outdoor amplified music does not violate the First Amendment. By contrast, a city ordinance regulating the volume of outdoor amplified music of only one particular genre would be unconstitutional.
Federal appeals courts have considered the First Amendment aspects of computer code in only a limited number of cases. The Second, Sixth and (in an opinion that was subsequently withdrawn on procedural grounds) Ninth Circuits have concluded that computer code can receive First Amendment protection. The Sixth Circuit, for example, wrote that “because computer source code is an expressive means for the exchange of information and ideas about computer programming, we hold that it is protected by the First Amendment.”
The question of when computer code is expressive is related to, but distinct from, asking whether the purpose of performing the resulting computation is expressive. As Anadol’s Unsupervised makes clear, computation can be used for expressive purposes. But what about the AI algorithms steering driverless cars? While there can be an expressive aspect to writing the software for driverless cars, when that software is run, the computations and the resulting decisions involved in navigating a vehicle through city streets are purely functional rather than expressive. Therefore, laws addressing the operation of driverless cars do not implicate the First Amendment.
As a thought experiment, suppose that the government were to prohibit training or use of AI models beyond a certain size. This is not so far-fetched. In March, thousands of people, including many notable AI experts, signed on to an open letter calling “on all AI labs to immediately pause for at least [six] months the training of AI systems more powerful than GPT-4,” and adding: “If such a pause cannot be enacted quickly, governments should step in and institute a moratorium.”
Would a government-imposed limit on AI computations implicate the First Amendment? Proponents would assert that it would not, underscoring that the regulation would target computation, not expression. They would argue that using large, compute-intensive AI models does not necessarily involve expressive activity, and that any impact on artists and others whose expression might be limited by the regulation would be only incidental. Those costs, they could assert, would be far more than outweighed by the alleged benefits to society of preventing the premature development and unchecked deployment of very large AI models.
But there is also a counterargument. The most compelling recent developments in AI—and those that are spurring the strongest calls for regulation—involve generative AI. As a post from IBM Research explains, generative AI “refers to deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on.” These systems commonly produce expressive content—often images and writing—as underscored by the multiple ongoing copyright lawsuits regarding the role of human-generated training data underpinning generative AI systems. The robust debate regarding who should have rights to the images and other works produced by generative AI is itself evidence that those works have expressive value.
An emerging generation of artists, composers and members of the broader public will use AI in extraordinarily innovative and creative ways. They could argue that government regulation limiting AI computation would have an impact on expression that is far more than incidental. After all, a key application of generative AI is to help produce prose, imagery, music and video. Thus, the argument might go, one of the most substantive consequences of a government cap on AI computation would be to constrain the production of expressive content. Under that reasoning, the First Amendment would indeed place some limits on the manner and extent to which government regulations could target AI computation.
The stunning recent advances in AI mean that computation and expression are increasingly overlapping in novel and rapidly evolving ways. The question “Is there a First Amendment right to compute?” might have seemed largely theoretical in the recent past. Today, it is one of many questions regarding AI that need answering.
This is an opinion and analysis article, and the views expressed by the author or authors are not necessarily those of Scientific American.