The First Amendment and AI “Slop and Swarms”

​AI scholar and USC Annenberg professor Kate Crawford has written a compelling thought piece for the Knight Institute. “Of Slop and Swarms: The First Amendment’s Next Test,” raises thorny questions. In fact, as much as anything, it’s a roadmap for anticipating the kinds of free speech issues that AI is likely to engender.

It’s well worth your time to read Professor Crawford’s essay from beginning to end, but the gist of her assertions centers on the juxtaposition of human agency and generative AI, which produces content.

“Generative AI,” Crawford cautions, “leaves First Amendment doctrine without a clear account of who, if anyone, is speaking.” And if the answer is that no actual person is speaking, how does the law treat that?

The “presumption of speaking” – of actual speech in the First Amendment sense of expression – would seem then to require both a human speaker and human control.

Such questions are entirely new territory from a First Amendment perspective, which is the point of Crawford’s essay. Particularly thorny issues that Crawford previews include:

• “Slopigation” – a term Crawford uses to describe “endlessly scaling litigation over AI content that no human directly expressed.” Who is responsible for defamation if the author is no one in particular?

 

• If AI itself is deemed to have human-style agency – viewpoints, for example (or, one imagines, creative rights) – then many forms of useful, democratic AI regulation could be upended.

 

• We could learn a lot from the historical experience of the U.S. Copyright Office, which developed a two-part threshold test for determining responsibility: 1) the amount and type of human creative contribution to a piece of content, and 2) human “authorship” stemming from human choices that have intended or at least somewhat predictable outcomes. Predictability, she argues, is inherently tied to human expression, creative or otherwise.

​

• An “express contribution” by a human author is another way to locate AI within existing law. Without such contributions by a person, there is no “speaker” in the traditional sense. Editorial choices, arrangement, and other forms of substantive intervention will all matter when debating the merits of “expression” in the age of AI.

Crawford returns to her central premise – the “presumption of speaking” in the First Amendment sense of protected expression requires both a human speaker and meaningful human control. This approach, says Crawford, tracks Justice Amy Coney Barrett’s stance in Moody v. NetChoice.

That same Supreme Court case raised (but did not settle) an adjacent question involving government regulation of social media content moderation. No one likes the idea of government overreach into what is clearly an expressive medium, but what happens when platform owners cede control and let large language models determine, for example, what is “hateful” and what is “permissible” speech? 

Other points to ponder at the intersection of free speech and AI include:

• Does the “inherently expressive” standard apply to AI-generated content produced and distributed at scale? Where does programming end and authentic human expression begin?

• In considering the boundaries of “speech” in the age of AI, is there a danger that communicative machines may end up being favored over the people who use them?

​

• The First Amendment needs a threshold test – and soon – to even begin navigating AI-generated content.

So stay tuned – probably for the next 20-plus years. In the meantime, read Professor Crawford’s thought-provoking essay.

DONATE & HELP US DEFEND YOUR FIRST AMENDMENT RIGHTS