- The principles of protecting AI-generated speech are the same as the protections established by the Framers
- The application of those principles may change (if narrowly) as technology advances
- North Carolina is starting to wrestle with applying restrictions to AI-generated speech, while others see AI in national security and economic development terms
The rise of artificial intelligence (AI) raises questions about regulating its use and abuse.
When considering laws regulating speech, we cannot be dazzled or feel threatened by the newness of technology into believing it is so new as to be outside the scope of the principles the Framers envisioned. We must instead drill down to those principles protecting our freedoms and apply them to that technology. AI’s potential disruptive power does not allow governments to violate the free speech rights of people using it.
As United States Supreme Court Justice Antonin Scalia said of free speech rights in 2014:
It’s the freedom of speech. It doesn’t matter whether you’re speaking in semaphore or a cellphone or any type of modern technology. You apply the same First Amendment principles to the new technology that you applied to the old.
There is little doubt that Scalia would add AI to the list of technologies to which those same First Amendment principles could be applied.
Justice Neil Gorsuch struck a similar chord at his 2017 confirmation hearing, but went further, saying, “So, the technology changes, but the principles do not. And it cannot be the case that the United States Constitution is any less protective of the people’s liberties today than it was the day it was drafted.”
What about liability if something generated by AI harms someone? The principle of “he who causes harm should compensate the victim” should apply. If a failure of an AI system causes harm, the producer of that system should provide compensation. If a misapplication of AI causes harm, such as an attorney relying on AI to produce a legal brief without verifying its correctness, the user should compensate.
Likewise, our free speech rights applied to AI do not eliminate the traditional legal protections against libel or fraud. Deception generated through AI is no better or worse in principle than deception through other means. AI-generated speech used to damage a person’s reputation deceptively should be subject to libel suits, and those using AI to commit fraud should face the legal consequences of that fraud.
That said, there can be narrow circumstances in which changing technologies can cause a rebalancing of how traditional principles are applied. We may be at such a point with AI-generated child pornography.
Child pornography is banned by the federal government and every state. The Supreme Court has upheld bans on not just the production of child pornography, but also its distribution. In upholding New York’s ban of the distribution of child porn in New York v. Ferber (1982), the court held that “the advertising and selling of child pornography provide an economic motive for, and are thus an integral part of, the production of such materials.” That link between child porn and the actual sexual abuse of children provides a compelling government interest in banning its distribution.
But what if no children are harmed? In Ashcroft v. Free Speech Coalition (2002), the high court found that “[v]irtual child pornography is not ‘intrinsically related’ to the sexual abuse of children.” In other words, the court found that governments had a compelling interest in stopping the distribution of sex acts involving children, but not in preventing the idea of child sex when no actual children were involved.
Justice Clarence Thomas warned in his concurring opinion in Ashcroft that advances in technology could make it impossible to differentiate real from virtual depictions, which would necessitate a change in the court’s view of virtual child pornography:
[T]echnology may evolve to the point where it becomes impossible to enforce actual child pornography laws because the Government cannot prove that certain pornographic images are of real children. In the event this occurs, the Government should not be foreclosed from enacting a regulation of virtual child pornography that contains an appropriate affirmative defense or some other narrowly drawn restriction.
Twenty-three years later, we may be at the point Thomas warned us about.
North Carolina House Bill 375, “AI/Ban Deceptive Ads,” would make the production, possession, or viewing of “generated child pornography” a felony. It also would ban the distribution of a “fabricated intimate image of another person” without that person’s consent. The first person arrested for possessing AI-generated child porn would almost certainly sue the state, setting up a test of whether the courts believe we are at the point Justice Thomas cautioned us about in his Ashcroft concurrence.
The General Assembly has considered other restrictions on AI use and will likely do so again.
State regulations like those being considered in the General Assembly may be rendered moot if a bill recently passed by the U.S. House of Representatives also passes the Senate. Tucked into H.R. 1, the “One Big Beautiful Bill Act,” is a provision that would prevent states from enforcing their laws regulating AI for the next 10 years. Supporters say the ban is designed to “stop a confusing patchwork of state AI laws that have cropped up nationwide and give Congress space to craft its own AI legislation while preserving American leadership.”
That approach is what the American Enterprise Institute’s Klon Kitchen called a course correction to one that “prioritizes American innovation, strengthens our competitive edge, and ensures that AI remains a force for growth and strategic advantage.”
As federal and state officials wrestle over how or if they should regulate AI, they should neither use the fear of it to stifle our First Amendment speech rights nor permit AI users to commit fraud or other abuses of speech.
AI may be new, but the principles we should use to regulate it are not.
