Journal of Free Speech Law: "A New Frontier for an International Right with No Frontiers: Freedom of Expression & Generative AI Outputs," by Evelyn Mary Aswad

The article is here; the Introduction:

In recent years, a lively scholarly discourse has emerged about whether and how the U.S. First Amendment protects generative AI outputs. Some have argued that such outputs are protected at the very least by the rights of AI users to receive information and to create their own speech. Others would not recognize gen AI outputs as receiving such First Amendment protections. The issue of whether gen AI program creators are entitled to free speech protections for such outputs has also spurred a variety of reactions.

As this discourse on the First Amendment and gen AI unfolds, it is also important to reflect on what the global freedom of expression standard has to say on the matter. This standard will affect discussions about national and regional regulatory approaches to gen AI throughout the world. In addition, global corporate responsibility standards call on companies to respect international human rights norms in their operations, which may also impact how businesses that provide gen AI services approach their activities.

Part I of this Article explores the scope of the existing global free expression standard. Part II considers the standard's application to gen AI outputs, including in various governmental and corporate contexts. Ultimately, this Article maintains that the global free expression standard protects the rights of individuals to seek and receive information of any kind, including gen AI outputs.

In addition, if human speakers share gen AI outputs as part of their own speech, this global standard also protects those speakers' right to impart information. Governmental attempts to restrict gen AI outputs are therefore subject to the standard's safeguards on how this human right can be limited. And companies providing general-purpose gen AI services should also respect human rights, including freedom of expression, in their operations.

Comments

[Murcushio]

It's an interesting article, but it seems somewhat wrongheaded in many ways.

In particular, it assigns these algorithmic models agency they do not have. LLM's and other genAI tools cannot have "interests" or "preferences" or "bias," all things that require agency, and which the article repeatedly attributes to them.

These things are tools, not people, and they're operated by corporations, not the government. That makes it pretty cut-and-dried, at least to me. Google (or Facebook, or whoever) are capable neither of granting no abridging speech rights. They can develop their stupid little toys to generate whatever output they want in response to any input they want, and nobody has had their rights abridged thereby.

I'm generally more concerned with responsibility, or liability, rather than rights in this arena. Obviously, the models themselves cannot be assigned blame for how they are used; blame requires agency. You can't put a crowbar on trial for being used to pry open a locked door during a burglary, after all.

It's pretty easy when it comes down to individuals. If you're a lawyer and you use these tools to try and lawyer for you, and they don't work, then YOU are held responsible, in the same way as if you'd used a crowbar to force entry.

But in that situation, it would be a very foolish person who filed suit against the crowbar manufacturer; assuming that it had at all operated legitimately (like, that it wasn't advertising "The Breaking and Entering 5000, our most efficient tool for going equipped yet!") they're not at fault. They could only possibly be at fault if they had some reasonable way of preventing their crowbars from being used to commit crimes and didn't avail themselves of it.

Which is something that those who provide LLMs often do have control over. Nikon can't stop you from using your camera to produce CSAM, but you can certainly make an LLM used for image generation not do that.

Individuals will certainly, and are being, held accountable for what happens when they use these tools, although I have seen attempts to deny responsibility and pin it all on the AI, which is just sad. The question is, to what extent are the providers of said tools responsible?

THAT'S the big deal.