“While the NetChoice decision, on its face, is a procedural decision applying to social media companies, the discussion engaged in by the Court hints at a number of critical issues relating to the intersection of AI and free speech, the First Amendment, and defamation.”
The United States Supreme Court recently ruled in a pair of cases (Moody v. NetChoice, LLC, 603 U.S. ____ (2024)) regarding the constitutionality of laws pertaining to moderation of content on social media platforms. While the cases were reversed and remanded largely on procedural grounds, and the statutes at issue allegedly pertain to social media platforms, vagaries in statutory drafting, the Court’s analysis, and the recent integration of artificial intelligence (AI) into social media platforms raise important considerations for both the application of the laws to AI tools and platforms, and give us a glimpse into potential issues that are likely to arise when courts address matters pertaining to AI in the future.
The Statutes and Their Applicability
In 2021, Florida and Texas passed laws regulating content moderation on social media platforms. Texas’ House Bill 20 and Florida’s Senate Bill 7072 each purport to regulate large social media companies and other internet platforms but differ in the entities they cover and the activities they limit. Because of differences in the statutes, one likely applies to artificial intelligence companies while the other does not.
The Florida statute defines a “social media platform” as “any information service, system, internet search engine, or access software provider that: (1) [p]rovides or enables computer access by multiple users to a computer server…; (2) [o]perates as a sole proprietorship [or entity]; (3) [d]oes business in the state; and (4)…[h]as annual gross revenues in excess of $100 million [or h]as at least 100 million monthly individual [monthly users].” Fla. Stat. § 501.2041(1)(g). The most popular GenAI platforms of today likely meet Florida’s definition of “social media platform” and would therefore bs subject to the law. The statute generally addresses three types of conduct: censoring, deplatforming, and shadow banning. Two of these activities – censoring and shadow banning – relate to activities that are likely not applicable to most artificial intelligence tools, specifically posting content by users and eliminating or reducing the exposure of one user to other users on the social media platform.
However, the third type of conduct – deplatforming – pertains to “the action or practice by a social media platform to permanently delete or ban a user or to temporarily delete or ban a user from the social media platform for more than 14 days.” Fla. Stat. § 501.2041(1)(c). Because deplatforming, unlike censoring and shadow banning, is not reliant on user posts or exposure, the restrictions on, and requirements pertaining to, deplatforming may apply to AI platforms, such as if an AI platform revokes a user account for abuse of the system. Additionally, the Florida statute appears to apply to AI platforms with respect to providing users with notice about changes to its rules, terms, and agreements before implementing any changes and may not make any such changes more than once every 30 days. Fla. Stat. § 501.2041(2)(c).
The Texas statute uses a narrower definition of “social media platform” that is narrower than the definition used in the Florida statute, “’[s]ocial media platform’ means an Internet website or application that is open to the public, allows a user to create an account, and enables users to communicate with other users for the primary purpose of posting information, comments, messages, or images…”. Tex. Stat. § 120.001(1). Because AI platforms generally do not enable users to communicate with other users for the primary purpose of posting information, comments, messages, or images, AI platforms would generally fall outside of the Texas statute.
These differences highlight the issues created by the lack of a comprehensive AI legal framework in the United States – companies doing business in Florida need to consider whether they are subject to the Florida statute (and comply with same) while companies doing business in Texas have no such concerns.
The Lawsuits and the Court’s Reasoning Highlight Future
NetChoice LLC and the Computer & Communications Industry Association, both trade associations, filed facial First Amendment challenges to the state laws; both district courts entered preliminary injunctions enjoining the statutes from being enforced. See 546 F. Supp. 3d 1082, 1096 (N.D. Fla. 2021); 573 F. Supp. 3d 1092, 1117 (W.D. Tex. 2021). The states appealed with split results – the Eleventh Circuit upheld the injunction while the Fifth Circuit reversed. See 34 F. 4th 1196 (11th Cir. 2022); 49 F. 4th 439 (5th Cir. 2022). Both decisions were appealed, and the Supreme Court granted certiorari to address the circuit split.
The Court reversed and remanded on procedural grounds – finding that the courts below failed to “explore the laws’ full range of applications – the constitutionally impermissible and permissible both – and compare the two sets” as is required in facial challenges to statutes. 603 U.S. ____ (2024). However, the Court also found it necessary “to say more about how the First Amendment relates to the laws’ content-moderation provisions, to ensure that the facial analysis proceeds on the right path in the courts below.” Id. This was particularly due to the Fifth Circuit’s decision that “held that the content choices the major platforms make for their main feeds are ‘not speech’ at all, so States may regulate them free of the First Amendment’s restraints.” Id. It is in this analysis that the majority opinion touches on several issues
The Court repeatedly stated that the introduction of technology does not change First Amendment law, “…we have repeatedly held that laws curtailing…editorial choices must meet the First Amendment’s requirements…[t]he principle does not change because the curated compilation has gone from the physical to the virtual world.” Id. The Court dismissed the idea that the use of technology would alter its analysis stating “[d]espite the relative novelty of the technology before us, the main problem in this case – and the inquiry it calls for – is not new.” Id.
We have repeatedly faced the question whether ordering a party to provide a forum for someone else’s views implicates the First Amendment. And we have repeatedly held that it does so if, though only if, the regulated party is engaged in its own expressive activity, which the mandated access would alter or disrupt. So too we have held, when applying that principle, that expressive activity includes presenting a curated compilation of speech originally created by others.
Id.
The Court therefore analyzed existing First Amendment law in the context of companies and editorial content. In Miami Herald Publishing Co. v. Tornillo, 418 U.S. 241 (1974) addressed a Florida law that required a newspaper to give a political candidate a right to reply when the newspaper published “criticism and attacks on his record.” Id., at 243. The Court held that Florida statute violated the First Amendment because it interfered with the newspaper’s “exercise of editorial control and judgment” as forcing the paper to print what “it would not otherwise print…intrudes into the function of the editors.” Id., at 256, 258.
The Court’s analysis assumes, in all instances, that a human is involved in the creation, curating, or editing of content. However, what happens if content is created or edited by AI? In her concurring opinion, Justice Barrett raises this same question:
Consider, for instance, how platforms use algorithms to prioritize and remove content on their feeds…[i]f they create an algorithm to help them identify and delete that content, the First Amendment protects their exercise of editorial judgment – even if the algorithm does most of the deleting without a person in the loop…But what if a platform’s algorithm just presents automatically to each user whatever the algorithm thinks the user would like…? The First Amendment implications…might be different for that kind of algorithm. And what about AI, which is rapidly evolving…has a human being with First Amendment rights made an inherently expressive ‘choice…not to propound a particular point of view?’
603 U.S. ____ (2024) (Barrett, J. concurring at 2-3)(quoting Hurley, 515 U.S., at 575). Justice Barrett concludes that “the way platforms use this sort of technology might have constitutional significance.” Id. While she does not reach any conclusions, Justice Barrett certainly lays out the case that the output of at least some artificial intelligence technologies may not qualify as speech that is entitled to First Amendment protections.
The United States Copyright Office (and other copyright offices across the globe) have determined that Generative Artificial Intelligence (“GenAI”) does not qualify as an ‘author’ and is therefore not entitled to receive copyright protection for the output of GenAI. “Copyright has never stretched so far, however, as to protect works generated new forms of technology operating absent any guiding human hand…Human authorship is a bedrock requirement of copyright…The act of human creation…was thus central to American copyright from its very inception.” Thaler v. Perlmutter, 687 F.Supp.3d 140, 147 (D.D.C. 2023). A court may likewise conclude that a human touch is necessary for the output of GenAI, or the editorial decisions of an AI tool, to be entitled to protections under the First Amendment. Alternatively, a court may determine that there is a difference between the standards for being an author under copyright law and being a speaker entitled to First Amendment protections.
Additionally, AI tools are not created in a vacuum – many foundational models are trained on data written by humans, many are programmed by humans, and many have safeguards and guardrails put in place that were decided on, and programmed by, humans. It is therefore arguably inaccurate in most-to-all cases to determine that content created, and decisions made, by AI tools were taken entirely devoid of human input. The question is one of degrees – at what point is the human touch so remote that it is unreasonable to consider AI an author or editor entitle to the same protections, and subject to the same liabilities, as humans? These are all questions that will need to be addressed at some point.
The AI Issue May Need to be Addressed Relatively Soon
As a result of social media platforms incorporating artificial intelligence tools into their products – such as Facebook’s incorporation of Meta AI where the Facebook platform utilizes Meta AI to summarize comments to user posts – the questions raised by Justice Barrett may need to be addressed sooner rather than later. For example, does the AI comment summary tool censor comments and does that censoring violate either of the state laws even though the original comment may still be readable? What if a particular comment is not summarized and is also downranked so it becomes hidden? Moreover, what if social media companies are using AI tools to engage in content moderation, such as “shadow banning” as defined by the Florida statute?
While the NetChoice decision, on its face, is a procedural decision applying to social media companies, the discussion engaged in by the Court hints at a number of critical issues relating to the intersection of AI and free speech, the First Amendment, and defamation. These issues will likely need to be addressed sooner rather than later and are important considerations for all developers and users of AI technologies.
Image Source: Deposit Photos
Author: phonlamai
Image ID: 240332172
