Week 4 - What is 'Pure' Speech?

Welcome back Free Speakers!

Last week, we determined that Character.AI had standing to bring First Amendment claims in our hypothetical. The next step is determining whether the outputs are speech, and to what extent they are speech.

Character.AI wants their LLMs outputs to be considered pure speech.^[1] Well, what does that mean, exactly? Let’s first look at the way the courts have separated speech. The Supreme Court has said “the Constitution looks beyond written or spoken words as mediums of expression,” meaning ‘speech’ can be more than just words on a page.^[2] That said, there are distinctions between the types of speech. We can look at the ways the First Amendment protects analyzes speech to elicit different definitions for different expressive conduct. 

In Week 2, we talked about the difference between strict and intermediate scrutiny. The First Amendment differentiates between types of speech in this way: core and peripheral protections. Regulations that limit core speech directly receive strict scrutiny, while regulations that limit core speech incidentally (through time, place, and manner restrictions) undergo intermediate scrutiny. The Court found that destroying draft cards to signal opposition to the Vietnam War was expressive conduct.^[3] Expressive conduct contains ‘speech’ and ‘non-speech’ elements.^[4] In O’Brien the government sought to regulate the ‘non-speech’ elements of the conduct. Resultantly, it underwent intermediate scrutiny (as it didn’t implicate content-based restrictions). In Tinker v. Des Moines, the Court found that wearing an arm band in protest of the Vietnam war was “pure speech… entitled to comprehensive protection under the First Amendment.”^[5] The conduct in Tinker and O’Brien both lacked spoken word, yet Tinker was afforded core First Amendment protections. That’s because the armbands in Tinker were wholly expressive. The burning of draft cards in O’Brien, however, affected the war effort. People aren’t going to be fighting for the country if they’re all burning their draft cards. The ‘non-speech’ element of O’Brien’s conduct could be regulated under intermediate scrutiny standards (being the government must present an “important or substantial interest” unrelated to the restriction and that the government’s restriction “is no greater than [] essential” to achieve their goal). 

So what does this have to do with pure speech?

For Character.AI’s outputs to be considered pure speech, they must prove there are no ‘non-speech’ elements of their conduct to be regulated. They don’t want to have the government encroaching on their AI’s training off a pesky technicality. So our next question follows to be: do Character.AI’s LLM outputs contain speech elements? 

When we look at the conduct in question (the conversations between C.AI and Sewell), we can see that the conversation is written word. This observation should theoretically quash further analysis, though the Court, in its order on the defendants’ motion to dismiss, states otherwise. The Court calls forth the expressive conduct test. The Court does so under the expressed purpose of “determin[ing] whether conduct is sufficiently similar to speech so as to warrant First Amendment protections.”^[6] The Court cites  Spence v. State of Wash. and Texas v. Johnson as supporting cases.^[7] The Supreme Court in Spence found that because Spence did not use “printed or spoken words,” the Court had to use the expressive conduct test.^[8] In other words, the Court conducted the test to make sure there was actually an expressive point to the activity he was doing. The Court assessed there were enough “elements of communication” in his action for it to be protected under the First Amendment.^[9] In Johnson, Johnson burned an American flag at a political protest.^[10] The Supreme Court found that the expressive conduct test must be applied when deciding if conduct had enough “communicative elements” to implicate the First Amendment.^[11] In both cases that the Court in Garcia cites, there were substantial non-speech elements at play. Those non-speech elements added some ambiguity in the expressive intent of the conduct—if any. The Court needed the expressive conduct test to discern the expressive intent. In the case at bar, however, the LLM outputs can be taken at face value for their expressive intent, because the outputs are words. The Court defends their inquiry of the outputs’ expression by citing Holloman v. Harland. In Holloman, a high school student was punished for “silently raising his fist during the daily flag salute” at school.^[12] The Court in Holloman found in their analysis that, “it does not ultimately matter whether Holloman’s act is characterized as ‘pure speech’ or ‘expressive conduct’ because this circuit… appl[ies] the same test in assessing school restrictions on either kind of expression.” Again, in Holloman no spoken or written words are at play, which warrants—or in fact requires—the use of the expressive conduct test. In cases where spoken and written words are the conduct under scrutiny, courts do not concern themselves with this test.^[13] The Supreme Court’s First Amendment jurisprudence presumes words to be expressive. The expression is explicit, because the words indicate the expression on their face. It is this distinction that differentiates pure speech and expressive conduct. The use of the test in this case is unwarranted, as it is not needed.

I disagree with the Court’s opinion on this issue for the aforementioned reasons, and I’ll continue my analysis under the assumption that I am right (brazen, I know). 

But perhaps there’s a different way to examine expression, and all the philosophical questions that come with it. Next week, I’ll continue my analysis on outputs being pure speech or not.

See you Free Speakers then!

References

[1] Garcia v. Character Technologies, Inc., 6:24-cv-01903, (M.D. Fla. Apr 04, 2025) ECF No. 98 https://www.courtlistener.com/docket/69300919/98/garcia-v-character-technologies-inc/

[2] Hurley v. Irish-American Gay, Lesbian, and Bisexual Group of Boston, Inc., 515 U.S. 557 (1995) https://supreme.justia.com/cases/federal/us/515/557/

[3] United States v. O’Brien, 391 U.S. 367 (1968) https://supreme.justia.com/cases/federal/us/391/367/

[4] Id. at 367.

[5] Tinker v. Des Moines Independent Community School District, 393 U.S. 503 (1969) https://supreme.justia.com/cases/federal/us/393/503/

[6] Garcia v. Character Technologies, Inc., 6:24-cv-01903, (M.D. Fla. May 21, 2025) ECF No. 115 at 30 https://www.courtlistener.com/docket/69300919/115/garcia-v-character-technologies-inc/

[7] Id. at 30.

[8] Spence v. Washington, 418 U.S. 405 (1974) https://supreme.justia.com/cases/federal/us/418/405/

[9] Id. at 376.

[10] Texas v. Johnson, 491 U.S. 397 (1989) https://supreme.justia.com/cases/federal/us/491/397/

[11] Id. at 404.

[12] Michael Holloman, on Behalf of and As Next Friend of His Son, Michael Holloman, Plaintiff-appellant, v. George Harland, Fawn Allred, Defendants-appellees.michael Holloman, Jr., Plaintiff-appellant, v. Walker County Board of Education, Defendant-appellee, 370 F.3d 1252 (11th Cir. 2004) https://law.justia.com/cases/federal/appellate-courts/F3/370/1252/522043/

[13] See Cohen v. California, 403 U.S. 15 (1971) https://supreme.justia.com/cases/federal/us/403/15/; Snyder v. Phelps, 562 U.S. 443 (2011) https://supreme.justia.com/cases/federal/us/562/443/. These cases examine free speech issues concerning written words. The Court does not examine expressive intent in these cases.