Published by Stanford Law Review, “Speech Certainty: Algorithmic Speech and the Limits of the First Amendment” refers to the principle of “speech certainty”—the idea that what makes speech is the speaker knowing what they said and when they said it—rooted “in the text, history, and purpose of the First Amendment.” Based on this principle, the authors argue that the First Amendment does not protect the output of machine learning algorithms.
Abstract 
Machine learning algorithms increasingly mediate our public discourse—from search engines to social media platforms to artificial intelligence companies. As their influence on online speech swells, so do questions of whether and how the First Amendment may apply to their output. A growing chorus of scholars has expressed doubt over whether the output of machine learning algorithms is truly speech within the meaning of the First Amendment, but none have suggested a workable way to cleanly draw the line between speech and non-speech. This Article proposes a way to successfully draw that line based on a principle that we call “speech certainty”—the basic idea that speech is only speech if the speaker knows what he said when he said it. This idea is rooted in the text, history, and purpose of the First Amendment, and built into modern speech doctrines of editorial discretion and expressive conduct. If this bedrock principle has been overlooked, it is because, until now, all speech has been imbued with speech certainty. Articulating its existence was never necessary. But machine learning has changed that. Unlike traditional code, a close look at how machine learning algorithms work reveals that the programmers who create them can never be certain of their output. Because that output lacks speech certainty, it’s not the programmer’s speech. Accordingly, this Article contends that the output of machine learning algorithms isn’t entitled to First Amendment protection. It reveals that the question of how an algorithm works is constitutionally significant. With the Supreme Court in Moody v. NetChoice demanding further inquiry into what constitutes protected expressive activity for social media platforms, that question can no longer be ignored. By failing to distinguish between traditional and machine learning algorithms, we risk sleepwalking into a radical departure from centuries of First Amendment jurisprudence. Protection for the output of machine learning algorithms would, for the first time in the Constitution’s history, protect speech that a speaker does not know he has said. Speech certainty provides a novel and principled approach to conceptualizing machine learning algorithms under existing First Amendment jurisprudence.