Algorithms and speech.

• Author: Benjamin, Stuart Minor
• Position: First Amendment applicability to algorithm-based speech

One of the central questions in free speech jurisprudence is what activities the First Amendment encompasses. This Article considers that question in the context of an area of increasing importance--algorithm-based decisions. I begin by looking to broadly accepted legal sources, which for the First Amendment means primarily Supreme Court jurisprudence. That jurisprudence provides for very broad First Amendment coverage, and the Court has reinforced that breadth in recent cases. Under the Court's jurisprudence the First Amendment (and the heightened scrutiny it entails) would apply to many algorithm-based decisions, specifically those entailing substantive communications. We could of course adopt a limiting conception of the First Amendment, but any nonarbitrary exclusion of algorithm-based decisions would require major changes in the Court's jurisprudence. I believe that First Amendment coverage of algorithm-based decisions is too small a step to justify such changes. But insofar as we are concerned about the expansiveness of First Amendment coverage, we may want to limit it in two areas of genuine uncertainty: editorial decisions that are neither obvious nor communicated to the reader, and laws that single out speakers but do not regulate their speech. Even with those limitations, however, an enormous and growing amount of activity will be subject to heightened scrutiny absent a fundamental reorientation of First Amendment jurisprudence.

INTRODUCTION I. WHAT IS AT STAKE II. THE CENTRALITY AND EXPANSION OF SUPREME COURT JURISPRUDENCE A. A Note on Broadly Accepted Sources and Forms of Reasoning B. Expansion and Exceptions III. SUPREME COURT JURISPRUDENCE AND ALGORITHM-BASED DECISIONS IV. PRODUCING A DIFFERENT RESULT A. Relying on Particular Theories of the First Amendment B. An Algorithm-Based Line That Works V. SCOPE A. Requiring Communicating About Your Editing B. Regulations of Speakers Not Aimed at Their Speech CONCLUSION INTRODUCTION

More and more of our activity involves not merely the transmission of bits, but the transmission of bits according to algorithms and protocols created by humans and implemented by machines. (1) Messages travel over the Internet because of transmission protocols, coding decisions determine the look and feel of websites, and algorithms determine which links, messages, or stories rise to the top of search engine results and web aggregators' webpages. Most webpages have automated components, as do most online articles and all video games. (2) Are these algorithm-based outputs "speech" for purposes of the First Amendment? (3) That is, does the Free Speech Clause of the First Amendment apply to government regulation of these or other algorithm-based changes to bits? (4)

In this Article I address that question. I conclude that if we accept Supreme Court jurisprudence, the First Amendment encompasses a great swath of algorithm-based decisions--specifically, algorithm-based outputs that entail a substantive communication. We could decide to reject Supreme Court jurisprudence, or read it narrowly in order to limit its application. But for the purposes of this Article, I will not apply that lens to the existing case law. Instead, I will look to broadly accepted sources and forms of legal reasoning--which in the First Amendment context means primarily Supreme Court jurisprudence--and consider whether those sources lead to the conclusion that algorithm-based outputs are speech for First Amendment purposes. I find that the answer is yes for most algorithm-based editing.

For some, this answer will be unwelcome. A wide range of commentators have expressed concerns about potentially expansive interpretations of the scope of the Free Speech Clause, such that much, if not most, government regulation is subject to heightened judicial scrutiny. (5) Such a concern may motivate, at least in part, the contrasting answer that Tim Wu reaches in his piece. (6) One possible response to these concerns is to articulate a theory of the Free Speech Clause that excludes algorithm-based decisions, or, perhaps more modestly, search engine results (which have been the focus of some commentators). (7) Any such exclusion, however, will entail a radical revamping of our Free Speech Clause jurisprudence. And, as it turns out, there are interpretations that are consistent with existing jurisprudence (and, in my view, desirable on their own terms) that would limit the scope of the Free Speech Clause. Of course, one could find those interpretations insufficient, but I conclude that the inclusion of algorithm-based decisions in the First Amendment's protections does not substantially advance the argument for a radical revamping.

In a previous article I asked how difficult it would be to find that mere transmission of bits constituted speech. (8) One way of framing that question is to ask how hard it would be to expand the definition of speech to include something (mere transmission) that ordinarily would fall outside it. In this Article I address the converse question: How hard would it be to narrow the definition of speech to exclude something that Supreme Court jurisprudence would encompass? What would such an exclusion mean for First Amendment jurisprudence?

1. WHAT IS AT STAKE

   A huge range of bit manipulations involves the use of algorithms. Computer code is a set of instructions and algorithms. (9) Every webpage relies on many different algorithms for its structure, not to mention its transmission over the Internet. Indeed, every networked device depends on an electronic network built in part on algorithms.

   Around the turn of this century, there was considerable focus on whether computer code itself was speech for First Amendment purposes, such that regulations on the distribution of code implicated the First Amendment. (10) The government had concerns about the proliferation of some computer programs (notably, those perceived as jeopardizing security), and it sought to regulate the circulation of the code itself--the instructions to a computer that would enable the feared activity. (11) For what it is worth, the few courts that considered the issue found by and large that regulations of computer code were regulations of speech. (12)

   My focus here is not on the distribution of code, and thus not on whether code itself is speech. Rather, I consider whether the outputs of that code--the decisions created by algorithms--are speech for First Amendment purposes. The question whether the First Amendment applies to regulation of search engine results is different from the question whether the algorithms used by those search engines are speech. Even if the algorithms are not speech, their products may be.

   What sorts of regulations of algorithm-based decisions might be at issue? The most prominent possibility, and the one that has inspired the most commentary, is the regulation of search engine results, and in particular (given its large market share) Google. A company frustrated by its low PageRank (which hurt its ability to find clients) brought an action against Google for tortious interference with contractual relations, and Google successfully argued that the First Amendment applied to its search results. (13) Another company frustrated by its rankings on Google unsuccessfully argued that Google's search engine is an "essential facility" that must be opened to access, (14) and Frank Pasquale argued that Google should be understood as a new kind of bottleneck deserving of regulatory attention-an "essential cultural and political facility." (15) Pasquale and Oren Bracha have also argued that the government should be able to regulate search engines' ability to structure their results, and that the First Amendment does not encompass search engine results. (16) Eugene Volokh and Donald Falk, by contrast, have contended that all aspects of search engines' results are fully protected by the First Amendment. (17)

   But, at least for some commentators, the concern is not limited to Google. For instance, Tim Wu has suggested that the important issue is computer-generated outcomes more generally. Wu has contended that "nonhuman or automated choices" should not be treated as speech for First Amendment purposes. (18) This framing is useful. Under the prevailing jurisprudence, the existence of anticompetitive concerns with respect to Google (or any other particular entity) might affect the application of First Amendment scrutiny but not whether the underlying activity is encompassed by the First Amendment in the first place.

   The apparent motivation behind excluding algorithm-based decisions from First Amendment coverage is understandable. More and more of our activity involves bits, and those bits are frequently guided and shaped by algorithms. The more fully algorithm-based decisions are treated as speech, the more broadly First Amendment jurisprudence will apply. And this has real consequences. Content-based government regulations of speech are subject to strict scrutiny, which is very difficult to satisfy. (19) Content-neutral regulations are subject to intermediate scrutiny, which is an easier test to pass but still much more rigorous than the rational basis review applicable to ordinary regulation. (20)

   Heightened scrutiny raises the costs of regulation, both in requiring more justification ex ante and in increasing the likelihood that the regulation will be rejected on constitutional grounds (since the chances of rejection on constitutional grounds for ordinary legislation are near zero). It could be that we, as a society, like this outcome because we decide that we want less government regulation of algorithm-related industries, but my point here is simply that we disincentivize regulation when heightened scrutiny applies. Subjecting every regulation that affects algorithm-based transmissions to intermediate scrutiny would have dramatic consequences.

   Consider the Court's recent opinion in Sorrell v. IMS Health...