Part I - Theoretical Approaches to the Disinformation Problem

2 The Internet, Democracy, and Misinformation

Robert C. Post

The structure of society is heavily dependent upon its means of producing and distributing information. As its methods of communication change, so does a society. In Europe, for example, the invention of the printing press created what we now call the public sphere. The public sphere, in turn, facilitated the appearance of ‘public opinion’, which made possible wholly new forms of politics and governance, including the democracies we treasure today. Society is presently in the midst of an information revolution. It is shifting from analogue to digital information, and it has invented the Internet as a nearly universal means for distributing digital information. Taken together, these two changes are profoundly affecting the organization of our society. With frightening rapidity, these innovations have created a wholly new digital public sphere that is both virtual and pervasive.

Law is of course a lagging indicator. We typically call upon law only after our lives have been disrupted, when we turn to law to intervene and restore order. But law will prove a blunt and ineffective instrument unless we first identify with precision the causes of our discomfort. We have already witnessed too many incoherent legal interventions in the developing arena of the virtual public sphere.¹

The theme of this volume is misinformation. This theme cannot be adequately understood unless it is theorized within the context of the new social structures everywhere emerging due to the Internet’s distribution of digital information. In this chapter I shall discuss six unique and novel dangers that the Internet may pose to democratic forms of self-government. These threats stem from aspects of the Internet that are qualitatively different from all previous forms of mass communication. Each of these threats has important consequences for the problem of misinformation, but because these threats are fundamental new, we only dimly understand them. We cannot fashion adequate legal responses to the flood of misinformation that prompts this volume until we first theorize the nature of these threats.

2.1 How the Internet Differs from Past Media of Mass Communication

Without purporting to be comprehensive, there are (at least) three ways in which the Internet differs from all prior mass media: zero marginal information cost, integration with life tasks, and interactivity.

First, digitized information spreads on the Internet in a frictionless way that is virtually cost-free at the margin. The price of sending information to 1,000 persons is no greater than sending it to 1 person. The price of sending information to someone on the other side of the globe is no greater than sending it to someone around the corner. As a result, the Internet differs from prior mass media in three important ways:

• Scale: Information spreads on the Internet on a scale that is orders of magnitude greater than that attained by any prior medium of communication. Facebook, for example, had 2,989,000,000 monthly active users during the first quarter of 2023.²

• Virality: Not only is the quantity of information distributed on the Internet greater than that transmitted by prior media, but the speed of that distribution is far faster.³ When we speak about the virality of information on the Internet, we refer to the almost unimaginably rapid pace at which information spreads from person to person in the virtual public sphere.

• Cosmopolitanism: The scale and virality with which information spreads on the Internet renders national borders almost irrelevant. This has put immense pressure on the integrity of national public spheres. In the past, we have conceptualized public spheres as tied to particular nation states. It used to make sense to speak of the German theater, of the English press, of the French novel, or of the American cinema. But the medium of the Internet is so cosmopolitan that we can now begin to glimpse the possibility of a virtual public sphere that is truly international. Although linguistic and legal barriers have so far prevented the realization of that possibility, it is easy to predict that in the not-so-distant future the Internet may foster a true and novel cosmopolitanism.

Second, the Internet is typically accessed through phones, which have become all-purpose tools for negotiating life tasks. We use phones to get directions, to order food, to contact friends, to monitor local news, to locate partners, to charge expenses, and so on. Traditional media did not permeate everyday life in this way. We may have watched a great deal of television, or spent every Sunday morning reading the newspaper, but the virtual public sphere is now integrated with everyday life in ways that dwarf our prior engagement with the traditional public sphere. Because the Internet is seamlessly integrated into our ordinary lives, its influence has become pervasive and inescapable. Our dependence on the communicative structure of the Internet for daily life practices is something entirely new in the world.

Third, the Internet differs from previous forms of mass communication because it is interactive. In traditional mass media, speakers unilaterally addressed large audiences. We ate popcorn while we watched movies in a theater; or we read what our daily newspaper had to tell us over brunch on Sunday mornings; or we listened to our favorite TV commentators over dinner. The Internet has rendered these isolated activities almost passé. It has given rise to wholly new social media like Facebook that are built on the principle of interactivity. Social media are constructed to sustain conversations in virtual space. This unique feature of the Internet has many profound social consequences, of which I shall focus on two: the loss of epistemological authority and polarization.

Traditional media featured professional gatekeepers who vouched for the authenticity and epistemological value of distributed information. The editors of newspapers and magazines staked their reputations on the quality of the product they published. Traditional mass media were controlled by elites who created structures of communication that were quintessentially top down. Social media like Facebook, by contrast, have no equivalent gatekeepers. Facebook may use algorithms to control feeds, but these algorithms do not guarantee the authenticity and epistemological value of the information they distribute.⁴ They instead facilitate decentralized and dispersed conversations among users.⁵ Those who participate in social media are thus less like the readers of a newspaper than they are like persons who gather to converse on a street corner or around the water cooler at work.

This has potentially important consequences for the creation of epistemological authority. In prior forms of mass communication, gatekeepers warranted the epistemological authority of the news they conveyed. But Facebook features no such elite gatekeepers. The structure of epistemological authority produced on social media is more like that created in self-reinforcing circles of gossip. Some have celebrated traditional gossip because it created nodes of resistance to socially dominant ways of thinking. Gossip has this capacity because it is not just about the exchange of information. It is also about the creation of group solidarity and identity.⁶ This solidarity underwrites self-validating forms of epistemological authority. The dynamics of a gossip circle become the measure of truth and falsity.

Traditional gossip is frequently dismissed as a premodern phenomenon. In contrast to mass media, gossip requires face-to-face interactions, which seems to render gossip irrelevant in the context of large nation states whose publics stretch over millions of persons. But the Internet creates, for the first time, the possibility of large, virtual gossip groups that are connected through the medium of the Internet. This has vast implications for the social construction of epistemological authority.⁷ It fractures public epistemological authority and disperses it into competing circles of gossip. It democratizes truth.

The creation of gossip groups also has important implications for the phenomenon of polarization. Although traditional mass media often targeted discrete groups that were potentially at odds with each other, social media actually create such groups.⁸ As social media increasingly integrate the virtual public sphere into the conduct of everyday life, so does its potential to create powerful groups whose influence permeates ordinary living. These groups can endow their members with identities that empower them to negotiate the tasks of everyday life. Such groups can acquire epistemological authority sustained by the social solidarity of the group itself. Because circles of gossip define themselves in terms of the distinction between those who are inside and those who are outside, interactive social media like Facebook can foster a terrifying tribalism, homologous to that which has come to dominate our public space. The combination of polarization and democratized epistemological authority creates a toxic brew.

2.2 Democracy and the Public Sphere

Taken together, the three structural ways in which the Internet differs from all prior forms of mass communication may create new and fundamental threats to our democracy. I shall discuss six such threats. Each has important implications for how we should address the problem of misinformation. These threats are significant insofar as we deem it imperative to defend democracy. Why might we deem that important? Democracy is the only modern form of government that instantiates the value of self-determination.⁹ Democracy is therefore the only modern form of government that respects equally all persons who are subject to state authority. It is also the only form of government that seeks to reconcile the value of individual autonomy with the need for a state strong enough to perform the services required in the twenty-first century.

Freedom of speech is indispensable for democracy, because democracy consists of ‘government by public opinion’.¹⁰ Democracies were made possible when the invention of printing in the fifteenth century facilitated the emergence of the ‘public sphere’¹¹ in the seventeenth and eighteenth centuries. What we now call the ‘public’¹² emerged within the public sphere. It was created by ‘the circulation of texts among strangers who become, by virtue of their reflexively circulating discourse, a social entity’.¹³

The public sphere, and its corresponding ‘public’, are presently maintained by an infrastructure of media, like newspapers or museums, which connect strangers to each other. To speak ‘in public’ is to speak to those whom one does not otherwise know, but whom one expects to reach through mass media that underwrite the public sphere. In our own time, social media and the Internet have created a new, vast, and comprehensive public sphere that is virtual.¹⁴

What we call ‘public opinion’ arises within the public sphere. The appearance of public opinion makes modern democracies possible. The public, in the words of Michael Schudson, is ‘the fiction that brings self-government to life’.¹⁵ All modern democracies must allow for the free formation of public opinion. If they do not, they no longer serve the value of self-determination and hence no longer deserve the appellation of democracy, whether liberal or illiberal. I shall use the term ‘public discourse’ to refer to the speech necessary for public opinion formation.

We can now ask whether the rise of the Internet and of the virtual public sphere poses distinctive threats to modern democracy. The Internet surely creates many dangers for democracy, but only some are truly novel. The widespread anonymity of the Internet, for example, can be disorienting. The possibilities of impersonation, of misattribution, of inauthenticity, are legion. But anonymity, with its accompanying dangers, was a phenomenon characteristic of the traditional public sphere, and law has had a few centuries to face down whatever issues it might pose. What I shall explore in this chapter are threats to democracy that uniquely arise from the digital circulation of information on the Internet. I shall discuss six such threats. Each will have important implications for the problem of misinformation.

2.3 The Internet and Potential Threats to Democracy

The first threat concerns the loss of epistemological authority. We know that every stable society requires a convincing form of epistemological authority. This is especially true for modern society, which depends so heavily on the authority of expertise. Division of labor has made expertise indispensable for nearly all aspects of modern life, from health to technology to energy to agriculture. But in recent decades this authority has come under sustained political assault from populist movements that reject the authority of expertise. Saliant examples include the contemporary attack on universities and the remarkable resistance to public health authorities during the COVID-19 pandemic.

Existing forms of epistemological authority have in the past been undermined by new structures of communication. The invention of printing in the fifteenth century had this effect. Prior to printing, bibles were hand-copied, and so were rare and expensive. Access to bibles was strictly controlled by the church, which prohibited lay persons from reading the words of the Gospel. Epistemological authority over salvation was monopolized by the church. The printing press, however, allowed the bible to be widely and cheaply disseminated. As people began to read the words of Jesus for themselves, the church began to lose control over theological knowledge. The upshot was the Reformation and two centuries of chaos, ranging from outright religious conflict to the English revolution to the Peasant’s war in Germany. All Europe struggled to contain the antinomian consequences of the communicative revolution produced by the printing press.

The widespread and virtually costless access to universal knowledge created by the Internet has had an analogous effect on today’s society. Now everyone can be an expert on everything. They need only look up the relevant information on the Internet. If we go to the doctor, we research our symptoms and feel free to challenge the medical authority of our physician. I now see signs in doctors’ waiting rooms to the effect: ‘Your Google search is not equivalent to my medical degree.’ The zero marginal information cost of the Internet, in other words, has potentially undermined the forms of epistemological authority by which modern society underwrites order and stability. The diminishment of epistemological authority is magnified by the existence of gossip groups, which construct their own epistemological authority based on the groups’ needs for social solidarity.

The implications for democracy of these developments are obviously profound. As we lose the ability to identify figures of authority whom the public can trust to distinguish truth from fiction, we correspondingly lose the capacity to establish common facts. Hannah Arendt rightly observed that we cannot inhabit a common political world unless we acknowledge shared facts.¹⁶ Democracy cannot survive in the absence of the epistemological authority necessary to create a shared political world. We need to be able to decide whether nicotine causes cancer, or whether climate change is produced by human action, or whether seat belts reduce the harms of automobile accidents. This can happen only if we are able to create authoritative disciplinary methods to underwrite our ability to answer such questions. By unleashing epistemological antinomianism, the Internet threatens the capacity of democracy for coherent self-governance.

It is in this context that the dangers of misinformation should be understood. It requires authority, which means it requires trust, to distinguish true information from false information. There was surely a great deal of untrue information distributed by traditional media. But misinformation has become a cardinal problem of the Internet age because zero marginal information has undermined the forms of authority society uses to distinguish falsity from truth. As Ross Douthat recently observed in the New York Times in the context of vaccine skepticism, the main theory of countering misinformation ‘seems to be to enforce an intellectual quarantine, policed by media fact-checking and authoritative expert statements. And I’m sorry, but that’s just a total flop. It depends on the very thing whose evaporation has made vaccine skepticism more popular – a basic trust in institutions, a deference to credentials, a willingness to accept judgments from on high’.¹⁷

The social problem of misinformation, as distinguished from misinformation itself, concerns the loss of epistemological authority. The circulation of untrue information is a symptom of this underlying social dislocation. One can treat the symptom, of course, but the underlying disease will likely manifest itself in other ways. And if treating the symptom means using the state to suppress free participation in public discourse, it may well mean losing the patient in an effort to save it. Those who distrust ‘trusted’ flaggers or experts will distrust them even more if their objections are officially suppressed. In a democratic society, the revenge of the repressed can be a terrible thing.¹⁸ In dealing with the undoubted problem of misinformation, we must negotiate between the Scylla of widely circulating falsehoods and the Charybdis of the loss of democratic participation. Under conditions of polarization, suppression that is experienced as illegitimate can easily lead to an existential opposition between friends and enemies that would undermine the very possibility of democratic politics.¹⁹

The second danger concerns the nature of a public in a democracy. At the turn of the twentieth century, sociological theorists such as Gabriel Tarde began to distinguish between a public and what they called a crowd or a mob.²⁰ Publics were created by mass communications like newspapers or magazines. Persons read these communications and then gathered in small groups to talk about them. The information contained in mass communications spread about as far and as fast as analogue communications could spread. This gave persons time to think about the information and to discuss its implications. Crowds or mobs, by contrast, are large groups of people created in the heat of a simultaneous exposure to common stimuli. Crowds interact immediately and in ways that are typically characterized as emotional.²¹ We describe crowds using metaphors such as contagion or frenzy. We might today make the contrast between what Daniel Kahneman calls system 1 and system 2 thinking, the former being immediate and emotional, and the latter being slower and more reflective.²²

A danger that the Internet poses to modern democracy is that the virality of information might convert the demos of a democracy from a public into a crowd. The flash mob is a physical representation of this social phenomenon. Mob thinking on the Internet is encouraged because the scarce resource in the virtual public sphere is not information, as we theorize in ordinary economic modeling, but instead attention. The Internet has accordingly developed methods to drive attention, which typically emphasize the arousal of emotional reactions like anger or affection. We thus slip from reflection into immediate and emotional reactions.

Democracy requires a public that thinks,²³ and thinking does not occur instantaneously. The virality and interactivity of the Internet, its integration in real time into the pressing tasks of everyday life, may well be inconsistent with necessary public reflection. As a medium, the Internet may privilege immediate reactions that marginalize self-conscious thoughtfulness and contemplation. The result is the production of a kind of mob mentality, characterized by rapid and instantaneous responses, that is inconsistent with democracy. The circulation of misinformation thus occurs in a context that is especially fertile for its uptake and use.

A third danger to democracy that might uniquely be posed by the Internet concerns issues of scale. The vast dimensions of the Internet frequently produce forms of harm that may best be characterized as stochastic. Previously we asked whether particular speech acts might cause particular harms. The Internet has rendered this kind of inquiry almost obsolete. Speech that is simultaneously distributed to millions of persons may produce harm in ways that cannot meaningfully be conceptualized through the discrete pathways of simple causality. We must instead think in terms of the statistical probabilities of harm. At present, however, we lack any legal framework capable of assessing stochastic harms in ways that will not drastically overregulate speech. If we were to suppress every form of communication that might cause harm if distributed in sufficiently large numbers, we would have precious little communication left unregulated.

This problem is particularly acute in the context of misinformation. No society punishes statements merely because they are false.²⁴ In traditional mass media, false information is regulated when it causes legally cognizable harm, such as loss of reputation or privacy. No epistemological authority has ever been strong enough to sustain a state that might seek to suppress every false statement in traditional mass media. A fortiori this is true of the contemporary Internet. If the problem of misinformation is to be addressed, therefore, the concept of harm must be refined. But at present we lack the conceptual tools to distinguish among the kinds of stochastic harms that may be caused by the nearly infinite variety of false statements widely circulated on the Internet.

A fourth danger to democracy that might be uniquely caused by the Internet concerns the definition of the public sphere itself. Almost every sophisticated democratic legal system offers special protections to speech that is about public officials or public figures, or that is about matters of public concern.²⁵ Speech that is distributed to the public at large is often ipso facto accorded such protections, as can be seen in the unique privileges enjoyed by the press. The modern architecture of freedom of speech heavily depends upon the fundamental distinction between public and private speech. Public speech is protected insofar as modern legal systems seek to preserve the free development of public opinion that insures both public accountability and democratic legitimacy. Most states permit the circulation of a good deal more falsity in public discourse than, say, in commercial speech. Speech between private persons, by contrast, is more heavily regulated to sustain the community norms that define and maintain personal dignity and respect.

The Internet fundamentally threatens this essential legal architecture by blurring the distinction between public and private speech. Even the most personal communication on the Internet can be more widely circulated than an article of public concern in the largest newspapers. On the Internet the personal becomes public.²⁶ Yet so much about our understanding of how freedom of speech works depends upon a clear demarcation between public and private spheres of speech and action. As this demarcation is blurred by the Internet, so is our ability to conceptualize how speech on the Internet is to be categorized and regulated. This has major implications for the regulation of misinformation.

A fifth unique danger to democracy posed by the Internet concerns cosmopolitanism. All systems of freedom of speech in modern democracies aspire to subordinate political authority to national public opinion. But because the Internet produces opinion that may be cosmopolitan rather than national, it threatens to undermine this entire framework of analysis. It is not clear why the American government should hold itself accountable to Russian public opinion. That is why foreign actors spreading misinformation from abroad are accorded very different legal treatment than domestic actors who are accused of disseminating misinformation. Once the Internet transforms the public sphere into an international phenomenon, however, any such distinction will become problematic, and we shall have to develop entirely new paradigms to explain and justify our protections for freedom of speech.

I shall conclude by sketching a sixth and very subtle danger that the Internet may pose to modern democracies. This is a danger that derives from the immense scale of the Internet. That scale makes its regulation incompatible with law. The Internet is far too big to be policed through the exercise of human judgment. During the first quarter of 2022, for example, Facebook alone took down some 151,900,000 pieces of content. These removals resulted in some 2,614,400 appeals.²⁷ And these calculations do not even begin to account for the large number of communications that were not taken down but that should have been.

No court, no human legal institution, has the capacity to oversee this volume of business. Human judgment simply does not operate at this scale. Content moderation on the Internet therefore mostly does not operate through human decision-making, but instead through the application of Artificial Intelligence (AI). The upshot is that the free formation of public opinion on the Internet, which is the lifeblood of any democracy, has essentially been delegated to AI.

The problem is especially acute in the context of misinformation. It is necessary to understand facts about the world to know whether a statement is true or false. When certain falsehoods are frequently repeated and become stereotyped, it is possible to program AI to detect and remove them. But AI is a poor instrument for determining in the first instance whether statements are true or false. For such tasks we require human judgment, typically exercised in the form of law. Over the centuries, law has earned the political legitimacy and epistemological authority required to determine disputed facts about the world.

The implication of this analysis is that we may be facing a potential crisis insofar as democracy depends upon public speech, and insofar as public speech is controlled by an AI that lacks the political legitimacy and epistemological authority of law. We need, therefore, to theorize the relationship between AI and law so as to endow AI with the legitimacy and authority required to govern the Internet. How to accomplish this poses a profound puzzle.

Law carries authority because it instantiates the human capacity for judgment, which, as Immanuel Kant taught us in his third Critique, ultimately depends upon an appeal to the sensus communis of humanity. Judgment, including legal judgment, depends upon our common participation in a shared community.²⁸ As we exercise judgment, we participate in and shape the nature of that community.²⁹ Judgments are validated by the reciprocal relationship between a community and its members,³⁰ which is why judges must be representative figures to pronounce law. Judges must commit to participating in the community that their judgments establish, which is one important reason why we trust their judgments and endow them with authority.³¹

AI cannot be a member of any human community. It cannot participate in, and hence construct a dialectical relationship with, any human community. AI therefore cannot pronounce law. At most AI can report factual determinations about the way that actual humans regard law. The decisions of AI are analogous to those of a jury that seeks to evade its responsibility to determine the ‘reasonableness’ of an action by taking an opinion poll of the ambient community. Juries are not permitted to act in this way because law is not a mere fact; it does not consist in mere information. Juries are required to exercise independent human judgment for the same reason as are judges. By exercising judgment, juries both participate in and define their community. This dialectical process is the essence of law.

It follows that AI cannot make content moderation decisions with the legitimacy or authority of law. The clear implication is that society will need to invent new ways to endow AI decisions with the kind of authority and legitimacy necessary to govern the Internet. One possibility is to take advantage of the fact that AI is not static. AI learns as it receives feedback about its decisions. Because AI algorithms learn through iterative training, politically appropriate participation in this training might offer the possibility of legitimating the decisions of AI. Whether or not these means are ultimately used, it is plain that the day-to-day governance of the Internet, which is of essential importance to the contemporary formation of public opinion, will remain estranged both from law and from political legitimation until we invent some mechanism to make AI accountable. The potential dangers to democracy are obvious.

2.4 Conclusion

I have identified six potentially unique threats that the Internet may pose to modern democracies: the loss of epistemological authority, the substitution of a crowd for a public, the creation of stochastic harm, the loss of the public/private distinction, the loss of the national public sphere, and the chasm between a regime of law and a regime of AI. The problem of misinformation is implicated in each of these threats. We cannot solve the problem of misinformation until we apprehend these dangers and develop strategies for their amelioration. That is the bad news. The good news is that there is much work for all of us to do.

3 Democratic Freedom of Expression and Disinformation

Andrew T. Kenyon

3.1 Introduction

When analysing disinformation, commentators often focus on major platforms and their influence on content circulation.¹ Some also examine institutional media, especially broadcasting. Platforms and media are both relevant; both are important in the communicative infrastructure underlying public speech. Whatever the focus, there is an almost endless examination of issues and suggestions regarding what to do about disinformation. Commentary defines false or misleading information in different ways,² compares it with historic practices of propaganda and persuasion, considers the emergence of large language models and content they could generate, documents varied legal responses, and considers what should be done. Here, I examine something that is relevant to that work but often not considered directly.

My interest is how democratic freedom of expression can be understood in quite different ways, and how this can affect legal responses to disinformation. This point may be recognised more outside the United States than within, in part because of the particular US understanding of freedom of expression. The currently dominant approach to the First Amendment limits what is possible when responding to disinformation. As communicative structures change, that is increasingly significant. Freedom of expression is arguably better protected in Europe than in the USA,³ in ways that are becoming increasingly important for democratic opinion formation and a communicatively legitimate democracy. Freedom of expression is not a complete response to democracy’s communicative needs, but it is important in supporting an infrastructure of communication sufficient for self-government. Alternatively, if the freedom is poorly understood it can obstruct that infrastructure. (Even in conditions that can be called post-democratic, freedom of expression can have a similar role supporting something closer to democratic opinion formation.⁴)

This chapter considers some differences in how freedom of expression is understood in US and European law and scholarship.⁵ The differences in understanding align with different responses to disinformation and help to clarify changes that will be needed for greater US responses to disinformation. It is not that European understandings of expressive freedom are ideal. As is common in Europe, there are multiple, overlapping and somewhat inconsistent ideas about free expression across the many settings involved, and at least parts of the European Union response are framed too closely to economic interests to meet the needs of democratic freedom of expression. Rather, the EU response can tend to protect ‘market freedom against national democratic contestation’ and enable platform business models.⁶ However, the approach shows more of what may be useful overall and offers an interesting model. As all this suggests, my interest is less in the details of legal responses to disinformation than the conceptualisations of communicative freedom that frame them. That said, the analysis of communicative freedom that follows goes further than any existing legal position within Europe, whether in EU provisions, the European Convention of Human Rights (ECHR), or national laws. Rather, it draws from all these sources to suggest an approach to freedom of expression that is more than nascent in the existing European ideas and could be useful when considering disinformation or public speech more generally.

Section 3.2 below briefly examines the changed context of public speech. Structures underlying public communication have changed, they are no longer dominated by mass media in the way they were in the twentieth century, and a broader range of actors is relevant. Distribution of expression is particularly important in the changes, along with the revenues that used to accompany it. Social media and search use commercially driven distribution very differently to what was imagined for past media distribution and commercial media financing. Section 3.3 moves to expressive freedom, focusing initially on US law and commentary. Debates about legal responses to disinformation resemble other content debates, such as whether vilification can be regulated in a communicatively legitimate democracy. The overlap is only partial, but examining the issue is instructive. It highlights differences in understanding freedom of expression and, perhaps especially, what freedom means within it. This helps to clarify what interpretations of expressive freedom support what legal approaches to regulation, including regulation relevant to disinformation. Section 3.4 draws out differences in European ideas of communicative freedom and takes their implications further. Then Section 3.5 examines some current EU measures relevant to disinformation, which have interest as a general model more than in the precise wording of their provisions or how they may operate in the short term. There are significant differences in how expressive freedom is understood which, as the chapter concludes, underline the value in reformulating the US understanding and in supporting a more comprehensive European interpretation of the freedom.

There are two further matters of introduction. First, while disinformation understandably receives attention, the term can clearly be misused. Johan Farkas and Jannick Schou have noted that post-truth is often ‘a discourse crafted by those who benefit from the status quo’.⁷ As they suggest, debates about disinformation can reduce the demos in democracy and focus on rational governance by experts over people’s voices and agency.⁸ The arguments can obscure a tension between democracy as representation and expertise on the one hand, and democracy as popular sovereignty on the other. That is not my intention. Those with political, financial and other power are always likely to have conflicts of interest about freedom of speech; that applies equally for disinformation. Even so, I think there are now differences in the way speech can reach and form publics and influence opinion formation that mean something is at stake beyond other longer-term changes in democratic practices.

Second, it is useful to consider at least two aspects of legal responses to disinformation. One concerns direct state restrictions on expression and whether accountability can be implemented for at least some disinformation. Another concerns structural or systemic responses to disinformation or responses relevant to public debate more widely. These might include state obligations to regulate private actors that affect public speech, support for communicative infrastructures more generally, and intervention in relevant actors’ financial models. Particular structural or systemic responses might be hampered by freedom of expression, allowed by it, or required by it, depending on how communicative freedom is understood. And the responses can encompass institutional media as well as platforms. Both aspects – restrictions and structural or systemic responses – can be understood as falling within the freedom in freedom of expression. In simple terms, US law often says no both to restrictions and to structural interventions, while European law says yes, partly at least. In addition, beyond freedom of expression law, and beyond law more generally, a wide range of social practices is clearly significant. For news and commentary, the practice of journalism is one obvious example.⁹ Public discourse has never required only law. Freedom of expression arises within social relations; it is nonsense to say the freedom only involves the absence of restraint or that law has no relevance to, and no effect on, those social relations.¹⁰

Current judicial interpretation of the First Amendment allows hardly any restriction of falsehoods in public discourse. Even deliberate lies are protected unless they fall into certain long-established categories. In addition, US free speech law does not require various state actions related to the infrastructures underlying public speech, such as actions supporting media pluralism and democratic opinion formation. On some analyses, the First Amendment does not even allow such structural or systemic action. At a minimum, it should. In Europe, things are different. Concerns exist about laws directly regulating disinformation, but more is possible than under US law. The second aspect, about state action, has clearer differences. For example, ECHR member states have positive obligations to make Convention rights effective.¹¹ This means states have obligations to regulate private actors where they threaten enjoyment of Convention rights, including expressive freedom for both speakers and recipients. And states may need to act in support of pluralism – they are its ultimate guarantors. This helps to explain why it can be said that freedom of expression is better protected than under the First Amendment. In Europe, positive dimensions of the freedom are legally recognised, not only negative ones.

3.2 Infrastructures of Public Speech

That the infrastructures underlying public speech have changed is obvious. As one aspect, consider an individual’s ability to reach a large audience when they believe an extreme political scandal has occurred that is not being publicly addressed. The belief is an ‘outlandish fiction’ in the example. It is false. Six decades ago, the possibilities for reaching people were limited:

The only way to reach a large audience in 1965 would have been through one of the major metropolitan newspapers or the three [US] broadcast networks. But these prided themselves on practicing serious journalism. They … all were by and large determinedly centrist in orientation [in US terms], with high professional standards, and none would have aired this outlandish fiction.¹²

In the example, Larry Kramer notes US supermarket tabloids did publish outlandish fictions, but suggests they were not taken seriously by almost any readers so they did not change the overall possibilities for influencing public speech. The general situation was similar in other democratic jurisdictions, although often commercial broadcasting was far less dominant and public broadcasting more prominent than in the USA, newspaper–political party relations varied in multiple ways, and there were different habits of media reception and use.¹³ But the basic point applied generally across established democracies. Reaching an audience required media.

Many challenges that are now raised about platform-distributed speech – the treatment of problematic content, where to focus attention, what to remove from circulation or limit in reach, questions of tone, and so forth – were formerly left to journalism or what can be called a journalistic-editorial function. This function was understood to have substantial, but not total, influence on content. Such an understanding was highly idealised, but it was the model at least. Journalists and editors made decisions about news and commentary (and broadly equivalent roles existed for entertainment and other genres). Of course, reality never matched the model, with media practices often including discrimination, elitism, or being subject to commercial or state domination.¹⁴ ‘Marginalized communities have long been “reduced” by the centrism and conservatism of traditional media.’¹⁵

Law attempted to deal with those challenges, again imperfectly. One set of legal responses was not focused on institutional media but was relevant to it. These were laws that limited certain content, such as defamation and copyright. Another set of responses involved structural measures aimed at promoting pluralism and independence across and within media organisations. (There were also content-related obligations in some jurisdictions – less so in the USA – about matters such as local content, children’s content and election material.) The extent and style of the structural measures varied enormously across jurisdictions, but some responses at least aimed to lessen the marginalisation and centrism noted above.¹⁶ Thus, some legal responses involved restrictions on expression; others were structural or systemic measures. Those responses still left a lot of room. Decisions about which issues received focus, which were ignored, what tone was used, and more were decisions for journalists and editors. The decisions were supposedly independent of financial and political domination. The media was responsible for publication and made choices about content that substantially influenced public speech. While law addressed issues related to content and structures, especially under some understandings of expressive freedom,¹⁷ it also left a lot to journalism. Many challenges around content were left to speakers, and in a mass media context that meant they were left to a journalistic-editorial function.

To return to Kramer’s example, things have changed. Today people and organisations can create professional-looking content (perhaps with explicit financial, political or geopolitical aims), promote its prominence in search results, and social media can ‘spread the story’ for its own commercial ends. Sufficient platform prominence can then ‘make the story newsworthy in a way the fragmented, competitive world of mainstream media could not or would not ignore’.¹⁸ As he states, ‘[current] technologies may not have created the problem, and they are not solely responsible for it, but they have enabled it – providing a necessary (indeed, indispensable) accelerant and catalyst for a change in degree so extreme as to amount to a change in kind’.¹⁹ The greatly increased ability to find an audience helps socially beneficial speech too; the indispensable accelerant magnifies speech of many kinds. But if more speech was always to the good – as US law so often states and wider free speech analyses frequently suggest – many of the current challenges would not be emerging.²⁰

To varying degrees, distribution of news and commentary is now separated from its producers, media are dependent on platform distribution, and commercial media have lost influence within advertising markets.²¹ The degree of change varies globally and partly depends on people’s habits of media and platform use but, overall, the change seems substantial. Even so, it is not that one form of distribution has replaced the other. Institutional media still produces content and makes it available directly to audiences, often doing so with less money and reshaping content for better reach in the varied distribution methods now used. If once media was imagined as a particularly dominant influence on public discourse, it now operates within wider infrastructures of public speech. And platform distribution’s commercially driven moderation and amplification are far removed from journalism’s public interest goals – even if those goals are highly idealised, never fully met, and have evolved across decades.²² In simplified terms, the communicative structure underlying public speech has changed from a situation of journalistic-editorial gatekeepers and curators with distribution largely under media control. Now platforms have important roles in distribution. In some ways, platforms are like both (or lie between) content providers and conduits. They have similarities to both institutional media and telecommunications from decades past.²³ But platforms also have resources and powers that exceed those examples.

It is useful to think about two elements in the contemporary situation. First, there are complex interactions between institutional media and other speakers on the one hand, and platforms on the other. Content moderation and amplification overwhelmingly follow platform interests and problematic speech has gained far greater prominence – disinformation, vilification, trolling and more. Platform ‘data harvesting and data-driven, programmatic advertising are astonishingly effective mechanisms for propagating disinformation, hate, and conspiracism’.²⁴ Second, institutional media continue to have influence, more successfully at national or transnational scales than for local news in many established democracies, although in notable instances institutional media also appear to promote disinformation due to the changes.²⁵

All this means journalism does less in terms of public opinion formation and platforms do more. The degree of influence varies across different audiences and locations, but platforms now ‘shape public communication and distribute attention’ through their ‘architectures, algorithmic amplification, and content moderation’,²⁶ using very different values than the model of journalism. Platform interests have no clear relation to – they do not correspond with – the public opinion formation that is imagined as necessary in a communicatively legitimate democracy. Rather, platforms moderate content ‘in order to sustain themselves’,²⁷ and further their own position, not to create democratically legitimate structures of public communication.²⁸ Platforms’ aims, along with changes to the scale and speed of expression, are fundamental challenges to past models of public discourse and democratic opinion formation. Automation at platform scale fits uneasily with those models.²⁹ The structural changes raise challenges about content, including disinformation. While not completely new, the challenges’ form, distribution and effects are different with platform-based public speech. This means the understanding of freedom of expression has more significance and, particularly, the state’s obligations regarding communicative freedom are more important. But how those obligations are understood is quite different in US and European contexts.

Before addressing that point, I want to clarify my use of journalism. The aim is not to recall some golden past – journalism’s weaknesses have long been evident. Rather, the aim is to point out that a journalistic-editorial function was assumed, and relied on, in the public discourse imagined within law. Even the US First Amendment assumes ‘a thriving press system’.³⁰ This has allowed law to defer many decisions about the content of public discourse to journalism. That role has been partially sidelined – it is less dominant but not completely removed – which makes it unsurprising that questions emerge about the content and curation of public speech. As Robert Post has suggested, a ‘potential crisis is brewing insofar as democracy depends upon speech, and insofar as speech is controlled by politically unaccountable algorithms’.³¹ I agree that democracy depends in part on speech – a legitimate democracy is a democracy of expression, even if that is not its only requirement.³² Further, the content of public spheres may well have always needed curation. Tarleton Gillespie states that ‘it may not sit well with idealized notions’ of free expression (I would say some idealised notions of free expression) but ‘it may nevertheless be true that healthy public spheres actually require their information landscapes be curated’.³³ The need for curation is the interesting point here, rather than a deference to negative freedom under the First Amendment. In important ways, past curation of public debate was journalistic. ‘Well before the digital age, institutions that curated content and made judgments about quality and civility were important elements of the public sphere.’³⁴ In the future, that sort of role could be supported by some approaches to freedom of expression.

3.3 Expressive Freedom, Ascriptions and Fear

What is legally allowed or required by freedom of expression in established democracies differs enormously. In the US context, there is an argument that restricting objectionable expression would control the boundaries of public discourse in a manner that breaches the First Amendment. It is a well-rehearsed analysis, especially for hate speech.³⁵ As is also well known, the issues are understood differently in European law. Admittedly, some European writers take a broadly US-style approach – for example, against anti-vilification laws³⁶ – and some US scholarship has long called for a more active interpretation of the First Amendment.³⁷ But the basic position is that US law on freedom of expression forbids the control of hate speech within public discourse while European law requires it, at least for some forms of vilification. That characterisation might mislead in specific instances, but at a general level captures what freedom of expression means and the state’s role. The issues are not identical in vilification and disinformation, but the point is significant for how the obligations on a democratic state arising from fundamental or constitutional rights are understood.

Arguments against vilification laws often emphasise that democratic legitimacy is lost where states regulate public discourse. It is said that, if states control speech about matters of public concern, citizens are not free to participate in self-government as they should be able to do. For example, Ash Bhagwat and James Weinstein write that, ‘unless justified by extremely weighty reasons, rules forbidding the expression of particular viewpoints or perspectives on matters of public concern violate the basic democratic commitment to formal equality by selectively denying citizens their fundamental interest in equal political participation’.³⁸ The approach requires that public discourse, which includes much hate speech, is almost completely unrestricted. Greater restriction might be allowed within the ‘election domain’ of speech,³⁹ and is possible outside public discourse.⁴⁰

In a key US decision for disinformation, Alvarez, the Supreme Court struck down a law against falsely claiming to have received military honours.⁴¹ Lying about a miliary honour could not be penalised. The plurality opinion said the First Amendment means government has no general power ‘to restrict expression because of its message, its ideas, its subject matter, or its content’. Instead, content-based restrictions are possible ‘only when confined to the few historic and traditional categories of expression long familiar to the bar’.⁴² The categories include defamation, obscenity, fraud and expression that is intended and likely to cause ‘imminent lawless action’. Within those categories, false information can be controlled although there are still hurdles to meet, such as US defamation law’s requirement of knowing or reckless falsity in speech about public officials or public figures.⁴³ Even those who support the general US approach can criticise the plurality’s wording. For example, Weinstein would allow restriction of more falsehoods than the categories literally provide. He argues that knowing falsehoods outside public discourse should be open to restriction, and subject to less than strict scrutiny by courts. However, knowing falsehoods about ‘government, science, and history’⁴⁴ – lies within public discourse – would remain protected from general restriction. Public discourse, including much disinformation, would remain highly protected. For Weinstein, the plurality opinion did not match the public discourse focus that the First Amendment has or should have.

On this view, people have freedom to speak, judge public speech and form opinions relevant to their democratic role without (almost) any legal restrictions on public speech. Only restrictions within the few historic categories are possible, and then only if applicable tests are met. Similar to Post, Weinstein explains that the US constitution involves an ascription of rationality within public speech – an ascription of autonomy in Post’s words – which prevents state control of public discourse.⁴⁵ In addition, there is a more empirical concern that ‘government officials cannot be trusted to fairly and accurately identify and prosecute knowingly false statements in the often highly ideological context of public discourse’.⁴⁶ These two elements – constitutional ascription and fear of government action – explain a lot about US law relevant to public speech.⁴⁷

I want to draw out something that can underly such analyses. At times, the ascribed rationality or autonomy amounts to, or at least suggests, a belief that democratic freedom exists without state action. Or it amounts to a belief that state action would only make matters worse. The analysis means government cannot set the boundaries of public discourse – the speech thought most relevant to self-government – because that would prejudge the domain of public discourse itself. Thus, government cannot limit hate speech unless it also falls into one of the historic categories. And, as Weinstein and all the Supreme Court explain in Alvarez, government cannot limit disinformation about matters of public discourse, again unless it falls into one of those historic categories.⁴⁸ Outside those categories, the agenda for public debate can only be set by its participants. Public discourse’s boundaries – what is acceptable speech and what is not – should be formed through public discourse.

Various points might be raised about this understanding of democratic citizenship. I note two here. The first is whether vilification should be a category the First Amendment allows to be restricted, as allowed by free expression in Europe. The second point is slightly different, asking whether it is plausible to ascribe the autonomy needed for democratic opinion formation as existing without state action affecting speech. The action might involve limitations on speech, as in the first point, but it might also involve action against non-state entities that affect expression, and various structural or systemic supports for expression. This suggests several types of potential response as part of protecting free expression: content-related laws restricting some speech; state action against non-state entities that affect speech – here, think of platform regulation; and wider structural or systemic measures that affect the communicative environment. (There could also be positive content laws.⁴⁹) The US approach is narrower: expressive freedom generally means freedom from state action that directly restricts speech. It is not that European approaches do not believe in autonomy or self-government, but there is a different understanding of what that needs. And the difference includes how expressive freedom is understood.

As all this suggests, freedom of expression can be understood in different ways. For example, if communicative freedom is interpreted in terms of non-domination – in any of the popular versions in the neo-republican literature⁵⁰ – then state action may be required in communicative environments that overwise involve domination.⁵¹ Sufficient autonomy and freedom would not exist where there is only the absence of express state action. And if freedom is understood to have both negative and positive dimensions – to concern limitations on state interference with expression, and requirements for enablement and the effective protection and support of communicative rights – then again state action may be needed.⁵² Such action is certainly not prohibited. European examples show freedom of expression can be understood that way, to entail limited direct restrictions on speech and enablement to make it a more effective freedom rather than only a formal one.⁵³

First Amendment analyses often focus on the freedom’s negative dimensions in line with much US law. But that focus can suggest, or be taken to mean, that the freedom has no important positive dimensions relevant to public discourse. Not even US law goes quite that far in relation to speech per se.⁵⁴ But it substantially does that for the structures underlying public debate. Thus, US writing commonly suggests a legitimate democratic state cannot restrict public discourse (outside the historic categories and the tests imposed within them) and that freedom of expression does not require the state to do anything supportive in relation to public discourse, such as regulating private entities or making wider structural or systemic interventions. I note again the continuing US scholarship that takes a fuller approach to what freedom in free speech requires, as well as work that allows a more active state as a political choice rather than a matter of free speech law.⁵⁵ While important, that work does not (yet) reflect much US law.

If communicative freedom has positive as well as negative dimensions, more is involved in democratically legitimate freedom of expression than the First Amendment currently provides. Rather, the approach should examine what environment of public speech is needed for people’s democratic autonomy and what that entails for law. This is the point at which US law on free speech is highly atypical. If communicative freedom only concerns the absence of direct state interference with public speech, then much disinformation will not be responded to legally whether through restrictions or wider structural or systemic responses. Responses will be left to industry, which theoretically could do all that law might require, although history suggests it will not. Instead, people are said to be sufficiently free to participate, form opinions and act as democratic citizens where the state does not regulate the content and boundaries of public discourse. On this analysis, the overall communicative infrastructures that underlie public speech do not raise matters of free speech. Communicative structures are removed from questions of communicative freedom. As Thomas Streeter stated three decades ago, under a formalist approach to the First Amendment, ‘free speech has come to mean freedom for private communicating entities (in practice, usually media businesses) against any sort of unwanted political interference, and questions of media structure have become almost entirely disassociated from matters of free speech’.⁵⁶ The approach is widely criticised. Perhaps increasingly so⁵⁷ because the ‘communicating entities’ now extend well beyond media organisations, and the commercial interests now involved appear far less suited to structuring a democratically legitimate public discourse.

3.4 Expressive Freedom, Structures and States

Things are different in Europe. There, the freedom’s positive dimensions have many legal consequences.⁵⁸ Here I highlight two obligations they impose on states, which flow from a different understanding of what freedom involves in freedom of expression. The first is the indirect horizontal effect of rights. This justifies – and sometimes requires – that states regulate private parties who impede the effective enjoyment of rights,⁵⁹ including freedom of expression.⁶⁰ The approach encompasses the interests in expression of both speakers and audiences, or what are sometimes called the active and passive sides of free expression.

The second obligation is pluralism. While differing aspects are seen in different European jurisdictions, overall, democratic freedom of expression should aim for a situation of sustained plural public speech. This requires structural diversity, independence and transparency, all in support of individual and collective opinion formation. It should involve media with:

• structural diversity of organisations, people, funding and content, with

• independence from domination by political and economic interests, and

• transparency regarding ownership, control and content.⁶¹

As well as news and commentary, this pluralism would cover entertainment and wider cultural material.⁶² Aspects of the approach are seen in German and French constitutional decisions, European Court of Human Rights (ECtHR) cases, and decades of academic scholarship, among other places. The case law and scholarship together suggest this sort of model for diversity, independence and transparency underlying public speech. Positive dimensions of communicative freedom involve choices; there is not only one way to meet positive rights. This differs from the freedom’s negative dimensions where, in simple terms, the injunction is ‘do not restrict expression’ unless relevant legal tests are met. Here, it is whether the state has acted to support a sufficiently effective right, within the state’s real but limited range of choice.

Both these aspects of communicative freedom – state regulation of private parties and structural or systemic actions, such as those aimed at supporting pluralism – can be seen in European initiatives relevant to disinformation, which I consider in the next section. Before doing that, it is worth noting these positive obligations align with calls to focus on infrastructures rather than instances of speech when considering disinformation. For example, Clare Wardle, who has been influential in recent disinformation scholarship, has argued against focusing on the truth or falsity of each item of speech.⁶³ She emphasises the issue is systemic and concerns the narratives that platforms enable. Similarly, Alicia Wanless draws on research into information conflicts and influence operations to suggest that focusing on disinformation ‘is not useful’ and leads ‘democratic societies to miss larger problems’. It is not disinformation that needs regulating as much as ‘the ecology of the information environment more generally’.⁶⁴ Moving attention to the wider context in this way can also be linked to greater recognition of the collective or institutional aspects of free expression (or recognition of the greater impact those aspects now have). It is not only a concern with particular items of speech that affect individuals or organisations; rather, the approach involves recognising free speech as ‘a public good or value in a pluralist society’⁶⁵ and acting in support of it.

3.5 Some EU Measures Relevant to Disinformation

The general understanding of free expression outlined above is relevant to many European actions about disinformation. As András Koltay shows in his chapter, there are multiple and partially overlapping initiatives at state and EU levels.⁶⁶ Some are directly relevant to disinformation; others encompass it in part while being aimed elsewhere. The measures include national laws against certain categories of speech, such as defamation and racist hate speech. The categories that can be regulated are wider than under US law, and the hurdles to limitation are lower. In addition, there is greater recourse to criminal penalties for some categories of speech, and more scope for states to extend the categories if social harm can be seen. But for these measures there is a family resemblance to the US approach. Speech within certain categories – none of which concern disinformation as such but can include disinformation – can be subject to legal control.

As Koltay details, the EU is also developing multiple other measures.⁶⁷ For present purposes, this can be illustrated by the Strengthened Code of Practice on Disinformation,⁶⁸ the Digital Services Act (DSA)⁶⁹ and the proposed European Media Freedom Act (EMFA).⁷⁰ The measures together reflect positive and negative dimensions of democratic free speech, and they do things consistent with the positive dimensions of free expression in a way that is harder to imagine in the USA. The measures include indirect horizontal requirements on private actors, such as DSA obligations on platforms, and positive state actions, such as those supporting media pluralism in the EMFA. The emergence of this approach makes more sense ‘when one recognises the broad European understanding of free expression. Together, the Code, the DSA and the EMFA suggest some ways in which freedom of expression may evolve in situations where control and curation are not dominated by a journalistic-editorial role.

How each of the three measures concerns disinformation could be analysed at length. Each is substantial; for example, the Code runs to 20,000 words and includes 44 commitments and more than 100 measures about how commitments will be met and reported. But outlining just some aspects suffices here. My interest is not so much the requirements’ strengths and weaknesses – there are substantial doubts about how effective this iteration of measures will be. Rather, it is to understand the responses as a model that may well be consistent with freedom of expression, arguably one better suited to the current context than an overwhelming focus on negative dimensions of expressive freedom.

Some things needed for sustained plural public speech remain similar with changed communications. At the level of media organisations, this includes having multiple organisations with varied financial bases and missions, varied staffing, management and participation, independence from political and commercial domination, and transparency of ownership, control and content. The EMFA aims to support some of these needs uniformly across the EU. Among other things, the proposal addresses the rights and duties of media services and recipients, including ‘the right to receive a plurality of news and current affairs content, produced with respect for editorial freedom’.⁷¹ It aims to safeguard the independence of public service media.⁷² And it structures some interactions between very large online platforms and media services, including a requirement for good faith ‘meaningful and effective dialogue’ when an editorially independent media service considers a platform is frequently restricting or suspending its content.⁷³

At the time of writing, the EMFA faces ongoing debate within the European legislative process,⁷⁴ and some strong criticism.⁷⁵ Here, I leave aside concerns about some aspects’ plausibility and suitability,⁷⁶ and significant questions about its legal basis – the EMFA arguably covers matters within the powers of member states not the EU.⁷⁷ In addition, while trying to do more to support democratic needs the EMFA has a market focus. This is unsurprising given EU powers, but it is awkward given the aims for media freedom and does not sit easily with some member state approaches. Consider, for example, French and German judicial statements that the freedom should not be ‘made the object of a market’ and that ‘Market opportunities may be a question of economic freedom, but not of freedom of opinion’.⁷⁸ However, whether taken at EU or member state level, my interest is the types of action and how they are understood as being consistent with freedom of expression. The EMFA illustrates a desire to reinforce important aspects of plural and independent media in the contemporary communications context as one part of protecting democratic freedom of expression, within a broader set of responses affecting disinformation.

Unlike the EMFA, the DSA has passed. It includes a range of obligations for platforms, and further requirements for very large online platforms and very large online search engines (often labelled VLOPs and VLOSEs).⁷⁹ Among other things, these very large actors need to assess and mitigate systemic risks posed by their systems. The DSA does not explicitly mention disinformation in its Articles – the term appears repeatedly in the Recitals – but disinformation is clearly intended to come within systemic risks. Under Article 34, risk assessment needs to be proportionate to the severity and probability of systemic risks, and include these systemic risks:

1. (a) the dissemination of illegal content through their services;

2. (b) any actual or foreseeable negative effects for the exercise of fundamental rights, in particular the fundamental rights to human dignity, … to respect for private and family life … to the protection of personal data, … to freedom of expression and information, including the freedom and pluralism of the media …

3. (c) any actual or foreseeable negative effects on civic discourse and electoral processes, and public security;

4. (d) any actual or foreseeable negative effects in relation to gender-based violence, the protection of public health and minors and serious negative consequences to the person’s physical and mental well-being.⁸⁰

So systemic risks related to illegal content, fundamental rights, civic discourse and electoral processes, public security, gender-based violence, public health and more need to be assessed. Assessments need to consider the effects of services’ recommender systems (and other algorithmic systems), content moderation systems, advertising systems, data practices, and terms and conditions.⁸¹ Article 34’s wording was strengthened during the legislative process so media pluralism is now explicitly mentioned, and negative effects on civic discourse and electoral processes need not be linked to intentional manipulation of services.⁸² While the words look powerful in some ways, there are many doubts about their application in practice.⁸³

As well as assessing systemic risks, very large actors need to mitigate them by putting ‘in place reasonable, proportionate and effective mitigation measures, tailored to the specific systemic risks identified … with particular consideration to the impacts of such measures on fundamental rights’.⁸⁴ Article 35 sets out a large list of possible mitigations, including making changes to services’ design, features or functioning, terms and conditions and their enforcement, content moderation, recommender systems and advertising systems. Very large actors are also subject to regular independent audit (which may end up focusing on compliance with obligations to conduct assessments rather than assessments’ validity).⁸⁵

Direct state action against disinformation is not being taken here. Rather, influential private entities need to assess and mitigate systemic risks posed by their systems. The requirements go beyond illegal content.⁸⁶ They seek to regulate aspects of content curation. Of course, very large actors are likely to use restricted interpretations of risks and mitigation, subject to regulators, courts or legislatures demanding more.⁸⁷ But my point is that the model appears broadly consistent with European ideas of free expression. Or, at least, it is seeking to find a way to deal with harmful but legal content that furthers the multiple interests of multiple actors in freedom of expression. Clearly, many of the challenges linked to communicative freedom are hidden within Article 35’s brief reference to mitigating systemic risks while also considering the effects on fundamental rights. Multiple parties are affected in terms of freedom of expression, and it is far from clear how ‘particular consideration’ will occur. That is just one reason to expect future legal disputes drawing on free speech arguments about exactly what the provisions mean, how platforms apply them, how regulators enforce them, and whether regulators are sufficiently independent. There are also challenges posed by individual entities assessing and mitigating risks which may be broader than the risks seen by any one very large actor. And the approach does not challenge platform business models.

This is unsurprising given the EU’s internal market powers, but the problematic effects of platforms’ data-based business models have been raised repeatedly in commentary. If freedom of expression includes freedom from market domination of public expression – a clear element of European approaches in general – it is not obvious how that is possible if commercially driven platforms come to dominate public speech. Indeed, this appears to be a central weakness in the approach in terms of democratic freedom of expression.⁸⁸ I would agree with Rachel Griffin that European traditions of public service media are based on the judgment that ‘wholly privatised broadcast media systems cannot serve the public interest even if they are well-regulated’, which makes it ‘inherently problematic that online media are governed by profit-driven conglomerates, even if they are subject to human rights obligations’.⁸⁹ However, the general idea of requiring private actors to do things related to the content they distribute is important. In some ways, it is seeking to model the curation of public speech once provided, in theory at least, by a diverse media environment and its journalistic-editorial function. Doing something like that – even if doing it better than this initial attempt – appears to be part of the future of democratic freedom of expression.

The Code is a more detailed catalogue of what platforms (and others such as advertising brokers) might do about disinformation, particularly electoral and issue-based political advertising. A handful of commitments illustrate the Code’s approach. These include:

• defunding the dissemination of disinformation and improving policies, systems and controls over content monetisation;

• preventing the use of advertising systems to disseminate disinformation in the form of advertisements; and

• adopting a common definition of ‘political and issue advertising’, clearly labelling it as paid content and providing ‘clear, comprehensible, comprehensive information about why’ it is shown to users.⁹⁰

There are also system design commitments to minimise the viral spread of disinformation, which might include recommender systems improving ‘the prominence of authoritative information’ and reducing disinformation’s prominence ‘based on clear and transparent methods and approaches for defining the criteria for authoritative information’.⁹¹ The Code does not specify those criteria, although it includes commitments to work with fact-checking organisations.⁹² The Code is intended to be recognised under the DSA, so complying with it could be relevant for very large actors showing they have made appropriate mitigation efforts. While the Code is sometimes seen as a form of co-regulation or partially enforced self-regulation – with the European Commission influencing its development and revision⁹³ – membership is entirely voluntary, and the Code is not directly enforceable. For example, when Twitter since renamed X) withdrew from the Code in 2023 reports suggested this could end ‘a headache for the Commission, given the platform’s lack of compliance’ with the Code and the Commission’s lack of power to ‘force them out’.⁹⁴ Overall, the Code sets varied objectives, with their meaning and implementation substantially left to platforms, alongside continuing involvement of various EU-linked organisations, and undertakings about reporting and research. In comparison, the DSA could have greater influence if it is strongly applied, but all three initiatives are useful to understand the multifaceted EU approach.

A related matter is how truth is determined in these measures. The Code and DSA leave platforms to determine what amounts to disinformation, in the first instance at least. DSA powers of regulators and courts may lead to those determinations being overruled. But what was once deferred to journalism is now in part deferred to platforms and fact-checkers (suggesting the value in examining the fact-checking organisations).⁹⁵ Of course, platforms cannot perform that public speech role in the same way as journalism, whether journalism’s actual practice or its idealised role. As well as platforms’ data- and attention-centred commercial focus, the speed and scale of platform-distributed speech means content decisions cannot be made as they are for journalistic content in institutional media.

Paolo Cavaliere has examined legal standards used to assess truth,⁹⁶ comparing ECtHR decisions involving journalism and EU advertising rules. While the ECtHR is not an arbiter of truth in general, it distinguishes truth and opinion in journalism cases. When assessing public interest expression, it uses a journalistic or process-based understanding of truth. It is not that a particular fact has necessarily been proved true; rather, appropriate journalistic practices provided sufficient certainty about the material before publication. In effect, the ECtHR relies on a journalistic-editorial function for truth. In contrast, advertising laws focus on a publication’s misleading quality, which need not necessarily involve falsity. While ‘respect for appropriate procedures can exonerate even factually false statements’ in journalism, the CJEU ‘has ruled out … a professional diligence defence for misleading advertising’.⁹⁷ Cavaliere suggests the DSA and the Code lean more towards an advertising standard, but the ECtHR requires political and public interest content (whether paid or unpaid) to be assessed on a journalistic standard.⁹⁸ Perhaps the outcome of this apparent conflict will depend on the approaches of fact-checking organisations, which might mirror traditional journalistic aims in some ways at least. If so, that would lessen a conflict between the EU provisions and ECHR protections for freedom of expression (with which European Charter provisions should be consistent). If not, EU aims could be frustrated because platform ‘content moderation decisions will … be assessed for their compliance with the level of protection of freedom of expression required by the ECHR’.⁹⁹

3.6 Discussion and Conclusion

Analyses of the First Amendment and disinformation can emphasise two points – a constitutional ascription of autonomy existing without state action and a fear of state action. Other approaches to democratic free speech do not follow the first point in the same way. To me, they are more attuned to the reality of democratic choice. They see state action as ubiquitous – the state is always involved in one way or another in ordering, enforcing or tolerating private actors that affect speech.¹⁰⁰ And this links to the state’s positive obligations to protect freedom of expression, to attempt to make the freedom real and effective.

However, the second point about fearing government action is relevant beyond the USA. In Europe, there is clearly not the same fear of government action as in US law. That would be impossible given the understanding of freedom of expression, but at the same time there is understandably no great appetite for state determinations of truth. This can be seen in the reluctance of the ECtHR to support limits on false speech, and the way in which measures like the DSA and the Code mean that states – whether executives, parliaments or courts – are not determining truth directly. The EU responses to disinformation involve legislating ‘for more societal responsibility for very large online platforms’ while leaving platforms to decide ‘if and how to deal with any systemic risks to freedom of expression’.¹⁰¹ The European measures do not outlaw false facts as such. For example, only Holocaust denial is clearly established as a false fact that receives judicial protection in ECtHR decisions.¹⁰² Other historical debates are allowed. This leads Irina Katsirea to suggest the ECtHR would not act as arbiter of truth in general, and criminalising disinformation in news appears problematic under Article 10 of the ECHR.¹⁰³

In a mass media context, there were important ways in which the legal approach to freedom of expression relied on journalism. The journalistic-editorial function decided much about the content of public speech, and where and how public attention would be focused. Those decisions were made at a particular scale and speed, with the relevant actors considering individual pieces of content. Now, legal and regulatory approaches seek to deal with significant changes, and freedom of expression is highly relevant to what can and should be done. Admittedly, the versions of freedom of expression outlined above are somewhat exaggerated, and the European model goes further than existing law. However, the law and ideas about expressive freedom seen in Europe suggest the model that I have sketched. The idea of positive obligations flowing from fundamental rights is clearly established and it suggests how current European initiatives have consistencies with important aspects of free expression. The EU measures are not as explicit as they should be in relation to the multifaceted quality of freedom of expression – ways in which interests are held individually and collectively across speakers and recipients. And much of the complexity is submerged in the current approaches, such as the DSA’s requirements for very large actors to mitigate systemic risks with due regard to fundamental rights. How to do both those things at once – mitigate risks and pay regard to fundamental rights – involves subtle evaluations. Whether platforms can do that is questionable, perhaps especially for platforms imbued with US ideas of rights and the broader scope that is needed here for rights moving beyond individual interests. It may mean that very little is done. As well as the very large actors, the European Commission, with ‘relatively little institutional experience in freedom of expression principles’,¹⁰⁴ faces regulatory challenges.

However, the models of communicative freedom and the illustrative responses to disinformation in Europe suggest how different dimensions of freedom of expression might be supported in a changed context of public communication. Details of the Code, the DSA and the proposed EMFA are less significant than the ideas they illustrate. US analysis, in comparison, is left looking for technical arguments to adapt earlier regulatory approaches, such as cable access provisions or telecommunications regulation, or sidestep the First Amendment, or remain reliant on platforms’ leeway under US law to moderate content almost without restraint.

That free speech law has long relied on journalism reflects longstanding practices of who decided about content and the scale and speed of decision-making. Platform moderation is not the same. It applies complex, networked decision processes at a scale and speed far removed from what has often been imagined for rights such as expression. It is systemic, and so are many current legal responses. But in being systemic, measures like the DSA still leave a lot of discretion to platforms. They face challenges in mass moderation,¹⁰⁵ including how, if at all, public values can be included in recommender systems.¹⁰⁶ As Gillespie has observed, ‘platforms now function at a scale and under a set of expectations that increasingly demands automation. Yet the kinds of decisions that platforms must make, especially in content moderation, are precisely the kinds of decisions that should not be automated’.¹⁰⁷ While that raises conceptual questions for freedom of expression, a fuller approach to freedom – one that understands its positive as well as negative dimensions – can go some way towards accommodating the changes.

My aim has been to explain something about why US approaches take the form they often do, and to highlight how different approaches to freedom of expression are possible. Much comes down to an understanding of freedom, which in turn relates to the state’s role. The differences are not minor. The ‘conceptual and normative differences between a US-oriented, First Amendment tradition and a positive rights, European and international human rights perspective’¹⁰⁸ should not be underestimated, with ‘differences in underlying expectations on governmental involvement in speech regulation’ being ‘fundamental’.¹⁰⁹ While some past commentators have found the differences problematic, they now appear increasingly problematic for democratic freedom of expression. Communicative freedom has positive and not only negative dimensions. This means there are obligations on states, which can include measures to tackle disinformation, though not necessarily by direct prohibition. The state obligations are challenging to implement and may be abused. But believing in a purely negative freedom without support or enablement seems at least as problematic or, given current infrastructures of public speech, more so.