Transnational organizations and hegemonic donor states such as the United States have concerned themselves in recent years with building the state capacity of developing and transitional states. The United States made efforts to develop the military and police capacity of postwar Iraq (after tearing down its prior version of those institutions) (Gates Reference Gates2010). The European Union has made efforts in Eastern Europe to build anti-corruption capacity (Börzel and van Hüllen Reference Börzel and van Hüllen2014). There have been substantial global efforts to build state capacity in Somalia (Menkhaus Reference Menkhaus2014). At least to some extent, these efforts are motivated by the self-interest of those leading the capacity development efforts – the more functional Somalia’s state is, the more it can control the piracy that harms its neighbors. And it may be that a functional Somali state would be better at controlling piracy than external states trying to do so themselves, for example, because of its superior local knowledge or sociological legitimacy. In other words, the enterprise of capacity development is often predicated on the ideas that governments that are unable or unwilling to control their own citizens create negative externalities for the rest of the world, and that the rest of the world is unable to control those behaviors directly, without helping the state with direct access to the people causing the problem to build the capacity to intervene.

This chapter argues that, from the standpoint of government regulators, analogous incentives exist with respect to platforms. Governments have strong reasons to control much of the pathological behavior that appears on networked internet platforms, from commercial fraud on Amazon to fake news on Facebook (although the example of fake news also illustrates that government reasons are far from unequivocal and may be outweighed by other interests, such as individual freedom). And it is likely that governments may be ineffective at directly regulating some of this behavior, for example, because of the tendency of platforms and their users to span jurisdictional boundaries, because of the property rights of owners of those platforms, or simply because of the bureaucratic burden of regulation without direct access to the immediate site of behavior in the form of platform code and interactional surfaces. Such governments accordingly have reasons to promote the development of a kind of analogy to state capacity in platforms in order that they may require or encourage those platforms to control behavior that appears via their infrastructure.

2.1 Why Is Conduct on Platforms So Hard to Regulate?

Some of the ways in which many commentators worry about externalities caused by platforms represent what we might think about as pure conflicts of interest between those platforms and society at large. Social media companies tend to profit by collecting large amounts of information about user activity and using that information for ad targeting. That is (arguably) bad for society as a whole, insofar as it allows advertisers to manipulate consumers and leads to the risk of dangerous third-party data leaks. But it’s not bad for companies (the data leaks are – but the manipulation is precisely the service they offer to advertisers). Similarly, “sharing economy” transactional platforms tend to make money by evading regulations governing services like taxicabs and hotels as well as workplace law in general. Again, that’s a problem for the rest of us. Under such circumstances, the best we can do as a society is use the legal system to simply coerce companies to act in ways that are less self-serving and more public-serving by, for example, requiring them to count their drivers and delivery people as employees rather than contractors or prohibiting them from sharing user data between different corporate functions.

By contrast, other kinds of external harms – the sorts within the domain of interest alignment described in the Introduction – likely also harm platform company interests, at least under certain interpretations of those interests. The clearest examples involve various kinds of fraudulent behavior which can harm companies’ long-run interests in running platforms that their users can trust for reliable information. For example, viral disinformation on social media undermines the implicit value, at least over the long run, that those users can attribute to information found on those platforms. The same goes for counterfeit products and fraudulent ratings on transactional platforms.Footnote ¹

With respect to such harms, it is a genuine puzzle (which this book aims to address) as to why the companies cannot simply stop them. This puzzle is highlighted by the fact that there are important senses in which technology platforms have vast governance advantages over physical world governments.

First, unlike physical world governments, company personnel can observe anything (though not everything at once) that happens under their control (with the sole exception of end-to-end encrypted platforms like WhatsApp – but even there, the encryption is itself a platform design choice under company control). They need not install surveillance cameras and peek through windows, they simply need to query a database. Such observation is not costless – it still requires paying personnel to actually examine some piece of on-platform behavior (a product for sale, a tweet), but the costs are substantially lower per observation than for states to send police to follow people around, get search warrants, and so forth. Moreover, some (but not all) “observations” in the platform context can be automated by, for example, deploying artificial intelligence to identify certain kinds of conduct. While artificial intelligence and other automated methods of observation are no panacea, they have certainly proven effective in a number of important cases. For example, companies have devised methods to automatically fingerprint and then identify duplicates of known child pornography images so that they may be removed from platforms without any human having to look at them.Footnote ² It is also possible to have a kind of human-in-the-loop observation where artificial intelligence identifies behavior to be reviewed by a human. Governments can automate enforcement too (speed cameras exist), but governments still have to expend costs to translate behavior in the physical world into data, whereas companies have data all the way down.

Similarly, many forms of rule enforcement are, at least in the abstract, much cheaper for platform companies than for physical world governments. The most common tools of enforcement in the platform context are various forms of restriction or removal either of user accounts or of the specific conduct of user accounts (such as posted content). But, once again, these enforcement techniques do not require the building of prisons or the operation of courts, or the sending of heavily armed police officers to people’s houses – they simply require the design of software to create affordances for enforcers within companies to alter user privileges and content, and then the operation of those tools – the click of a button within some internal system to change a field in a database.

But the relative ease of governance of platforms is to some extent an illusion, for platform companies are still subject to two core abstract-level problems that they share with physical governance.

The first is the difference between observation and knowledge, namely interpretation or what I will describe in Chapter 3 as “legibility”: While platform companies can “see” any behavior on the platform by querying a database, interpreting that behavior, and in particular, determining whether that behavior is harmful or harmless, compliant with company rules or noncompliant, is much more difficult. In many ways, this interpretation is likely to be more difficult for platform companies than physical world governments, for platform companies – particularly the most successful companies at the greatest scale – have to deal with a vast amount of behavioral diversity and novelty, and are much more socially and culturally distant from many of their users. A bunch of lawyers and engineers in Menlo Park are likely to have a much harder time understanding the behavior of people in, say, Luxembourg or Kenya than the local government of a municipality composed of representatives of the people they govern.

The second is the problem of incentives. Not every person within a company’s decision-making structure may have the incentives necessary to actually govern problematic behavior. Personnel whose job is to keep government officials happy may be more vulnerable to external pressure by politicians who benefit from disinformation. Engineers and product designers on “growth” teams who are rewarded for increasing activity may have an incentive to build features that facilitate all activity, whether those are harmful or helpful to a company’s long-term interests. Stock markets may give executives an incentive to permit behavior that promotes short-term profits at the expense of long-term profits.

Those two problems are, respectively, the subjects of Chapters 3 and 4 of this book – I contend that they are the core challenges that vex the enterprise of platform governance in the domain where company interests (properly understood, from a long-term perspective) and social interests are aligned.

2.2 Why Build Capacity?

The global community has long wrestled with the problem of states that are unable or unwilling to regulate the harms that users of their territory inflict on the rest of the world. Piracy is a classic example: When a state fails to adequately patrol its territorial waters, it can become a haven for pirates who then prey on international shipping (e.g., Pham Reference Pham2010). But there are other kinds of criminal activity that can result from governance failure (or governance unwillingness) and can predominantly affect people outside a given country, ranging from terrorism to intellectual property misuse and the cottage industry of internet scams in certain towns in Eastern Europe (Subramanian Reference Subramanian2017; Bhattacharjee Reference Bhattacharjee2011).

This problem may arise because a country lacks sufficient ability to fully regulate behavior on its territory. Maybe, for example, the government lacks legitimacy with the population, or the funding to train and deploy adequate security forces. Moreover, it may lack adequate incentive, either simply in virtue of the fact that much of the cost of the harm is externalized or because leaders have mixed motives, as when the criminals pay bribes to continue their activity. The piracy example is again apt: Percy and Shortland (Reference Percy and Shortland2013) have plausibly argued that even improvements in state capacity are ineffective in controlling some cases of modern piracy in view of the fact that the activity provides local economic benefits and its harms are externalized.Footnote ³

Nonetheless, it may be difficult for other states to directly regulate this behavior, even though it inflicts harm on their citizens or in their territories. They may not, for the simplest example of a reason, have their own physical access to the territory where the international harms are being generated. And they may lack sociological legitimacy with the people of that territory, even more so than an unpopular domestic government – it should be unsurprising, for example, that the people of Somalia would be unwilling to be governed by the United States and France.

Accordingly, a core strategy that other states have adopted to address the problem of externalities generated by under-governance is to attempt to assist the state which is failing to govern in building or rebuilding its capacity to do so, and providing it with incentives to use that capacity. On our running example, Bueger et al. (Reference Bueger, Edmunds and McCabe2020) describe maritime capacity-building efforts in various nations in the region to address piracy from Somalia. Similar efforts have been made in order to improve the ability of countries to fight international terrorism within their territories (e.g., Bachmann and Hönke Reference Bachmann and Hönke2010).

Observe the incentive structure underneath such capacity-building enterprises. Under-governing states create certain kinds of external harms in view of the fact that they fail to control the behavior of people operating from their territory. For convenience, I label those kinds of harms governance failure externalities. Moreover, such states not only fail to control harmful behavior, there’s an important sense in which they provide resources with which individuals may carry out that behavior, and hence provide an incentive for that behavior. The most obvious example of such resources is a headquarters – by allowing (deliberately or inadvertently) pirates or terrorists or international fraudsters or whomever to operate from their territory, such states shield them from the direct control of others – Westphalian notions of sovereignty impose a barrier to more effective states directly acting to control the harmful behavior. Another related example of a resource that failed states provide is access to victims – by failing to control an important region of the ocean located among an important trade route, a state effectively provides pirates with a known location in which victims may be found. Regardless, the point is the externalities: other states, and international organizations, have self-interested reasons to provide capacity-building assistance and incentives to lower the cost of the home state regulating domestic behavior with negative spillover effects.

I do not wish to describe this problem in a way that buys into problematic US-centric or developed country-centric representations of global problems and responsibility. Wealthier and more powerful countries can also impose governance failure externalities on the rest of the world. For the most obvious example, the failure of the United States to control the carbon emissions emanating from its territory imposes vast harms on the rest of the world. In terms of long-run global welfare, greenhouse gases are far more damaging than pirates. In principle, it would make sense for the rest of the world to help us Americans control our polluters. Unfortunately, given the vast power and wealth disparity between the United States and most other nations, our troubling belief in American exceptionalism, and the fact that our governance failures are built deep into our polarized politics and ineffective constitutional institutions, there’s probably precious little that other countries could do to help us.Footnote ⁴

Platforms create precisely the same kinds of governance failure externalities. Here, some care is in order: a corollary of the point which I have repeatedly emphasized that only some platform harms are within the domain of interest alignment with society is that not every harm that platforms impose on the outside world is a governance failure externality. Sometimes, the platforms themselves do things that directly harm third parties – such as when they violate people’s privacy or squash their competitors. But other times, the externalities come from the independent behavior of users making use of platform affordances.Footnote ⁵ In the platform context, the most obvious examples revolve around various kinds of fraud or confusion – when viral disinformation circulates on social media, for example, or when counterfeit N95s plague Amazon. And much like in the international context, the affordances of platforms both facilitate this behavior and make it harder for governments to control (primarily by increasing its quantity and speed, as well as cross-border impact).

More specifically, the relative inability of governance “donor” states to act directly on the people of recipient states is replicated in the platform context. As a whole, companies tend to have both access to more information about user behaviors and access to a wider array of tools (such as major interventions on their own platform design) to regulate them than do governments. By contrast, even though ordinary platform users would presumably prefer to be able to use their market power to mitigate some of the harms they receive from platform companies (especially if that sometimes means, e.g., accidentally buying fake N95s in a pandemic, or being deceived by political lies), governments have much more leverage over platform companies than the users of those platforms do, in virtue of the fact that governments, of course, have more coercive tools available (they can freeze assets and lock up executives). Users have very little power over platforms themselves (their only viable option is exit, network externalities lead to very high switching costs for individuals, and mass coordination is costly).

This suggests that a regulatory capacity-building approach dedicated to the goal of empowering companies to protect the interests of users and society – which this book argues will also require empowering ordinary people to intervene on platform choices – is an appropriate way to think about how we, as citizens of democracies wishing to control our environments, ought to bring ourselves, as users of these platforms, under control. Of course, there are dangers as well to the project of capacity building – which I will address below – most obviously the rightful fears that many of us have about enhancing the power of unaccountable private platform operators over behavior in a context where increasing shares of important human interaction happen via those platforms. But only when we enable ourselves to think directly about governance capacity do we even become capable of properly raising those questions.

2.3 How to Build Capacity?

In the abstract, it is helpful to think about the enterprise of capacity building as giving a putative governor (as a first pass, a platform company – but the remainder of the book will call into question the notion that it is companies in particular who should be doing the governing, as opposed to a combination of companies, workers within companies, civil society, perhaps governments, and most importantly ordinary people) both the incentive and the tools to control unwanted behavior.

In the context of states, relevant examples of such capacity-building efforts include the enterprises of rule-of-law development and of security assistance. Powerful states and their associated NGOs have offered less powerful states a variety of tools for more effectively extending their authority over their territories, including military training and arms, legal personnel training, and building of physical infrastructure for legal administration such as courthouses; of particular relevance for the argument to come, some of these resources (such as training) have also been extended to civil society actors within recipient countries rather than governments directly, with some success in achieving the goals of legal development (Blattman, Hartman, and Blair Reference Blattman, Hartman and Blair2014). Rule-of-law development efforts also have intervened directly on state incentives, such as with constitution-drafting projects; additionally (and rather more controversially) economic development programs directed by entities such as the World Bank and the International Monetary Fund have also aimed to intervene on state incentives with “structural adjustment” programs which impose obligations on states to do things like privatize public services as a condition for receiving international loans.

Governments have analogous tools with respect to companies. The United States or Europe could, for example, impose changes on the corporate structure of platform companies or on the ways that decisions are made within that structure by alterations to corporate law. In other corporate contexts, we have seen examples such as the Sarbanes-Oxley Act, enacted after the Enron scandal, which imposed new requirements on the boards of directors and outside auditors of public companies. We can understand this as directly analogous to (but hopefully more coercive than) the constitution drafting assistance offered by the world’s liberal democracies in places such as Afghanistan. Foreshadowing the proposals articulated in Chapter 6 as well as this book’s Conclusion, governments could also directly insert third parties into the governing structures of companies, for example, by mandating certain kinds of representation on corporate boards, by giving third parties a veto on some company decisions, or even by directly acquiring company stock and exercising decision rights on behalf of empowered third parties. This too is a governance tool with which governments are familiar. For example, Article 27 of the European Charter of Fundamental Rights provides that workers have a right to “information and consultation.” As implemented by an EU directive, workers are entitled to that “information and consultation” on, among other things, “decisions likely to lead to substantial changes in work organisation.”Footnote ⁶ Member states have further implemented this right by providing for worker consultative bodies within companies.Footnote ⁷

While there are limits, such as the American constitutional protections for property (and, in the case of social media companies, free speech), the very brief survey above should suffice to illustrate that governments have the tools to intervene quite deeply into the decision-making processes of platform companies. In view of the significant (and legitimate) coercive power that countries exercise over those companies, we also have reason to think that the use of those powers for purposes of capacity development has greater prospects of success than capacity-building projects in the context of states.

This book concerns itself with the enterprise of governance in the abstract, that is, with the kinds of institutions (organizational forms) that ought to be adopted in order to permit platforms to effectively govern (and permit us to effectively govern them). In the context of the analogy to governance development in the international community, this is analogous to saying that the scope of concern should be the promotion of institutions like democracy and the rule of law, rather than the prescription of the particular substantive legal and economic rules which recipient countries ought to adopt.

The institutional approach recognizes that the concrete rules that platforms – or even platforms in specific sectors, such as social media – ought to adopt will potentially vary widely across platforms, across underlying social contexts, and across time (just as with countries). Moreover, many of the real-world rules problems facing platforms are sufficiently contested and potentially irresolvably difficult in their form as rules, even if there are correct answers to them. What, for example, is the normatively correct position for a platform to take on disputed claims about treatments for COVID-19, or allegations about the behavior of a US president and their family, for example? It seems to me overwhelmingly likely that the best answer to this category of problem will operate in a combination of the domains of what Rawls (Reference Rawls1999, 73–78) called “imperfect procedural justice” and “pure procedural justice.” That is, that the answer to the question will focus on how the decision is to be made (and, by extension, who is to make it), rather than what the decision is to be, sometimes because the procedure is our best albeit imperfect way to get to an externally defined right answer, and sometimes because whatever answer follows from the procedure (as with democratic policy choices) is for that reason the right one.

Another component of the institutional approach adopted in this book is the supposition that the concrete interventions that are likely to be effective in platform governance are more in the domain of product design in terms of the high-level affordances available to users as well as in tools and methods of governance, rather than in low-level enforcement details or volumes of work carried out (such as the number of person-hours available to do content moderation). Sahar Massachi, a former Facebook software engineer in Civic Integrity and one of the founders of the Integrity Institute, sometimes refers to this as the difference between adding more traffic cops and adding speed bumps to the roads.Footnote ⁸

With respect to social media platforms, free speech ideals also place important constraints on the scope of potential government regulation. While few countries have free speech traditions quite as strong as those of the United States, the value is recognized in every liberal democracy and by international human rights law. Under US law, it would be impossible to directly restrain many of the pathological speech phenomena on social media (or to force the companies to do so), such as by banning the spread of conspiracy theories and so-called “fake news” or racist speech. Countries with hate speech regulations and the like, of course, might be able to go further, but even there – as with most efforts to regulate widely dispersed behavior via law – the practical difficulty of such regulations and the cost of enforcement is widely recognized. The same point holds with respect to transactional platforms. If, for example, states and municipalities really wanted to directly regulate every person who offered up a room on Airbnb, well, they could do so, but the effectiveness and practical scope of those regulations would be limited by the sheer number of people involved and the difficulty of the government directly monitoring so much behavior.

By contrast, regulations of the decisional infrastructure of technology companies themselves, as well as their design, and the affordances they offer – as the institutional approach recommends – are somewhat more insulated from free speech challenges insofar as they do not directly target expressive activity and could also apply to non-expressive activity. To be sure, regulations of companies are not without limits. In the United States, it would probably be unconstitutional to impose something like a “fairness doctrine” directly on social media companies, requiring them to engage in politically neutral content moderation (the misguided views of the State of Texas and the Fifth Circuit in a recent bill and case notwithstanding). It would also be unconstitutional, although somewhat more difficult to prove in a specific case, to impose regulatory disadvantage on them in ways motivated by political viewpoint, such as various Republican proposals to abolish Section 230 of the Communications Decency Act in retaliation for companies’ interpretation of their rules to remove conservative speech from their platforms. But other sorts of institutional interventions are free from such constitutional barriers. For example, many interventions on the workplace structure of companies, such as a proposal given in the Conclusion to this book to give more job-related power to the offshored contractors currently serving as content moderators in social media companies, would not be directly related to speech at all. Also not directly related to free speech would be the use of antitrust law to break up companies with excessive market power (discussed later in this chapter). Similarly, government efforts to assist the voluntary building of user and public-dominated governance institutions for companies through industry-wide action like that proposed in Chapter 6 would be unlikely, because of their voluntary character, to run afoul of companies’ First Amendment rights; for similar reasons, the interposition of an uncoerced choice by companies to build such institutions (albeit with the government’s help) would at least arguably insulate the government’s role from challenges on the basis of the First Amendment rights of users as well.

2.4 Objection 1: Is Private Governance an Oxymoron or a Danger?

Before we delve into the account of how we might promote the effective governance of users by platforms, some interrelated preliminary objections must be addressed. The first identifies a kind of teleological mismatch between the goals of governance development and the ends of private, profit-motivated, entities. Even in the ideal form of the so-called “new governance” models under which the state recruits or collaborates with private entities to govern the public, there’s an important sense in which the state remains in the driver’s seat as much as possible – for example, by contracting with private entities to carry out governance tasks, recruiting the support of private entities for pieces of an overall project of infrastructure or community-building, or convening private for-profit entities to participate in their own regulation rather than the regulation of third parties.Footnote ⁹ By contrast, the most plausible ways to develop private platform governance involve platforms regulating their users, and, in view of the international nature of the major platforms, it’s unlikely that governments will be able to exercise close supervision over the conduct of that governance. Indeed, close supervision by governments is almost certainly undesirable, due to the oppressive nature of some governments and the hegemonic nature of others – issues to be explored further below.

We have strong reasons to be wary of the delegation of governing responsibilities from states to private companies. At least in theory, states are oriented toward (and accountable to) the good of their people, whereas companies are oriented toward the good of their shareholders. In law, for example, the whole enterprise of “consumer protection” reflects a recognition that giving companies the power to dictate the behavior of people who engage in non-shareholder economic relationships with them (e.g., with excessively empowering contractual terms) makes individuals vulnerable to exploitation.

Even when states, rather than companies, have been the beneficiaries of capacity-building efforts, those efforts have often led to unintended consequences relating to the abuse of that power. For example, antiterror capacity-building programs in Kenya have also built the government’s capacity to engage in repressive actions against local minorities (Bachmann and Hönke Reference Bachmann and Hönke2010). Similarly, with respect to companies, we have good reason to fear that increasing their control over the conduct of users on their platforms could facilitate any number of inappropriate choices made by the companies over the years. For perhaps the most important example: Every platform company is widely criticized for violating the privacy of users; many scholars have suggested that such violations in the form of surveillance are the core business model of companies (e.g., Zuboff Reference Zuboff2019). But by asking companies to do more governance we might be exacerbating their surveillance by giving it a kind of normative cover. After all, the same kinds of practices that offend our notions of privacy and raise the specter of surveillance might also be useful in effective governance. Consider database merging and user profiling across platforms – many of us object to, say, Facebook’s using its infamous “pixel” to track our behavior across non-Meta-operated websites and even allegedly build profiles of users independent of their Meta-affiliated accounts (so-called “shadow profiles”) (Garcia Reference Garcia2017; Aguiar et al. Reference Aguiar, Peukert, Schäfer and Ullrich2022; Bekos et al. Reference Bekos, Papadopoulos, Markatos and Kourtellis2022). But what if that information is used to identify “inauthentic” users, such as Russians attempting to subvert elections?Footnote ¹⁰ Suddenly, it seems perhaps a bit more acceptable, and perhaps even something we might demand of companies. Looking down the game tree, we may resist conferring additional governance resources (and responsibilities) on companies in order to avoid giving them further incentives and tools to conduct such surveillance.

This problem of surveillance – indeed, the general danger that capacity improvements for governance purposes (i.e., where company interests align with the public interest) may turn into capacity improvements for exploitative purposes (where company interests conflict with the public interest) – is real and serious.

However, I contend that we can avoid objections that point to the above-described problems by attending more carefully to the ways that (a) states and companies are subject to similar incentives (so state regulation of users directly, even if possible, might not be better than governance development); and (b) we can build capacity improvements for platform governance that do not necessarily empower companies as such – but rather empower other ecosystem actors with incentives aligned both with the public and with companies on governance issues – including users, nonuser citizens, civil society, workers and other important constituencies.

To begin, it is notable that many of our most useful accounts of state development suppose that leaders have self-interested motives, and, historically, it is difficult in the extreme to suppose that the nascent stages in the early development of the high-capacity nation-state were purely or even mostly public oriented. Indeed, a substantial part of the point of the literature on the development of governance capacity and the state itself is the design or evolution of institutions that can re-align the incentives of top-level and mid-level leaders to cause them to pursue the public good in their own private interest, or the identification of private interest based causal frameworks for the emergence of such institutions (e.g., Olson Reference Olson1993; De Lara, Greif, and Jha Reference De Lara, Greif and Jha2008; North, Weingast, and Wallis Reference North, Weingast and Joseph Wallis2009; sometimes this can include the development of private legal frameworks as well, e.g., L. Liu and Weingast Reference Liu and Weingast2018).

Unsurprisingly, the worries which I articulated above also appear frequently in the context of states. Take the problem of surveillance. It’s useful to help states prevent crime, terrorism, subversion, and so forth. And it’s incredibly dangerous to enable states to conduct political repression. I’d go so far as to say that the problem of surveillance in the context of states is the most classic and intractable example of the persistent tension between security and freedom. Shorter version: J. Edgar Hoover existed. Assuming that someone has to watch what people say on social media to control the spread of things like viral misinformation, there is little reason to think that the state is any less likely to abuse the power to do so than companies.

Moreover, it seems to me that the key problem with the objection to private governance is that it assumes a too-thick distinction between “public” and “private” which is, to some extent, merely a quirk of our place in history. Rather, we ought to understand that the “public,” both in the sense of governments and in the sense of other ways of noncommercially organizing people (e.g., via civil society, social movements, and even purely artifactual organizations), can be integrated into the decision-making processes of private companies in ways that are both private and public – private because the companies remain for-profit businesses with private owners, whose decisions are in substantial part motivated by profit – yet public because those decisions can both be controlled in the public interest and improved in the private interest by integrating noneconomic forms of human organization, and hence human incentives, into them.

Critiques of private governance can also merge into critiques of the undemocratic nature of such forms of governance – and hence the implicit claim that what counts as democratic is limited by existing forms of public authority. As an example, consider two recent papers, one by Blayne Haggart and one by Haggart and Clara Iglesias Keller (Haggart Reference Haggart2020; Haggart and Keller Reference Keller2021). The two papers criticize several major platform governance initiatives such as the Meta Oversight board and proposals by several governments, as well as leading scholarly proposals, on the basis of their lack of democratic legitimacy.Footnote ¹¹ In particular, they focus on a conception of democratic deficit framed around what the Haggart/Keller paper describes as their lack of “input legitimacy,” that is, the failure of major decisions to go through democratically accountable institutions.

There is force to their objections. For example, they object to the design of the Meta Oversight Board in virtue of the fact that its decisions ultimately are still in the control of the company, both because the company makes the rules and because the company sets the terms of civil society and other participation (e.g., by choosing the initial chairs). And, they criticize David Kaye’s (Reference Kaye2019b) human rights-focused approach for being overly concerned with protecting companies and their users against censorship efforts by governments, and failing to distinguish between democratic and authoritarian governments and the differing degrees of trust which ought to be extended to them in controlling online speech.

Yet the vision of those papers is limited by their rigid conception of the shapes democratic institutions might take. Both papers assume a fairly cramped Westphalian statist notion of democracy, in which it is impossible for a decision-making process to be legitimate unless it is associated with a specific, bounded, demos.Footnote ¹² But if we value democracy because of the normative importance of individual and collective autonomy and the practical advantages of superior decision-making, I contend that those values can be achieved outside the context of the nation-state (back to the Deweyian point about how democratic institutions need to proceed from an understanding of the relevant underlying public and, well, its problems). Similarly, I think Kaye has the better of the argument with Haggart and Keller with respect to the danger of excessive control over platform speech even by nominally democratic states – we need only consider the threatened abuses of government power in weak democracies run by demagogues such as the United States during the Trump regime or India during the Modi regime to illustrate the risks involved. More generally, even robust democracies have long-term minorities, indigenous peoples, and other groups of people identified with, but not fully represented by, the nation-state that happens to rule the physical territory. For example, Spain is widely seen as a stable and legitimate democracy, but it is far from obvious that it can be understood to fairly speak for its Basque and Catalan citizens; how much less can the United States be understood to fairly speak for its Native American citizens.

It seems to me that the goal of achieving “input legitimacy” is a worthy one, but that we ought to focus our attention on ways to do so that include, but are not limited to, the nation-state. Rather, this book will argue for platform governance reforms that can genuinely incorporate both democratic peoples organized into such states and national minorities, global civil society and social movements, and other forms of meaningful human organization not encompassed by the modern state. I will further argue that such reforms can actually make platform governance more effective at governance – that is, at writing rules that work and can be effectively enforced to control harmful externalities. The state, as well as platform companies, must drive such reforms, because it is states (and multi-state organizations like the EU) and companies that currently have the power to do so. But those empowered by the reforms cannot be limited to states and companies.

A better approach is the one described by Hannah Bloch-Wehba (Reference Bloch-Wehba2019, 67–71) in what I believe to be the closest affinity in the existing literature to this book’s governance development approach. While Bloch-Wehba does not argue for capacity building as such, she does argue for the building of global democratic institutions to govern platforms as a form of what we might call legitimacy building, to contrast with capacity building. By doing so, she recognizes in a way that Haggart and Keller do not, that novel contexts for the exercise of power call for novel forms of democratization that are not wedded to pre-existing institutional forms.

The objection to private governance as raising a kind of teleological conflict between the interests of private companies and the public interest represented by states also ignores the scope limitation given at the beginning of this book. The scope of governance development as I propose is limited to the category of problems mentioned in the Introduction, where there is a minimal alignment of interest between platform companies and the rest of us. “Minimal” stands in for the recognition that platform companies might have complicated and conflicting interests, especially when considering the difference between short-term and long-term perspectives. “The rest of us” stands in, more or less, for the people of reasonably liberal and democratic states with interests in socially healthy behavior, an avoidance of fraud and malicious political subversion, and the like. The notion of minimal alignment of interests is also meant to recognize that sometimes the role of public policy is in part to give companies an incentive to follow the side of their interest which they share with the rest of us. Consider the problem of viral political misinformation: This is clearly against the public interest (poisoning democratic debate with lies is bad for everyone except the malicious actors spreading them). And it is partly against company interests – becoming known as a place where your quirky uncle Terry learns to be afraid of vaccines is not good for the long-term health of companies (or uncle Terry), as both users and advertisers are likely to be driven away. However, it is also partly consistent with their interests, understood from a different (more short-term) perspective: Viral misinformation is still activity, and it’s activity that some users vigorously consume; in an economic model that monetizes engagement, toxic engagement still counts. Governance reforms might make a difference in those kinds of problems by giving companies the incentive and the ability to focus on the long-run benefits of getting rid of the viral misinformation rather than the short-run benefits of allowing it.Footnote ¹³

By contrast, the problems where such alignment of interest does not exist are not amenable to governance reform. Consider the surveillance problem again: critics of the platforms represent it, perhaps rightly, as a pure clash of interests between companies and the rest of us. They make more money by doing and facilitating more surveillance; we enjoy more privacy, freedom, and power over our own lives by being subject to less of it. These problems are serious and critical, and we need to find a way to address them – but they are not the subject of this book, except insofar as they are indirectly implicated by this book, for example, because we need to craft our governance reforms with an eye toward not making the conflict of interest between platform companies and the rest of us worse.

That being said, in an ideal world, well-crafted governance development might also change platform incentives in the right direction even with respect to those situations where there is a genuine clash of interests between companies and the rest of us. In particular, suppose I’m right that more effective governance – over issues where companies’ interests are aligned with the rest of us – requires giving ordinary people more power within and over companies (as I argue in the rest of the book). Those same reforms might also give the rest of us enough influence to at least partly nudge company incentives over a little bit in places where they are not aligned. More bluntly, once we democratize the platforms, we might discover that our new democratic institutions can help them find revenue models that are less dependent on surveillance, the exploitation of workers, the evasion of local regulation, and all the rest of their abusive behaviors.

2.5 Objection 2: Does Platform Capacity Building Over-empower the Already Powerful? Is It Colonial? Imperial?

Here in the United States, our problems tend to dominate not only our own but the world’s minds. Because of our economic power, our ideologies also tend to dominate platform governance. Twitter’s executives used to call it “the free speech wing of the free speech party,” in line with America’s position on the global extremes of free speech and the fact that, at the time, its leadership was dominated by famously libertarian Silicon-Valley types. On the other end of the libertarian spectrum, the same pattern holds: America’s “leadership” (if we can call it that) in hyper-aggressive intellectual property regulation and the power of its media companies have exercised an outsized influence on global intellectual property law (e.g., Kaminski Reference Kaminski2014; Yu Reference Yu2006), doubtless including its enforcement by platform companies.

To a limited extent, Europe has also been able to propagate its values via platforms. Because of the size and wealth of its market, a number of aggressive EU internet regulations have exercised an international influence – the General Data Protection Regulation being the most prominent, but the Digital Markets Act and the Digital Services Act have also drawn particular attention, as have a number of proposed or implemented intellectual property regulations.

Accordingly, any interventions that might make these firms more effective at governing platform behavior potentially promote the further propagation of US and European values to people and places that may not want them. US free speech ideals might, for example, be applied in regulating conduct on social media in countries with much less libertarian norms; US and European intellectual property regulations might be used to censor the speech of users from countries and cultures with a very different attitude toward things like authorship and copying.

At the limit, Americans may even be making extremely contentious political judgments about the legitimacy of the leaders of other states. At least when Donald Trump was kicked off Twitter and Facebook, the executives making the decision were Americans, in American companies. But what happens if, say, Narendra Modi or Jair Bolsonaro were to incite a similar attack on their national legislatures after losing an election? Would we be as comfortable with the likes of Dorsey and Zuckerberg (or, heaven help us, Musk) making the same call?

Consider as a concrete example of the difficulty of these determinations the Thai laws against insulting the monarchy. There’s a legitimate debate to be had about whether those laws – restrictions on political speech, to be sure, but restrictions that are associated with a longstanding political tradition and that are, at least on their face, fairly narrow – are consistent with human rights principles. Platform companies find themselves thrust into such debates when, for example, the Thai government asks YouTube to take down videos committing lèse-majesté (DeNardis Reference DeNardis2014, 218; Klonick Reference Klonick2018, 1623).Footnote ¹⁴

Exacerbating this risk of colonialism is the notorious problem of cultural competence experienced by many platforms, which frequently lack political, social, and even linguistic knowledge sufficient to adequately interpret behavior or model the values of foreign cultures. Perhaps the most famous example is Facebook’s habit of taking down pictures of indigenous persons with culturally appropriate bared breasts (Alexander Reference Alexander2016), evidently on the basis of having misinterpreted the expression as pornography.

As with so many issues in internet governance, the problem of colonialism has not begun with the phenomenon of platforms. Rather, even governance strategies for the basic technical infrastructure of the Internet, such as technical standard-setting processes for things like IP addresses, WiFi, and the like, can easily exhibit features that could be described as having at least a semi-colonial character. DeNardis (Reference DeNardis2009, 201–5) describes how such standard-setting processes can become dominated by developed country politics, and, in doing so, embed economic advantages to developed countries, such as a preference for incumbent intellectual property holders in those countries rather than new entrepreneurs in developing countries.

Moreover, there is a distinctive historical fidelity to the worry about colonialism in private governance. A major driver of European colonialism as it actually occurred was the empowerment of companies like the Dutch East India Company, in a process that historical sociologist Julia Adams (Reference Adams1994, 329–34) described as “patrimonial” in virtue of the way that the Dutch state, under the influence of powerful merchant elites, granted the company legal and financial privileges and sovereign authority while structuring its operations in order to reinforce their own institutional power. Adams (Reference Adams1994, 335–36) argues that the organizational structure of the company facilitated colonial enterprises of military and economic domination up to and including genocide by providing a mechanism through which the state’s repressive goals could be laundered.

The enterprise of rule-of-law development too has been criticized as colonial – including in my own prior work (Gowder Reference Gowder2016, 170–71; see also Mattei and Nader Reference Mattei and Nader2008). Rule-of-law development projects often appear to rest on the assumption that developing nations lack legal systems of their own – because they don’t look exactly like the western ones – and attempt to impose the values and institutions of the countries – mostly the United States and Western Europe – doing the promotion on the nations in which such projects are carried out. Certain kinds of capacity-building assistance may be downright harmful to recipient countries, perhaps even imposed on them coercively. For example, the enterprise of capacity building has been deployed in order to “support” developing countries engaging in more vigorous intellectual property rights enforcement (May Reference May2004) – even though local economies may benefit from a lower degree of intellectual property protection than the international norm, and even though the international norm might be understood as being imposed by the greater trading power of powerful countries like the United States. My suggestion that platform governance regulation amounts to a kind of capacity development similar to rule-of-law development in the state context might seem to invite similar worries.

The conduct of major platforms has already been criticized as colonial. For example, Kwet (Reference Kwet2019, 12) criticizes Facebook’s “Free Basics” internet service, which provides network connectivity for free in developing countries as a way to acquire market dominance via preferring the company’s own services, in view of the way such market positions enable such companies to exercise control over communications in those countries and thereby, in Kwet’s (Reference Kwet2019, 12) words, “undermines various forms of local governance.” Similarly, Cohen (Reference Cohen2019, 59–61) interprets platform extraction of data from and surveillance of the global South – sometimes with the collaboration of local governments – through a colonial lens.

Moreover, other scholars have identified the dominance of US platforms with the concept of imperialism. As I read Dal Yong Jin’s (Reference Jin2013, Reference Jin2017) influential work on the subject, platform imperialism is primarily economic in nature, and revolves around the way that platforms leverage intellectual property law as well as the commodification of information created by as well as about their users to extract wealth from the global South. In this sense, it can be distinguished from colonialism, which I mean to refer to political and social rather than economic domination – that is, the undermining of the freedom of colonized peoples to determine their own fate. However, the two phenomena obviously tend to go together – the propagation of American (and European, etc.) cultural influence increases both American economic influence and American social/political influence, and the two sorts of influence can be self-reinforcing.

This imperial quality can also be a consequence simply of the use of platforms as a kind of standard due to their positive network externalities. As David Grewal (Reference Grewal2008) has cogently argued, positive network externalities constitute a form of power in that they degrade the usefulness and availability of alternatives, and hence deprive people of access to those alternatives. Even if nobody is in control of the object of one of these kinds of positive network externality coordination goods (or “standards”), such as a language or a currency being used as a lingua franca, Grewal’s work shows that nonetheless, they have the potential to suppress diversity and local autonomy. So, for example, the network externality-derived usefulness of Facebook or Google in the context of a global economy may impede the development of social media or search engines that are more responsive to local needs.

Another form of colonialism or neocolonialism (quasi-colonialism?) in the platform business revolves around their labor practices. In particular, Roberts (Reference Roberts2019, 182–200) describes how the commercial content moderation enterprise is embedded in colonial relationships, as offshored content moderators in the Philippines work in their (exploitative) jobs under conditions both of colonial cultural influence (this work is done in the Philippines partly because American colonial legacies have left workers there with a high degree of cultural and linguistic competence relative to American culture) as well as company-driven development paths in things like infrastructure choices.Footnote ¹⁵

Objections to private governance and objections to colonialism potentially come together, as one tool with which developed nations have objectionably dominated developing nations has been the integration of nonstate organizations into government. Even leaving aside historical examples such as the various East India companies, contemporary governance capacity-building efforts have been criticized for effectively replacing functions of local states with private organizations that appear to answer more readily to the donor states than to the domestic governments they are supposed to serve (e.g., Mkandawire Reference Mkandawire2002, 16–18).

In a world in which every government was legitimate and just, some potential solutions would suggest themselves, primarily involving platform rules being modified and applied in accordance with the requests of local governments. But, of course, this is not the case, and, in fact, the quasi-colonial character of platform governance is actually valuable to many living under authoritarian regimes, particularly with respect to social media. For example, one of the stories we might tell about the role of social media in the Arab Spring is that revolutionaries and dissidents took advantage of the fact that companies like Twitter imposed American free speech norms on content in countries like Egypt. For a more recent example, Brazilian populist demagogue Jair Bolsonaro attempted (albeit unsuccessfully) to prevent social media companies from removing his COVID-19 misinformation (BBC News 2021; Marcello Reference Marcello2021). Thus there is an inherent tension between the goal to promote the self-determination of peoples with diverse cultural and value systems and the goal to not be a tool of government oppression or demagoguery.

Nor can international human rights standards, on their own, resolve the tension between platform colonialism and platform culpability in governments gone bad, for human rights standards, like all laws and other normative principles, are subject to interpretive disputes, and it may still be objectionable for Western elites to make themselves the authorities to decide whether, for example, the request from some government to censor some content from one of its citizens comports with those standards. Moreover, the content of those standards themselves is disputed, especially by states that see themselves as having a religious identity shared with the entire community – thus, for example, the Cairo Declaration on Human Rights in Islam is widely seen as a response to the Universal Declaration on Human Rights which articulates a contrasting framework meant to be specifically Islamic.Footnote ¹⁶

Even among wealthy liberal-democratic nations, there is disagreement on the scope and applicability of human rights norms, and that disagreement has already leaked into the platform governance domain. Internet human rights group Article 19 has criticized Germany’s “NetzDG” intermediary liability law for (among other things) imposing liability on companies who do not help enforce Germany’s prohibition against glorifying the National Socialist party – a law that, in Article 19’s view, violates international human rights norms with respect to free speech.Footnote ¹⁷ To me, Article 19’s objection is deeply misguided: Germany has prohibited the National Socialist party for a very long time and we all know why. It’s hard to imagine a more compelling case for deference to a local interpretation of a global human rights norm in light of a country’s distinctive history, culture, values, and dangers. But this just goes to show that even the extent to which global human rights standards are subject to local variation is itself controversial.

2.6 Objection 3: The Whole Industry Is Bad – Radically Remake It by Force

One question that immediately arises as we consider the harms platform companies have caused is: Should we (as citizens of democratic states) simply put a stop to them? Should we ban their services (consistent with constitutional and international free speech and other limitations)? Should we tax them into oblivion?

Those who write about the major platforms almost universally emphasize the social harms they create. And, doubtless, those harms are significant. Amazon crushes small businesses and abuses its workers, while facilitating a seemingly endless supply of counterfeits and scams. Facebook and Twitter and Instagram and the like promote viral conspiracy theories, political polarization (maybe, the empirical jury is still out (Tucker et al. Reference Tucker2018)), and psychologically unhealthy forms of social self-presentation and “news feed” addiction. Airbnb and Uber and the like make it dramatically more difficult for us to democratically regulate industries like hotelkeeping and livery services, and have engaged in troublingly exploitative work practices under the guise of independent contracting.

But Facebook and (pre-Elon) Twitter and Instagram have also provided many of us the services listed on the tin: They have allowed us to maintain and build real connections with other people which we might not otherwise have maintained.Footnote ²⁶ There are friends from the many places I’ve lived with whom I keep in touch almost exclusively over social media; and the fact that the pictures of their dog or kid or whatever occasionally scroll through my feed makes it much easier to maintain the rewarding social relationships. Those relationships have social and emotional value that is real, and that was much harder to maintain before the existence of social media.Footnote ²⁷ Those relationships have economic value too: as has been well understood at least since Granovetter (Reference Granovetter1973), things like jobs and other valuable opportunities are often learned about through “weak ties” – that is, people with whom one is acquainted, but less close than, say, one’s intimate friends or the colleagues whom one sees daily – precisely the ties that social media most helps us maintain.Footnote ²⁸ And, indeed, I have learned of important publication and consulting opportunities from weak ties via social media.Footnote ²⁹

The same can be said for the transactional platforms. For all the harm that businesses like Uber and Airbnb have caused, the fact of the matter is that they meet (or at least met at the time of their creation) a real need. For perhaps the most obvious example, the taxi market was egregiously anticompetitive if not outright corrupt until Uber and Lyft came along. At one point, New York taxi medallions were selling for up to 1.3 million dollars (Harnett Reference Harnett2018). That’s blatantly corrupt: drivers would go into life-ruining levels of debt (Rosenthal Reference Rosenthal2019) to buy the privilege of participating in an artificially supply-constrained market; not all that different from student loans to go to medical school or law school, except that those professional degree programs provide an actual education necessary to do the job, whereas the taxi medallion system was just a pure government-sponsored cartel.Footnote ³⁰ Or they would have to rent their medallions from those who could afford to own them – so that the government monopolies gave way to investor monopolies in which a rentier class who (until Uber and Lyft came along) effectively took no risk (thanks to the artificially constrained supply) and could simply print money.Footnote ³¹ What possibly legitimate policy reason could there have been to create a system in which investors could speculate on occupational licenses? The level of total policy failure represented by the medallion system is just astonishing: by creating an artificially supply-constrained cartel and then putting membership in that cartel on the market, it managed to combine the worst aspects of capitalism with the worst aspects of central planning. Uber and Lyft did a good thing by blowing it up; the taxi drivers who were harmed in the process should seek recompense from the corrupt politicians and medallion oligarchs who built the predecessor system in the first place.

Similarly, Amazon, for all its brutal market power, in all likelihood helped save lives during the COVID-19 pandemic as its massive and painstakingly built logistics network meant that it could actually deliver necessities to people at semi-affordable prices rather than send them out into retail stores to infect one another with a deadly disease (and yet, at the same time, they probably killed a few as well by providing a platform for fraudulent sellers to distribute counterfeit N95 masks). Similar claims could easily be made of food delivery companies like Doordash and Instacart. In the beginning of 2021, as the federal government struggled to manage the distribution of the COVID-19 vaccine, President Biden reportedly considered recruiting companies like Amazon to help – and rightly so, since the kind of pervasive logistics network that can somehow reliably deliver a potholder to the door of almost any American in two days had the potential to be immensely helpful in getting vaccines into arms (Scola and Ravindranth Reference Scola and Ravindranth2021).Footnote ³²

The big mainstream platforms also cause undeniable harms. I take no position on whether the overall balance of social benefits and harms caused by platforms is in the black or the red. But I wish to urge caution here: nobody else is likely to be able to carry out that balancing act either. The harms they have inflicted on the world are probably incommensurable with their benefits, and that evaluation embarrasses the case for taking a prohibitionist approach to the companies.

For example: During the COVID-19 pandemic Amazon, Instacart, and similar delivery companies helped protect the health of those who could afford their services at the expense of workers, such as warehouse employees and delivery “contractors” (really, misclassified workers), who were unjustly placed at greater risk for the virus without being fairly compensated for taking that risk or, in many cases, having any real choice about the matter. This is a moral stain on the companies, on everyone who benefited from the harms inflicted on those workers, and on society at large. It is also a fairly common experience in capitalism, no different in any meaningful sense from the moral wrong committed for decades when Americans benefited from cheap and readily available energy to drive economic growth on the backs of brutally endangered coal miners. Critics from the left have been pointing out that workers are exploited under capitalism and endangered for the benefit of others for centuries, this has nothing to do with the platform economy. But can we honestly say that the world would be better off if Instacart had not existed during the pandemic? Or is the genuinely honest thing to say a variant of “something like Instacart should exist, but combined with massive redistributive taxation to ensure that those who benefit fairly compensate those who take the risk, and those who take the risk do so freely?”Footnote ³³ And in the absence of an obvious route to that alternative, does it make a lot of sense to try to engage in global evaluations of the with-Instacart and without-Instacart worlds?

Again, the same is true of the social platforms. Facebook and Twitter and Reddit and the like have unquestionably facilitated immense harms to our societies. Many people plausibly believe that Donald Trump wouldn’t have been president without them, and that a mound of corpses of unknown size (related, e.g., to mismanagement of the COVID-19 pandemic and brutal immigration policies) is thereby attributable to their existence. Maybe so, but none of us have any clue how to actually evaluate that counterfactual in a world in which, for example, Fox News and talk radio exist, and in which the existence of the Internet itself makes it almost inevitable that spaces for dangerous political groups will exist outside the scope of big technology companies.

At any rate, heavy-handed regulation is a notoriously ineffective strategy for controlling products that produce real benefits to their users, even if they also produce social harms. Consider Uber again. In principle, municipalities could ban un-medallioned transport-for-hire altogether, and could subject people who offered or used those services to criminal sanctions. But this kind of prohibition strategy as to products and services that are already popular does not, it is fair to say, have a terribly good record. In all likelihood, now that the ridesharing genie is out of the bottle, any such prohibitory efforts would both increase the pervasiveness and harms caused by the intrusion of law enforcement in day-to-day life (you think the corporate surveillance done by platforms is bad? imagine the government surveillance necessary to control ride-sharing!) and be unsuccessful at controlling the conduct; in fact, exacerbate the harms of the conduct by driving it to black markets. See, for example, alcohol prohibition, the war on drugs, anti-prostitution policing, and countless other examples.

2.7 A Cautionary Approach to Antitrust Law

A major policy lever that both the American left and the American right seem eager to pull – ranging from Elizabeth Warren (Kolhatkar Reference Kolhatkar2019) to Josh Hawley (Bartz 2021) – is the use of antitrust law to break up the platform companies on the order of Standard Oil or AT&T, or at least to prevent additional corporate acquisitions. In some sense, this is the opposite of a capacity development strategy, as the goal becomes not to improve companies’ capacity to govern user behavior on platforms but at least partly to defang it – and defanging it is at least part of the reason for such proposals (certainly it motivates Hawley’s, as the US political right have often decried companies’ alleged censorship of their views – more on this in Chapter 4). Nonetheless, to the extent that some of the social harms from platforms come from their market dominance, it is worth a brief discussion of this alternative strategy to close this chapter.

The premise of the use of antitrust law against platforms is that many of their pathologies are the consequence of market power: There are no realistic alternatives to Google for search or Facebook for social media (although surely the latter is plainly untrue), and hence the bad decisions of those companies are amplified across society at large, while network externalities make it unreasonably difficult for market discipline to control those decisions.

The political relevance of what many platforms – especially social media platforms – do adds fuel to the case for their breakup. As Tim Wu (Reference Wu2018, 132) has suggested, there may be a democratic case for increasing competition in the companies that control substantial shares of the public sphere similar to the case for efforts against traditional media concentration.Footnote ³⁴ This is certainly a point that would be welcomed by the American political right, as there is increasing reason to think that at least Facebook intervenes (however ineffectually) to stop far-right movements from getting a foothold on the platform (Horwitz and Scheck Reference Horwitz and Scheck2021) – and, alas, in the United States, there appears to be very little distance between the extreme right and the mainstream Republican Party. Similarly, Zephyr Teachout (Reference Teachout2020, 43–44) deploys the fundamental premise of this book – that platforms are behaving like governments – to suggest that we ought to see companies like Meta as in effect competing against the democratic authority of the United States, and hence as particularly amenable to being broken up.

However, conventional strategies of corporate dissolution may, unless paired with other substantial policy interventions, entail the sacrifice of important public interests that are actually served by the companies in their present forms. With respect to Amazon, notwithstanding all the company’s serious misbehavior, it is eminently reasonable to suppose that there are genuine economies of scale in a national logistics network of the kind that Amazon built.

With respect to social media, some breakup strategies seem to risk – perhaps deliberately, as in the Hawley case noted above – undermining the goal of governance to the extent they might encourage companies that serve niche markets, which, in turn, may cater to socially harmful interests. Consider that, for all the harms they inflict, Meta and Alphabet have strong incentives rooted in their scale to be compatible with mass consumption markets: If, for example, either Facebook’s News Feed or YouTube’s recommended videos became pervasively filled with Nazi-adjacent content, both large numbers of ordinary users and advertisers would likely flee the platforms. The slow-rolling destruction of Twitter after the Musk acquisition, with users and advertisers fleeing the platform as the extreme right is welcomed back, stands as a clear example.Footnote ³⁵

By contrast, consider the actual niche social networks that have arisen in recent years: companies like Gab and Parler outright designed to appeal to Nazi-adjacent users. And consider their niche predecessors in enterprises like 4chan or its even scarier descendant Kiwi Farms. Lacking the constraints of market operations at scale, like mass-market users, mass-market advertisers, press attention, and publicly traded stock, these companies had no reason to control the incitement found on their platforms; indeed, it took a large and hence publicly accountable company, Amazon (in its infrastructure rather than platform business), to bring one of them at least temporarily under control, to wit, by kicking Parler off its Amazon Web Services infrastructure following its role in facilitating the organization of January 6, 2021, coup attempt at the US Capitol (Romm and Lerman Reference Romm and Lerman2021; Guynn Reference Guynn2021). In a vivid illustration of the effects of social isolation on extremism, at least a handful of the insurrectionists spent some of their time in the Capitol showing off for the presumptively sympathetic audience to be found on Parler (Klein and Kao Reference Klein and Kao2021). According to legal pleadings filed by Amazon in its defense against a suit by Parler over the termination of its hosting service, Amazon had been urging Parler to control the threats of violence against Democratic officials such as Nancy Pelosi at least since November.Footnote ³⁶ While Amazon (and Apple, and Google) may certainly be criticized for not putting a stop to Parler well before January 6, there is zero reason to believe that Parler’s operators would have put a stop to its incitement on its own initiative ever.

This point draws on a larger worry about antitrust: to the extent there’s a normative justification for antitrust law in consumer services, it revolves around the idea that a more competitive market will provide a wider variety and better prices of those services to the general public. But where services might be specialized to socially harmful market segments, it might actually be a bad thing to encourage the growth of niche service providers, and a good thing to permit the homogenization associated with big centralized providers.

One might imagine a number of alternative strategies for platform company breakups that could avoid these problems. For example, different services could be disaggregated: Facebook and Instagram could be separated; Amazon’s delivery services and its network infrastructure (AWS) could be separated; Google search and YouTube could be separated. There might be some possibility of geographically centered breakups – such as Amazon logistics across US regions – or along the lines of proxies for geography such as language.Footnote ³⁷

On the social network side, larger companies might be able to achieve socially beneficial economies of scale in content moderation, for example, training machine learning models to identify certain kinds of content on much larger and more diverse datasets. However, in principle, antitrust strategies could be devised to avoid undermining such economies of scale, for example, by breaking up product sides of companies while leaving governance sides (such as “integrity” entities) intact. Voluntary industry cooperation could also ameliorate some of the difficulties of disaggregation in that respect.

On the whole, it seems to me – again, quite tentatively – that while antitrust breakups may be useful to ameliorate other harms that platforms impose on the world – such as invasions of privacy and impeding innovation by blocking market entrants – it is on balance unlikely to be effective as a solution to those problems that fall under the category of “governance” as used in these pages, within the domain of this book’s concern, that is, where long-term platform company interests and the interests of the rest of us are aligned.

The question of antitrust, however, fits more closely into the overall project of this book in two respects. First, the democratic reforms which this book defends may also be a component of a program to mitigate the public dangers of antitrust action of the sorts I question above. The core proposal of Chapter 6 is to create democratic institutions of platform governance that interact with, but are independent of, individual companies and operate across them. Creating such institutions might also make it less dangerous to break the companies up because those institutions themselves could embody some of the beneficial effects of scale, such as giving companies pressure to moderate content in accordance with the needs and desires of vast bodies of people (rather than wicked niches) and aggregating both practical knowledge as well as more brute tools like machine learning models or training data. In the world I envision in Chapter 6, governance economies of scale might be capable of being spread across the industry as a whole (depending on how much infrastructure to intervene directly in content moderation and other kinds of rule enforcement could be built around the participatory institutions I describe), and pressure from empowered ordinary people could be directed at the industry as a whole. Under such circumstances, perhaps even small platforms serving niche markets could benefit from overall governance reforms designed for the big companies.

From the other direction, antitrust law could be a vehicle for implementing those reforms. As Chapter 4 describes in some more detail, some of the governance pathologies of platforms result from poor internal governance, such as the excessive authority of social media company lobbyists over content policy decisions regulating the behavior of the same politicians whose favor they seek. In a recent article, Herb Hovenkamp (Reference Hovenkamp2021, 2021–32), one of the United States’ most prominent antitrust scholars, suggests using antitrust law to internally reorganize the decision-making structure of platform companies, for example, to empower different groups of stakeholders in order to prevent abuses of power.Footnote ³⁸ Hovenkamp’s approach is quite similar to my own – he draws, as do I, on Ostrom’s governance theory,Footnote ³⁹ and offers the compulsory integration of stakeholders in corporate governance structure as a tool to address platform abuses of power similar to – and in some cases overlapping with – the ones I address, such as Amazon’s abuse of data to discriminate against third-party sellers.Footnote ⁴⁰ Accordingly, it may ultimately be the case that the reforms offered in this book are less an alternative to antitrust law than a potential product of it.

At this point, I consider the broad approach of a governance development framework adequately defended. The rest of this book turns toward how such a framework would work – the characteristic problems of governance and institutional solutions to those problems shared both by states that struggle to govern and by platforms that struggle to govern (Chapters 3 and 4) and the sorts of concrete institutions that might be built to alleviate them in the platform context (Chapters 5 and 6, and the Conclusion).