Wednesday, June 12, 2019

Privatizing Censorship

Guest Blogger

New Controversies in Intermediary Liability Law

Michael Karanicolas

Privatization can be a controversial practice. To its proponents, it is an engine of efficiency, introducing a competitive atmosphere to stodgy and self-perpetuating bureaucracies. But there are also externalities which can come into play when governments abrogate direct responsibility over an area of administration. A private prison may be run at less cost to the taxpayer, but will it respect the rights of inmates and devote sufficient resources to their rehabilitation? Privatizing a water company could turn it profitable, but this might come at the cost of an increase in contaminants or a refusal to service unprofitable areas. Despite the common refrain that government should be run like a business, there is an important distinction between the core functions of these two types of entities. A private company’s purpose, its only purpose, is to maximize profit for its shareholders. A government’s purpose is to promote and protect the rights of its people.

Regulating speech is among the most important, and most delicate, tasks that a government may undertake. It requires a careful balancing between removing harmful content while providing space for controversial and challenging ideas to spread, and between deterring dangerous speech while minimizing a broader chilling effect that can impact legitimate areas of debate. The challenges in regulating speech are among the most vibrant and hotly debated areas of law and philosophy, with a voluminous history of jurisprudence and academic theory on how regulations should be crafted.

Today, this entire school of thought is being cast by the wayside, as the practical functions of content regulation are being increasingly handed over to an industry which is not only totally unprepared to handle the subtleties and technical challenges associated with defining the contours of acceptable speech on a global scale, but has, as far as possible, resisted taking responsibility for this function.

How did we get here?

In the early days of the commercial Internet, policymakers realized that the commercial and social potential of this new medium could best be realized if service providers were protected against direct liability for the words of their users. Without it, scalability of the kind achieved by Facebook and Twitter would never have been possible. However, this has turned into a double-edged sword. Having been allowed to grow without an expectation of policing their users, the world’s biggest tech firms were built around business models that make it very difficult to control how their products are being used.

Now, governments are demanding that the companies start taking responsibility, and impose content controls that suit their needs. In some cases, these involve fairly well recognized categories of harmful content, such as hate speech or child abuse imagery. Other examples revolve around content which is outlawed locally, but whose prohibition runs counter to global freedom of expression standards, from risqué photos of the King of Thailand to material deemed to violate conservative religious standards. In some instances, companies have entered into collaborative relationships with governments to remove content that is determined to be objectionable, notably (and controversially) in Israel. Demands for private sector cooperation are backed by a variety of coercive measures, including the imposition of large fines, threats to block a company’s website, and even the arrest and imprisonment of company employees.

The end result is a “privatized” system of content control, which is run at the behest of government authorities, but which is operated and enforced by the tech companies. To understand why this is problematic, consider the case of South Korea, where content enforcement decisions are made by the Korea Communications Standards Commission (KCSC), an administrative body whose members are appointed by the President. The KCSC is notoriously heavy handed, and frequently targets sites which criticize politicians or challenge sensitive policy areas. Their decisions are issued to the platforms, rather than to the users who post the material, and come in the form of non-binding requests for removal. Weak intermediary liability protections mean that, in practice, these requests are always followed. However, the fact that the decisions are not formally binding means that, technically, enforcement originates from the platform, rather than the KCSC, which strips users of any procedural safeguards, such as a right of appeal or even notification that their material is subject to removal.

This practice of “laundering” government content restrictions through the private sector allows for mechanisms of control which vastly outstrip what might otherwise be permissible in a democratic context. For example, Germany’s Network Enforcement Act (NetzDG), which came into force in 2018, requires companies to remove “obviously illegal” material within 24 hours of being notified of its existence. More recently, proposals from European Parliament could push the deadline for responding to “terrorist content” notifications to just one hour. No judicial or administrative process in the world operates this quickly. Similarly, traditional content restrictions were designed on the understanding that their applicability would be limited by the resources available for enforcement. But in the context of private sector platforms, enforcement is expected to be close to 100 percent, creating a vastly more intrusive system.

These issues are compounded by the fact that, due to the size and scale of the major platforms, the only practical avenue to developing moderation solutions that approach what governments are demanding is to lean heavily on automated decision-making systems. But while AI is relatively competent at screening for nudity, content that implicates hate speech or copyright infringement is vastly more difficult since it is inherently contextual. An identical statement made in Myanmar and in Canada could qualify as hate speech in the former but not in the latter, due to the fact that one country has a much higher level of underlying ethnic tension. Not only is AI presently incapable of making this type of determination, but it is questionable whether the technology will ever be able to do so.

Moreover, in a context where the legal framework sets a minimum standard of enforcement, with harsh penalties for dropping below that standard, platforms are incentivized to err on the side of caution and remove anything which even approaches the line. This problem has been widely documented with regard to the DMCA system of copyright enforcement, including clear instances where it has been gamed to target political opponents. Increasing automation will only exacerbate this tendency.

None of this is to suggest that tech companies should have no responsibilities with regard to the impact of their products on the world. But perspective is important. The resiliency of the Internet to pervasive forms of content control is a feature of the technology, not a bug. Just as we celebrate the inability of Vladimir Putin to remove an embarrassing image of himself or Xi Jinping’s struggles to stop Internet users from comparing him to Winnie the Pooh, it is these same characteristics that make it so difficult to clamp down on the viral spreading of video of the Christchurch attack.

The new privatized enforcement models, which are being embraced, to some degree, by virtually every developed democracy, threaten many key safeguards that were developed to prevent the abusive application of content restrictions. While there are clearly problems in moderating online speech that need to be addressed, the solution to these challenges must be crafted within well-recognized global norms of freedom of expression, including appropriate checks and balances, and not as a private sector solution to what is fundamentally a matter of public interest.

Michael Karanicolas is a human rights advocate who is based in Halifax, Canada. He is a graduate student in law at the University of Toronto and, as of July 2019, will be the incoming WIII Fellow at the Information Society Project at Yale Law School. You can reach him by email at michael.karanicolas at and on Twitter at @M_Karanicolas.

Older Posts
Newer Posts