Freedom of expression in the age of artificial intelligence: the risks and challenges to online speech and media freedom

In 1842, Ada Lovelace had declared that “the Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform. It can follow analysis, but it has no power of anticipating any analytical relations or truths. Its province is to assist us in making available what we are already acquainted with.”

Almost two centuries later, we are grappling with these same fundamental issues, as the impacts of artificial intelligence on society unfold. Technological advances in the past decade have fueled a boom in AI applications—in communications, banking, health care, manufacturing, criminal justice, and national security—in ways that are radically reshaping our private and public lives.

While proponents say these technologies hold the promise of tackling some of society’s most complex challenges, a growing number of policy experts and advocates warn that AI has as much potential to erode human rights, harm our democracies, and further entrench and amplify discrimination and global inequity at scale. And in the rush to innovate, it has become increasingly apparent that neither governments nor private companies have put adequate human rights protections and governance mechanisms in place to ensure that AI’s benefits to society outweigh the harms.

How does AI shape our information environment?

Internet intermediaries, and social media platforms in particular, have increasingly turned to AI systems to help personalize and disseminate news and information, as well as to flag and remove harmful or unlawful content. These applications have dramatically increased the already significant power that online platforms have in shaping our information environments, and in determining what content we access, view, read, and share online.

As AI systems play a central role in defining our news and information diets, we must ask: how are these systems influencing—or threatening—our rights to freedom of expression and access to information? What are the broader implications for the free exchange of ideas and opinions and for our ability to exercise the political rights and freedoms that foster healthy democratic societies?

The 2022 Digital News Report by the Reuters Institute of Journalism at Oxford estimates that 77 percent of all online news is accessed through so-called “side door” channels—search engines, social media, and news aggregators—that rely on AI systems to both serve and censor the news and information we consume.

These technologies and systems are designed to deliver content based on how much engagement it is likely to generate—as measured by the number of views, likes, and shares—rather than for its accuracy, diversity, or public-interest value. Platforms target individuals and groups with tailored content (and ads) based on algorithmically-generated collections of our data, touching everything from our political affiliations and preferences, to our gender, race, age, income, and other characteristics.

At the root of the problem is a profit model that is based on the mass collection of data about our digital activities and behaviors. This is what Harvard scholar Shoshanna Zuboff calls “surveillance capitalism.” [1] Many platforms today derive their revenue from targeted advertising, which is powered by opaque algorithmic systems designed to capture as much of our data as possible, and then use it to drive reach and engagement, and maximize corporate profits. This system creates the distribution channels through which divisive, discriminatory, hateful, and illegal online content flows.

How much do private platforms know about you?

A team of documentary filmmakers set out to answer this question—and to illustrate how the digital traces of everything we do online can be used to assemble powerful profiles about individuals, and even to manipulate our opinions and expressions. Made to Measure exposes the power of surveillance capitalism and how much private companies know—and can predict—about who we are.

This interactive film was co-produced by the Organization for Security and Co-operation in Europe, Office of the Representative on Freedom of the Media (OSCE RFoM), as part of the “Spotlight on Artificial Intelligence & Freedom of Expression (SAIFE)” project.

Watch

Policymakers and the public often call on social media companies to remove disinformation, hate speech, racist and discriminatory materials, terrorist propaganda, and other forms of problematic and unlawful speech. [2] In response, platforms like Facebook, Instagram, YouTube, Twitter, and TikTok have developed and deployed different types of AI to help police an immense universe of digital content that violates the law or that runs afoul of companies’ own policies.

But research has consistently shown that these tools are unreliable in identifying harmful online speech, including illegal hate speech and terrorist content. [3] Studies reveal that these technologies are context blind and not sophisticated enough to accurately detect other types of harmful or illegal content, such as hate speech. AI systems have immense potential to shape public opinion and political behavior—and equally immense potential for exploitation by malicious actors and political elites. In recent years, we have seen how these technologies and systems can be and too often are manipulated to target voters with disinformation in an attempt to influence elections.

The use of AI to moderate content therefore has mixed results: these systems can misidentify and remove legitimate content in some instances—a form of prior restraint that results in over removal and thus censorship—while failing to accurately or consistently spot harmful, abusive, or illegal content, in which case these materials continue to proliferate online. Both false positives and false negatives can have severe impacts on the online information landscape and enjoyment of freedom of expression of individuals and communities. The removal of legitimate speech by automated tools can restrict access to information, limit media pluralism, and threaten freedom of expression. Investigations by media and civil society have revealed numerous cases in which social media platforms have blocked or removed opposition posts and hashtags during times of political unrest. At the other end of the spectrum, the flood of hateful, racist, toxic speech can have a deeply corrosive effect on public discourse but can also perpetuate discrimination and other types of human rights harms toward individuals and groups both on- and offline.

Ample case studies point to how AI-based content curation and moderation can distort or reshape online news diets in ways that restrict or limit the free flow of ideas and opinions—essential prerequisites of healthy democratic systems. Research has shown how AI-based content curation can create so-called “filter bubbles” or “echo chambers,” which dramatically limit exposure to views and information that do not align with an individual’s already-established beliefs. While the effects of such “filter bubbles” on an individuals’ beliefs or behaviors is unclear—-and in some cases their existence has been found to have been overstated—what is certain is that filter bubbles can accentuate and drive political and societal polarization, which in turn can have a deleterious and harmful effect on the open exchange of opinions and ideas. [4]

While platforms often claim that their AI systems are advanced enough to detect and remove harmful materials, the continuing swell of hate speech, disinformation, and other types of problematic materials online tells a different story. In 2021, the UN Special Rapporteur on minority issues presented a report highlighting a significant increase in online hate speech and xenophobia during the 2020 pandemic, finding the majority of that content targeted minorities and women [5]. Recent investigations have also revealed that Facebook’s AI-powered content moderation systems remove only a small fraction of hate speech—less than 5 percent—which means that a vast majority of harmful, even illegal content on that platform remains online. [6]

AI and human rights: a global public policy challenge

Governments around the world are working to develop AI policies and regulations, as the pace of tech development and AI applications accelerates. Lawmakers in more than 60 countries have launched efforts to develop domestic AI policies, amid ongoing initiatives aimed at coordinating multinational AI governance frameworks. [7]

This process has sparked wider debates among domestic and international policy makers, business leaders, and civil society over how to design AI policies that foster tech innovation while also safeguarding human rights. Policymakers must work urgently, and together with a wide spectrum of stakeholders, to establish adequate human rights protections and governance mechanisms to ensure that AI works to benefit society and to advance, rather than erode human rights and fundamental freedoms.

This challenge is particularly pressing in the area of media freedom and freedom of expression that are essential to the functioning of healthy democratic systems and societies. Policymakers in many jurisdictions are turning to AI as a one-size-fits-all solution for tackling harmful or unlawful content online, despite the fact that these technologies are unreliable and pose serious risks to freedom of expression and the free flow of information and ideas online.

In ongoing discussions about the use of AI to govern online content, some governments and policymakers have prioritized national security concerns over the protection of human rights—which rights groups and experts warn sets up a false dichotomy between protecting state security and safeguarding human rights. Lawmakers should therefore take care to avoid pitting security concerns against freedom of expression rights—as lasting, comprehensive security cannot be achieved without respect for human rights and functioning democratic institutions.

1. Zuboff, Shoshana. 2019. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. First edition. New York: PublicAffairs.

2. “It’s the Business Model: How Big Tech’s Profit Machine is Distorting the Public Sphere and Threatening Democracy.” Nathalie Maréchal and Ellery Roberts Biddle. Ranking Digital Rights, March 2020, https://rankingdigitalrights.org/its-the-business-model/.

3. “Artificial Intelligence, Content Moderation, and Freedom of Expression.” Emma Llansó, Joris van Hoboken, Paddy Leerssen, and Jaron Harambam. A working paper of the Transatlantic Working Group on Content Moderation Online and Freedom of Expression, February 26, 2020. https://www.ivir.nl/publicaties/download/AI-Llanso-Van-Hoboken-Feb-2020.pdf

4. See “The truth behind filter bubbles: Bursting some myths.” Richard Fletcher. Reuters Institute for the Study of Journalism, University of Oxford, January 2020, https://reutersinstitute.politics.ox.ac.uk/news/truth-behind-filter-bubbles-bursting-some-myths. And “Beyond the Filter Bubble: Challenges for the Traditional Media and the Public Discourse,” in Are Algorithms a Threat to Democracy? The Rise of Intermediaries: A Challenge for Public Discourse. Birgit Stark, Daniel Stegmann, Melanie Magin, and Pascal Jürgens. Algorithm Watch, May 2020. https://algorithmwatch.org/en/wp-content/uploads/2020/05/Governing-Platforms-communications-study-Stark-May-2020-AlgorithmWatch.pdf

5. “Report: Online hate increasing against minorities, says expert,” United Nations Human Rights Office of the High Commissioner, March 2021. https://www.ohchr.org/EN/NewsEvents/Pages/sr-minorities-report.aspx

6. “The Facebook Files: A Wall Street Journal investigation,” Wall Street Journal. https://www.wsj.com/articles/the-facebook-files-11631713039

7. OECD.AI (2021), powered by EC/OECD (2021), database of national AI policies, accessed on 21/10/2021, https://oecd.ai.