Free Speech and Hate Speech

In January 2015, two gunmen walked into the Paris offices of Charlie Hebdo and killed twelve people, including the editor and several of the cartoonists. The magazine had published satirical drawings of the Prophet Muhammad, some of which depicted him as a terrorist. Within days, millions of people were carrying signs reading Je suis Charlie in solidarity. Within weeks, a counter-movement pointed out that Charlie Hebdo had also run cartoons that many people considered straightforwardly racist, and that “I am Charlie” was a strange thing to say if you belonged to one of the communities the magazine had mocked. Both sides had principled positions that were incompatible in practice. Welcome to every free speech debate ever.

Every democratic society claims to protect free speech, and every democratic society also restricts some speech. The question is never whether to draw a line but where to draw it and who holds the pen. The US, with its First Amendment tradition, protects speech that is illegal in most other democracies. Germany criminalizes Holocaust denial; France criminalizes certain forms of public racism; Singapore’s Protection from Online Falsehoods and Manipulation Act lets the government compel corrections and takedowns of statements it deems false. These are not differences of degree but of principle, and the history behind them is worth understanding on its own terms before we apply it to questions about content moderation on social media platforms.

The philosopher John Stuart Mill defined a harm principle, saying that the only legitimate ground for restricting individual liberty is preventing harm to others. This sounds simple until you try to apply it. Physical assault is harm. But is it harm to live in a neighborhood where public speech constantly signals that you are unwelcome? Is it harm to watch your community depicted as subhuman by mass media? The harm from speech is often diffuse, cumulative, and distributed unequally across social groups. Mill’s principle does not tell us how to weigh these harms against each other or against the interests of speakers.

History offers some uncomfortable answers to the question of whether speech causes harm. The blood libel circulated in Europe from the twelfth century onward. It led to pogroms, massacres, and expulsions for hundreds of years. No single pamphlet or sermon caused any particular death, but the accumulated weight of the accusation, repeated across generations, created conditions in which organized violence against Jewish communities was routine.

Similarly, in Rwanda in 1994, Radio Milles Collines broadcast a sustained campaign referring to the Tutsi minority as inyenzi—cockroaches—and read out the names and addresses of individuals on air. The station was not reporting violence: it was organizing it. Approximately 800,000 people were killed in 100 days, and several of the station’s organizers were later convicted of incitement to genocide by the International Criminal Tribunal for Rwanda. Roméo Dallaire, the Canadian general commanding the UN peacekeeping force in Rwanda at the time, argued afterward that jamming the radio station’s broadcasts could have saved tens of thousands of lives, and that the international community’s refusal to act was itself a moral choice.

Nazi propaganda did the same thing a generation earlier, drawing explicitly on blood libel imagery to depict Jews as predatory and subhuman, and Julius Streicher was convicted at Nuremberg on incitement charges despite having ordered no killings himself. These cases established a principle the postwar European constitutional designers took seriously: speech that systematically dehumanizes a group can constitute a crime against humanity without a direct causal link to any specific act of violence.

The postwar European approach to hate speech restrictions emerged from a specific historical trauma. Germany, Austria, and other countries that passed hate speech laws after 1945 did so because they had seen how free speech could be abused to destroy liberal democracy from within. The restrictions these countries placed on it were not concessions to illiberalism. They were necessary to make democracy possible.

The United States arrived at a different conclusion from a different history. The near-absolute speech protections that the Supreme Court developed across the twentieth century emerged from a country where the federal government had a long habit of using speech restrictions against political opponents. The Espionage Act of 1917 sent Eugene Debs to prison for giving a speech in Ohio opposing the draft. The Sedition Act of 1918 made it a crime to say anything “disloyal” about the military. Federal law was used consistently to prosecute socialists, anarchists, and labor organizers, but not the people who threatened them. Out of that history came a very strong judicial presumption against letting governments decide which speech was too dangerous to permit. The Supreme Court eventually settled on a test in Brandenburg v. Ohio (1969): speech is only unprotected when it is directed to producing imminent lawless action and is likely to produce it. Holocaust denial does not meet that test; neither do Klan marches through predominantly Jewish neighborhoods.

What is also relevant is that the country with the most speech-protective legal tradition has had the longest and most violent history of racially motivated political terrorism in any established democracy. The First Amendment did not prevent that violence. It mostly provided legal protection for those who organized it.

The history of how hate speech laws get applied in practice is instructive but not reassuring. Legal tools designed to protect vulnerable communities have a consistent track record of being aimed at those same communities. The sedition laws used to jail Debs were also used to deport immigrant organizers and break up labor movements. When southern states invoked breach-of-peace laws against civil rights demonstrators in the 1960s, they were using speech restrictions to protect the social order the demonstrators were trying to change. India’s Section 124A, a colonial-era sedition law passed under British rule in 1870, was used against Gandhi and Nehru before independence and has been used against journalists, farmers’ protest leaders, and climate activists since. In Turkey, the novelist Orhan Pamuk was prosecuted in 2005 under a law prohibiting “insulting Turkishness” after telling a Swiss newspaper that thirty thousand Kurds and a million Armenians had been killed in Turkey in the twentieth century. He was not convicted, but the prosecution made the point it was intended to make.

The Danish cartoon controversy of 2005 put these incompatible frameworks on public display. Jyllands-Posten, a Danish newspaper, published cartoons depicting the Prophet Muhammad, some in ways that explicitly linked Islam to terrorism. The editors defended the decision as applying to Islam the same critical scrutiny that secular European culture had for decades applied to Christianity; Muslim communities in Denmark and across the world viewed the publication as a targeted attack. Governments responded differently: some European politicians treated republication of the cartoons as an act of courage, while others condemned it as gratuitous provocation. The controversy could not be resolved because the parties disagreed about where editorial freedom ends and community harm begins, and did not share a framework that could adjudicate between those positions.

Anti-hate-speech law has provided real protection to LGBTQ+ communities. In 2013, the Supreme Court of Canada upheld a Saskatchewan Human Rights Tribunal ruling against William Whatcott, who had distributed flyers in residential neighborhoods with titles like “Keep Homosexuality out of Saskatoon’s Public Schools,” depicting gay men as predators and diseased. The court found that two of the four flyers constituted hate speech under the province’s human rights code and ordered Whatcott to pay damages. The decision was not without critics: some argued the flyers, though vile, should be protected as religious expression. But it established that speech targeting LGBTQ+ people on the basis of their identity falls within the scope of hate speech provisions, and that communities targeted by such material have legal recourse.

Content moderation by large platforms has added a new layer of private governance to this debate for which we have no good framework. The two most important case studies point in opposite directions. In Myanmar between 2017 and 2018, Facebook effectively was the internet for most of the country. Military-linked accounts used the platform to spread content depicting the Rohingya Muslim minority as terrorists and vermin, coordinating a campaign of ethnic cleansing that the United Nations later called a textbook example of genocide. Facebook’s content moderation, built around English-language content, failed almost completely. The company had fewer than five Burmese-speaking content reviewers for a country of fifty million people. It was not that the platform had a principled position on Rohingya content and chose not to act. It simply wasn’t paying attention.

When Twitter banned Donald Trump after January 6, 2021, the decision was made by a small group of executives applying a private company’s content policy. No court reviewed it, no legislature authorized it, and no democratic process governed it. Defenders called it a straightforward application of the platform’s own rules, critics on the right called it censorship by a liberal monopoly, and critics on the left pointed out that the platform had left the same account up for years while it spread disinformation and incited harassment. Everyone had a point.

When a company makes binding decisions about speech for billions of people across hundreds of political jurisdictions, applying its own policies without democratic accountability, no existing framework clearly applies. First Amendment analysis treats the platform as a private actor free to make its own editorial decisions, which is legally accurate and practically inadequate. European regulatory frameworks treat platforms as infrastructure subject to public obligations, which captures something real but was not designed for entities of this scale. The question of what obligations such platforms actually have, to whom they owe them, and who can enforce them, remains genuinely open. Meanwhile, the platforms’ owners are spending billions to ensure that whatever answer emerges will serve their interests.

see the whole series · email me

Dallaire2005

Roméo Dallaire with Brent Beardsley: Shake Hands with the Devil: The Failure of Humanity in Rwanda. Carroll & Graf, 2005, 978-0786715107

GartonAsh2016

Timothy Garton Ash: Free Speech: Ten Principles for a Connected World. Yale University Press, 2016, 978-0300226942.

Popper2011

Karl Popper: The Open Society and Its Enemies. Routledge, 2011, 978-0415610216.

Waldron2012

Jeremy Waldron: The Harm in Hate Speech. Harvard University Press, 2014, 978-0674416864.