Probably Speech, Maybe Free: Toward a Probabilistic Understanding of Online Expression and Platform Governance

Article Source: Knight First Amendment Institute Essay Series, August 21, 2019
Publication Date:
Time to Read: 2 minute read
Written By:



Because digital platforms operate on a large scale, bans on harmful speech will not work; rather, platforms’ limits on speech try to reduce the spread of speech that is “probably” harmful.


Policy Relevance:

Policymakers should ask who bears the cost of high content moderation error rates.


Key Takeaways:
  • The concept of “probability” has important implications for free speech.
  • Platform speech is governed by policies and technological infrastructures defined by the tech firms themselves.
  • Observers are concerned that platform self-regulation is inadequate, or suggest that the platforms are too fast and too big to be governed effectively.
  • Platform content has several probabilistic elements, including the following:
    • Moderation is activated when a threshold of user reports or other algorithmic triggers is met.
    • Ranking of news items changes as publishers attempt to drive traffic, and because of the activity of automated bots and fake accounts.
    • Disinformation is probabilistic, as fake news is ranked lower but not deleted.
  • “Probability” can refer to:
    • Expectations about the world arising from observations of natural events.
    • Attempts to control the world by punishing deviance from norms.
    • Experimenting to discover the conditions under which people or machines can confidently expect certain outcomes.
  • We should consider the environmental costs of the search for certainty; training the algorithms that create or detect fake news articles involves immense amounts of computing power and has a substantial carbon footprint.
  • Online speech platforms operate on a large scale, which makes human oversight difficult; digital platforms must operate using actuarial systems and probabilities to manage content.
  • We should carefully consider who has the power to decide whether a restriction that is 80 or 90 percent effective is “good enough.”
  • Key questions about probability and speech include the following:
    • How can proprietary systems' probabilistic models be challenged?
    • Who will suffer when probabilistic systems generate false positive and false negatives?
    • What error rates should be tolerated?
    • Should platforms be held accountable for high error rates?



Mike Ananny

About Mike Ananny

Mike Ananny is Associate Professor of Communication and Journalism at USC’s Annenberg School for Communication and Journalism. He studies how technologies and cultures of media production have the power to shape public life. His research focuses on the public ethics of communication systems, specifically intersections of journalism practice and technology design, the sociotechnical dynamics of networked news infrastructures, and the power of algorithmic systems.