Algorithmic Transparency: Peeking Into the Black Box

While we know that algorithms are often the underlying cause of virality, we don’t know much more—corporate norms fiercely protect these technologies as private property, leaving them insulated from public scrutiny, despite their immeasurable impact on public life.

Since early 2019, RDR has been researching the impact of internet platforms’ use of algorithmic systems, including those used for targeted advertising, and how companies should govern these systems.¹ With some exceptions, we found that companies largely failed to disclose sufficient information about these processes, leaving us all in the dark about the forces that shape our information environments. Facebook, Google, and Twitter do not hide the fact that they use algorithms to shape content, but they are less forthcoming about how the algorithms actually work, what factors influence them, and how users can customize their own experiences. ²

Companies largely failed to disclose sufficient information about their algorithmic systems, leaving us all in the dark about the forces that shape our information environments.

U.S. tech companies have also avoided publicly committing to upholding international human rights standards for how they develop and use algorithmic systems. Google, Facebook, and other tech companies are instead leading the push for ethical or responsible artificial intelligence as a way of steering the discussion away from regulation.³ These initiatives lack established, agreed-upon principles, and are neither legally binding nor enforceable—in contrast to international human rights doctrine, which offers a robust legal framework to guide the development and use of these technologies.

Grounding the development and use of algorithmic systems in human rights norms is especially important because tech platforms have a long record of launching new products without considering the impact on human rights.⁴ Neither Facebook, Google, nor Twitter disclose any evidence that they conduct human rights due diligence on their use of algorithmic systems or on their use of personal information to develop and train them. Yet researchers and journalists have found strong evidence that algorithmic content-shaping systems that are optimized for user engagement prioritize controversial and inflammatory content.⁵ This can help foster online communities centered around specific ideologies and conspiracy theories, whose members further radicalize each other and may even collaborate on online harassment campaigns and real-world attacks.⁶

One example is YouTube’s video recommendation system, which has come under heavy scrutiny for its tendency to promote content that is extreme, shocking, or otherwise hard to look away from, even when a user starts out by searching for information on relatively innocuous topics. When she looked into the issue ahead of the 2018 midterm elections, University of North Carolina sociologist Zeynep Tufekci wrote, “What we are witnessing is the computational exploitation of a natural human desire: to look ‘behind the curtain,’ to dig deeper into something that engages us. … YouTube leads viewers down a rabbit hole of extremism, while Google racks up the ad sales.”⁷

Tufekci’s observations have been backed up by the work of researchers like Becca Lewis, Jonas Kaiser, and Yasodara Cordova. The latter two, both affiliates of Harvard’s Berkman Klein Center for Internet and Society, found that when users searched videos of children’s athletic events, YouTube often served them recommendations of videos of sexually themed content featuring young-looking (possibly underage) people with comments to match.⁸

After the New York Times reported on the problem last year, the company removed several of the videos, disabled comments on many videos of children, and made some tweaks to its recommendation system. But it has stopped short of turning off recommendations on videos of children altogether. Why did the company stop there? The New York Times’s Max Fisher and Amanda Taub summarized YouTube’s response: “The company said that because recommendations are the biggest traffic driver, removing them would hurt creators who rely on those clicks.”⁹ We can surmise that eliminating the recommendation feature would also compromise YouTube’s ability to keep users hooked on its platform, and thus capture and monetize their data.

If companies were required by law to meet baseline standards for transparency in algorithms like this one, policymakers and civil society advocates alike would be better positioned to push for changes that benefit the public interest. And if they were compelled to conduct due diligence on the negative effects of their products, platforms would likely make very different choices about how they use and develop algorithmic systems.