第3章データ構造とアルゴリズム

In Algorithms of Oppression, Safiya Noble describes how search engines can amplify sexist, racist, and misogynistic ideas. While searching for “Black girls,” “Latina girls,” and “Asian girls” circa 2013, Safiya was startled by how many of the top search results and advertisements that appeared on the first page of Google Search led to pornographic results when her input query did not at all suggest anything pornographic. In contrast, searching for “White girls” did not include pornographic results. As algorithms become more commonplace in our daily lives, they also become a more potent force for determining certain social futures. Algorithms are immensely powerful in their ability to affect not only how we act, but also what we see, what we hear, what we believe about the world, and even what we believe about ourselves.

As the amount of input data increases, computers often need more time or storage to execute algorithms. This condition is known as complexity, which is based on the degree computational resources that an algorithm consumes during its execution in relation to the size of the input. More computational time also often means consuming more energy. Given the exponential explosion in demand for data and computation, designing efficient algorithms is not only of practical value but also existential value as computing contributes directly to global warming and resultant climate crises.

2 S.U. Noble, Algorithms of Oppression: How Search Engines Reinforce Racism. NYU Press, 2018.

Online social media platforms facilitate social relationships between users by allowing users to create and share content with each other. This user-generated content requires moderation, or methods for managing content shared between the platform users. Some researchers argue that content moderation defines a social network platform; in other words, content moderation policies determine exactly what content can be shared on the platform, which in turn defines the value of information. As social media platforms become increasingly prevalent, the information on these platforms plays an important role in influencing their users.

One approach for content moderation is to recruit human moderators to review toxic content, or content that is profane, abusive, or otherwise likely to make someone disengage. An algorithm could be developed to determine the toxicity of a piece of content, and the most toxic content could be added to a priority queue for priority moderation.

What are the consequences of this approach? How does the definition of toxicity prioritize (or de-prioritize) certain content? Who does it benefit? Consider the positionality of the users that interact with the platform:

• marginalized users of the platform, who may be further marginalized by this definition of toxicity.

• content moderators, who are each reviewing several hundred pieces of the most toxic content for hours every day.

• legal teams, who want to mitigate government regulations and legislation that are not aligned with corporate interests.

• social media hackers, or users who want to leverage the way certain content is prioritized in order to deliberately shape public opinion.