Fuck the Algorithm: Conceptual Issues in Algorithmic Bias

Algorithms can themselves be sources of bias, not just the data they process. The paper challenges the common claim that "algorithms themselves cannot be biased," arguing that algorithms are a type of thing capable of producing bias independent of input data—a distinction critical for assigning responsibility when algorithmic systems cause harm.

"Bias" has multiple meanings that need clarification in technical discussions. The paper distinguishes between statistical bias (mathematical deviation from expected values) and moral bias (unfair discrimination), showing how statistical properties can translate into morally significant harms in real-world applications like hiring, policing, and medicine.

Documented cases show algorithms as direct sources of discrimination. Specific examples include recommender systems that shape media consumption patterns, academic search engines that influence citation practices, and the 2020 UK A-level grading algorithm that produced discriminatory outcomes—demonstrating that algorithmic bias occurs across multiple high-stakes domains.

Understanding what algorithms "are" matters for accountability. By clarifying that algorithms are a distinct category of thing (drawing parallels to "political artifacts"), the paper establishes a conceptual foundation for determining who bears responsibility when algorithmic systems cause discriminatory harm.

Recognizing algorithmic bias is essential for preventing discrimination. Clear conceptual understanding enables better decision-making about how to design, audit, and govern algorithmic systems to prevent algorithmic mediation of unfair discrimination.