Fairness in Socio-technical Systems: a Case Study of Wikipedia

Bias in Wikipedia stems from complex interactions between people, policies, and systems rather than single algorithms. The paper reviewed 75 studies on Wikipedia bias and found that observed unfairness cannot be traced to one automated decision-making system, highlighting how biases emerge from the socio-technical ecosystem as a whole.

Different types of bias in Wikipedia require different fairness solutions. The researchers classified various biases and found that existing fairness criteria designed for machine learning systems don't directly apply to Wikipedia's problems, suggesting that one-size-fits-all fairness approaches are insufficient for complex platforms.

Understanding causal relationships between biases is essential for addressing them responsibly. By analyzing how different biases relate to and reinforce each other in Wikipedia's socio-technical system, the paper demonstrates that tackling fairness requires understanding these interconnected causes rather than treating symptoms in isolation.

Fairness expectations must be tailored to specific harms and contexts. The study identified different normative expectations of fairness associated with different problems in Wikipedia, suggesting that responsible AI governance requires context-specific definitions of fairness rather than universal standards.