Algorithmic Bias and the Power of Code: Investigating the Reproduction of Gender Structures in the Development of Computer Technologies

This study explores how everyday programming practices reproduce and amplify gender bias in software systems. By analyzing five widely used open-source repositories in the fields of recruitment and financial analysis (totaling over 200,000 lines of code), and conducting in-depth interviews with ten developers, we identified three common patterns of bias: rigidly coded binary gender labels, stereotype-driven feature engineering (such as vacancy years), and the negation of non-binary genders. Static code analysis marked these trends, while dynamic testing on the gender-balanced dataset showed that the false negative rate for female users could be up to 15 percentage points higher, and there was a 10% inflation in the risk score. The interview revealed that organizational pressures (tight deadlines, lack of built-in fairness tools, and lack of procedural guidelines) led designers to be unable to mitigate bias. Based on the above findings, we propose a method for classifying bias-prone code features, an integration strategy for automatically detecting bias in continuous integration, and targeted code review guidelines. This achievement not only provides empirical evidence of code-level unfairness but also presents practical suggestions for integrating fairness into the software development lifecycle, thereby promoting a more equitable social technology system.