Generative AI (GenAI) often reproduces and amplifies societal stereotypes, such as portraying software engineers predominantly as young, white males. Such outputs reflect biases embedded in the training data and are rooted in historical discrimination and systemic inequalities. Addressing these biases requires more than technical fixes; it demands a socio-technical perspective that combines algorithmic approaches with a critical understanding of the social contexts and ideologies in which these technologies operate. We argue that software engineering (SE) researchers and practitioners should engage with constructs and theories from the social sciences to study, explain the persistence, and support the mitigation of bias in GenAI. We collate and discuss social theories relevant to SE research and practice, highlight risks associated with blind theory transfer, and call for more systematic interdisciplinary collaboration between SE and the social sciences. By carefully grounding GenAI research into relevant social perspectives that fit the SE context in the age of AI, the SE community will be better equipped to design inclusive tools, develop equitable solutions, and articulate their societal impacts.

@inproceedings{paperbot12000,
  title = {Exploring Societal Biases in Generative AI using Social Science Constructs and Theories},
  author = {Muneera Bano and Rashina Hoda and Didar Zowghi},
  booktitle = {ICSE},
  year = {2026},
  doi = {10.1145/3786581.3786924}
}