In-group bias in the context of AI refers to the ways in which AI systems can unintentionally or intentionally exhibit biased behaviors that favor certain groups over others, often based on the data they are trained on. This bias can emerge from training datasets that reflect existing societal biases, leading to AI models that reinforce rather than mitigate inequality when making predictions or decisions. As AI systems are increasingly integrated into real-world applications, addressing in-group bias is critical to ensure fairness, accountability, and trustworthiness, and is an area of focus for AI ethics and policy discussions. Techniques for mitigating such biases include careful design of training datasets, bias detection algorithms, and ongoing monitoring of AI decisions in diverse environments.
The concept of in-group bias originated from social psychology with its roots tracing back to research in the 1970s, but the specific application to AI gained more attention in the late 2010s as AI technologies became more ubiquitous and concerns about algorithmic fairness grew more pronounced.
Key contributors to understanding and addressing in-group bias in AI include researchers and institutions focused on AI ethics and fairness, such as the AI Now Institute and scholars like Timnit Gebru and Joy Buolamwini, who have highlighted the real-world impacts of biased AI systems on marginalized groups.