Programmatically Identifying Cognitive Biases Present in Software Development

Mitigating bias in AI-enabled systems is a topic of great concern within the research community. While efforts are underway to increase model interpretability and de-bias datasets, little attention has been given to identifying biases that are introduced by developers as part of the software engineering process. To address this, we began developing an approach to identify a subset of cognitive biases that may be present in development artifacts: anchoring bias, availability bias, confirmation bias, and hyperbolic discounting. We developed multiple natural language processing (NLP) models to identify and classify the presence of bias in text originating from software development artifacts.