Ensuring Equitable Guidance: A Context-Based Approach to Bias Detection in AI Advisors

Nowadays, the increasing use of virtual conversational assistants raises important societal and ethical challenges. In fact, these systems can reflect and amplify social biases present in the dataset they are trained on, potentially leading to unfair or discriminatory behaviour in their interactions. Biases can heavily influence decisions of agents used as advisors, in areas such as employment and education. As a result, they can limit opportunities and reinforce social inequalities. This contribution introduces a methodology for detecting social biases in AI dia-logues by taking into account the specific conversational context. As a case study, the emergence of social bias in an academic advisory practice is considered. The proposed methodology, along with the required resources and knowledge structures, relies on a social theory-based model that considers social context as a first-order construct in agent deliberation.