On 9 January 2025, the Financial Conduct Authority (FCA) published the latest in its series of research notes on bias in artificial intelligence (AI).

The FCA has been carrying out research in the area of AI bias to help deepen its understanding of AI and its potential impact on financial services, and to spark discussion on these issues. It hopes that the notes will be of interest to those who build models, financial firms, and consumer groups in understanding complex debates on building and implementing AI systems.  

This latest paper explores bias in a natural language context. It looks in particular at word embeddings – mathematical representations of words that capture their meanings and relationships to other words or phrases – which the FCA explains are a valuable tool for many natural language processing applications and continue to be widely used in industry as a cost-effective and easy-to-deploy alternative to large language models. The research seeks to uncover how biases in word embeddings could be identified and removed at source through current methodologies.

The paper finds that while it is possible to measure some aspects of language bias and mitigation techniques can remove some elements of gender and ethnicity bias, there are limitations to current methods. The FCA notes that it welcomes further research in this area, as well as the development of new techniques to mitigate and measure bias.