Bias in Taxonomy Design and Application: Sources, Solutions, and Implications for AI Tools and Algorithms (TM09)

Although taxonomies and controlled vocabularies seek to organize information and make data retrieval more efficient and accurate, their creation, maintenance, and application can introduce various forms of bias, which can have downstream effects on artificial intelligence tools and contribute to algorithmic bias. Bias can be introduced from various sources, including cultural and linguistic bias, term framing, legacy systems, authority bias, over-simplification, and language limitations. Our three speakers will each present use cases within taxonomy, discussing their specific sources of bias, as well as the impacts on AI tools and algorithms, such as discriminatory outputs and decision-making. The speakers will conclude with a discussion of ways to mitigate the issues, including inclusive design and regular audits.