Why AI Is W.E.I.R.D. And Shouldn't Be This Way

Recent years have witnessed remarkable advancements in AI, with language and vision models that have enabled progress in numerous applications and opened the door to the integration of AI in areas such as communication, transportation, healthcare, and arts. Yet, many of these models and their corresponding datasets are W.E.I.R.D. (Western, Educated, Industrialized, Rich, Democratic) and they are reflective of a small fraction of the population.(*) In this talk, I will show some of the limitations and lack of representation of current AI models, especially for commonsense and contextual knowledge, and highlight the need for cross-cultural language and vision models that can capture the diversity of behaviors, beliefs, and language expressions across different groups. I will also explore ways to address these limitations by developing models that are re-centered around people and their unique characteristics.

(*) W.E.I.R.D. is  an acronym widely used in psychology to indicate the limitations of many of the studies carried out in the field