Uplifting Lower-Income Data: Strategies for Socioeconomic Perspective Shifts in Large Multi-modal Models
This addresses bias in AI models for underrepresented socioeconomic groups, but it is incremental as it builds on existing prompting techniques.
The paper tackled the problem of biased Large Multi-modal Models due to unequal representation of socioeconomic groups in training data by proposing and evaluating prompting strategies using geographic and socioeconomic attributes, resulting in improved model performance on lower-income data.
Recent work has demonstrated that the unequal representation of cultures and socioeconomic groups in training data leads to biased Large Multi-modal (LMM) models. To improve LMM model performance on underrepresented data, we propose and evaluate several prompting strategies using non-English, geographic, and socioeconomic attributes. We show that these geographic and socioeconomic integrated prompts favor retrieving topic appearances commonly found in data from low-income households across different countries leading to improved LMM model performance on lower-income data. Our analyses identify and highlight contexts where these strategies yield the most improvements.