Evaluating the Cultural Relevance of AI Models and Products: Learnings on Maternal Health ASR, Data Augmentation and User Testing Methods

  • Oche Ankeli, YUX; Ertony Bashil, YUX; Dhananjay Balakrishnan, YUX; Serigne Fall, YUX; Oluchi Audu, YUX; Melissah Weya, YUX; Yann Le Beux, YUX

How do we ensure that AI systems are not only technologically advanced but also deeply resonant and beneficial for the specific communities they aim to serve? As AI becomes more integrated into critical sectors like healthcare, this question has never been more important.

This session by YUX explores that question through key learnings on evaluating the cultural relevance of AI models and products. Drawing from their work fine-tuning an ASR model in Wolof for maternal health, YUX shares insights on social norms, evaluation datasets, and user testing methods for assessing AI systems. They also introduce LOOKA, a platform designed to scale human evaluation in AI development.

次を見る