How can we extract actionable meaning from image data?