Margaret Mitchell, an AI ethics researcher at Hugging Face, tells WIRED about a new dataset designed to test AI models for bias in multiple languages.