Can AI avoid offending someone with a strong opinion?