My point is, that if you trained an AI on Parisian French, people in Quebec would be annoyed but could adapt to the differences if they wanted to use the system

My point is, that if you trained an AI on Parisian French, people in Quebec would be annoyed but could adapt to the differences if they wanted to use the system. People in southern Germany may be annoyed they have to speak high German, but they know it. There are very few people, if any, who only understand Cajun French and the issue. Also, the biggest difference would be the pronunciation differences vs words being different. The word differences are easily modeled after any ASR, if the developer understands the differences and plans for it in their NLU models.