For example, lets say you write an API that translates text. Should you assume the text the human sent in is correct and return an error? That way you can make better translations when they send the right data. Or should you try to be helpful and automatically fix spelling errors you think you've found in order to make the API more human friendly? Would make it easier but reduce accuracy.

You seem to be in the second camp, reduce power of API and make them easier to use. But lots of people think that the best way to serve humans is to make more powerful and strict API's.