That said, you're still generally stuck between a rock and a hard place when it comes to a complicated type definition as the non-annotated alternative (e.g. in a dynamically-typed language) isn't necessarily any less obtuse.

Lots of people simply prefer to know at compile time that their assumptions were wrong than having to reverse-engineer how they were wrong from stack traces at runtime.