Surely this will depend on what your software does and the data it works with.
Huge amounts of software, particularly embedded software, doesn't have any sort of single structured database or a simple set of data files that you could just dump. Even where you do have these things, the dump won't be worth much unless the DB schema or file structure can be successfully interpreted in order to reconstruct the meaning of the underlying data within some other system.
That might or might not be easy, depending on how complicated the original data formats were. If it's something common and widely understood, perhaps some text a user wrote or a photo they saved, that's one thing. But what about something more specialised? Suppose you're moving to a new home. How do you export the training data from the AI in your old smart front door lock, which now recognises your friends and family's faces, so you can set up the smart locks on your new home to recognise the same people?
Unless there are industry standards for representing whatever data you're working with, providing portability between applications or devices that offer similar functionality from different manufacturers/developers is always going to be a problem. Encouraging standardisation is certainly beneficial here, but imposing some sort of blanket obligation on the manufacturers/developers to provide import/export functionality in the absence of any relevant format probably isn't.