"That explanation makes it easier to understand. Unfortunately, it’s impossible to know which parts of the database are protected by copyright and which aren’t; in that sense, there isn’t much that can be done."
There is something very easy to be done: You just do not use the model unless all of the material that went into it was properly licensed.
If you know that a box of stuff contains some stolen material, and you buy the entire box knowing that some of the content is stolen, it doesn't suddenly become morally just or acceptable or even legal just because you do not know exactly which parts are stolen.