It seems like this particular alogrithm has two main components that are highly important, and human-oriented: a) The search space is limited with human-suppli

It seems like this particular alogrithm has two main components that are highly important, and human-oriented: a) The search space is limited with human-supplied options. b) The output isn't deterministic, is offers a series of best guesses. I would say this is closer to a NN-powered spellchecker than an authoritative source on the restoration of the work. It still requires a decent amount of interaction from an actual expert for the system to work.