Anything they're capable of picking up _isn't_ what you're actually concerned about as an interpeter. Sure they're good at picking reflectors in the shallow portion of the volume. No one cares about picking reflectors. That's not what you're doing as interpreter.
A good example is the faults in that video. Sure, it did a great job at picking the tiny-but-well-imaged and mostly irrelevant faults. Those are the sort of things you'd almost always ignore because they don't matter it detail for most applications.
The faults you care about are the ones that those methods more-or-less always fail to recognize. The significant faults are almost never imaged directly. Instead, they're inferred from deformed stratigraphy. It's definitely possible to automatically predict them using basic principles of structural geology, but it's exactly the type of thing that these sort of image-focused "automated interpretation" methods completely miss.
Simply put: These methods miss the point. They produce something that looks good, but isn't relevant to the problems that you're trying to solve. No one cares about the well-imaged portion that these methods do a good job with. They automate the part that took 0 time to begin with.