What I meant to say was that such a rich interactivity should not be the default experience for simple and common cases. For simple and common cases, the technology should just work automatically without requiring much user interaction, thereby promoting usability and discoverability. However, you are absolutely right that for more sophisticated cases, instead of letting the user fall off the cliff, we should invest in a rich interactive experience where the AI can partner with the user to help complete more sophisticated tasks.
The thing is I'm a firm believer in putting the user at the centre of decision-making in automated processes, especially in AI tools that combine multiple sources of data without a clear of model of how they arrive to their solutions.
Too often these process provide their result as a foregone conclusion, and leave the user helpless if it is not the right one. By providing hints on how the process has arrived at that outcome, the user can form a mental model of how it works and learn how to use it more efficiently, or for which situations it is not suitable.
It annoys me that they research and apply complex algorithms to analyse the available data, detect the most likely ways to fill the document, build alternative programs to solve the problem and rank which is the most useful; only to then choose one of the inferred options and discard all the rest of the analysis work, getting only a (possibly incorrect) guess of the data completion task.
If, once the work is done, the user were allowed to see what the process was to detect the data, and were allowed to choose between the discarded options or give additional clues as to what the original intention was (the "active-learning session with the user to resolve the ambiguity" that the author of the article discards), the technique could be controlled more precisely and be useful in more situations, not just the ones that work the first time. Giving control to the user would make the technique more robust in my opinion instead of being hit-or-miss.