Indeed! Substring search with rudimentary Boolean would be good enough for me the vast majority of the time. The UI could be like this:
Find documents with words that
[Contain|v] [_____________]
[And|v] [Contain|v] [_____________]
[And|v] [Contain|v] [_____________]
[And|v] [Contain|v] [_____________]
[_] Case Sensitive [_] Search file name only
"|v" represents a drop-down list. The left-most drop-down would be And, Or, & Not, and the middle would be Contain, Start-With, End-With, & Equal. The Or's would be grouped together. For example:
Words that [Equal] AA
[Or] [Equal] BB
[Or] [Equal] CC
[And] [Equal] DD
Would be "show all documents having the word DD and at least one word of AA, BB, or DD."
I've coded UI's like this multiple times, and they rarely confuse users (after experimenting a little).
If MS merely perfects bicycle search science, we don't need rocket science. Industry is too eager to shove in buzzwords rather than improve what mostly works, making bloated buggy messes. The most annoying excuse is "but it's not enterprise/web-scale". Most projects are not giant. Why bloat up the 98% for the 2% biggies? The web-scale thing is probably not AI-related, but you get the idea. Like AI, web-scale trigger's me-too-ism. I'm putting your whippersnappers on a buzzword diet until you git off my lawn!
(This is Monday, so there's probably a Major Typo. Yes, that is partly a military joke.)