This is a great example of how, from an ML/NLP standpoint, Instant-Search promoted confirmation-bias. If the idea is to learn how a user searches, that user should not be told how they should search.
Hmm, is it not just a matter of turning it off? Or are you doing something more advanced than using the standard Google search page?
My problem is I look at keywords from the results and want to use them to refine my search, and as soon as I type the first letter of my refinement, I lose access to the words I need.
And as you say, results on half words have never helped anyone, and there are already suggestions for what these half words might expand to.
If you have an account it stays off? I always hated instant and disabled it, and never had to disable it again...?
I have never once had my search experience enhanced by this anti-feature, and most of the time it's a hindrance, because I'll start reading the search results, then based on that come up with a new query and start typing it while continuing to scan the page. In non-instant world, maybe 60% of the time I'll end up finding what I want to click on as I continue to scan the page, then stop typing and click on that (or I'll be reading the name of someone that I want to google from the except from the page). In instant world, I have to read the whole page, decide if I need to make a new query, then type it (which is the exact opposite of the intent of instant, mind you).
The worst thing about Instant was that if you started typing something that didn't have results it wanted to display (because maybe it was a NSFW-type query or because it just didn't have good results for the incomplete string), rather than leaving the old page up until it had something it wanted to show, it would just blank the old page. Horrible.