Why AI Is Not An Advance As A Search Engine
My only experience of AI thus far is its search, verbal, and image generation capabilities. I doubt that’s the most important feature of AI, or the most complex technical capability. But I have to say that in terms of the search and verbal capabilities, I’m underwhelmed. My reaction is similar to my take on fully autonomous unpiloted automobiles- it’s brilliant, up to that crucial last fraction of decisionmaking. I guarantee that AI search demands more precision from me than Google 1.0, where I’d usually be able to narrow a search easily by typing in a topic phrase and then adding in one or two unlikely keywords that would bring up search results on topics that might be quite disparate, but delimited in accordance to my instructions. From there, I could easily pick out the precise result I was seeking somewhere in the first page or two of results. Whereas AI search insists on trying to outguess me, and unless I find a way to supply it with a “proper” context, it throws up The Set Of All Keyword Search Results Most Popular With People Searching For The Most Usual Topics. The extra keywords I might add to a topic inquiry actually confuse the program, because it insists on “overthinking” the inquiry, and i end up faced with reams of extraneous irrelevance. I actually do better by giving the program fewer information cues.
The thing is, there’s a method in my additional keyword choices, and now the machine only reads it as madness. I might be seeking one particular news article where I recall those exact words showing up, and adding them would winnow out any result that didn’t contain them.*
Twenty years ago, the search engine would simply fetch for me. Now it wants to guess why I’m making the search. The programmers apparently think that’s helpful or even essential for improving search, because the sort of work they do leads them to imagine that everything gets sorted along some Ultimate One True Linear Path, to find the proper result. It isn’t helpful.
I’ve noticed this about all sorts of programmed computer features for years—there’s often an allergy to parallel processing, or to lateral leaps of logic that might provide shortcuts or alternate routes to achieve the results I’m seeking. I don’t want a computer that’s programmed to anticipate my intentions or refer to my archive to try to read my thoughts or any of the other narrow-minded pretensions that I notice from having to grapple with how wrong the programming can be in that regard. I just want Google 2001. I’ll do the rest.
(*For example, I used to find author quotes much more easily by recalling a unique word or phrase used in the quote that I was seeking, for example. I admit, that could often be an elusive quest. But now it’s practically a hopeless task. The machine wants to know the topic that the author was opining on. Sometimes I used to search for quotes with a couple of keyword phrases and find the one I was looking for, without even knowing the author. Nowadays, without knowing the author, forget it.)
