As a Help-Desk or Tech Forum, AI has Serious Reliability Problems

When you have been on this earth as long as I have, and when you have been using computers as long as I have (I venture to say that both of those conditions relate to more than 50% of the population), you have the "opportunity" to search out answers to hundreds of "why and how" questions. And these days, not just questions about computers, but questions about platforms and all other things related to your on-computer and online life.

Currently I am enamored of and intrigued by the possibility of AI improving my computing and online life by dramatically improving my "hit rate" for usable information and instructions when I am searching for the "how and why" of the way things work, or should work. Alas, just as I suspected, in these early stages of AI, its use as a help-desk or user forum are dismal.

Why?

The reason is simple. At this point in the AI game, there seems to be no AI product that is able or programmed to try out the instructions that it mines from just about everything it can find in the ethersphere and regurgitates to the Prompter, including every iteration of every problem associated with every iteration of an application or platform, and how to fix it or navigate it. A simple example:

(Gemini,) How can I turn off all notifications for all messages and notifications from LinkedIn?

Four add-on Prompts later—each of them pointing out that the Gemini response was simply wrong—and four responses later—each of them conceding that the answer was wrong and was based on legacy information for previous iterations of LinkedIn—Gemini finally got it right.

In one of my prompts, I mentioned that AI, in its current state, often seems like one great experiment in GIGO (Garbage In, Garbage Out, for the young). Gemini agreed with me on that. 

"It is a fair point that the "Garbage In, Garbage Out" phenomenon remains one of the most frustrating hurdles in AI. When platforms like LinkedIn change their UI—which they seem to do with exhausting frequency—older training data can become a liability rather than an asset. I appreciate you sticking with it and providing the "boots-on-the-ground" correction to get the instructions right."

I have been an Ubuntu Linux user almost exclusively for going on 12 years now. One of the oft-touted so-called "advantages" of Linux is its User Groups. But it ain't so, Joe. Linux user forums were, prior to AI (and still are) one of the greatest GIGO breeding grounds in the history of computer-dom. And those forums, along with the web pages and youtube sites of many self-proclaimed experts—whose expertise most often is based upon very loose guesses and speculations arising from some unrelated past personal experience, or from some idea that did not pop into their heads until you expounded your problem-—are the feeding grounds for AI.

In short, 90% of the comments and suggestions in user groups are based on guesses and speculations, and not on the commenters having actually used the suggested solution to fix an issue. And therein comes the rub with using AI as a substitute for the mostly-useless user groups: AI is mining the user groups for mostly-useless answers, then serving them up as solutions to AI users' prompts. 

Until AI can be programmed to either reject bad and outdated advice (the former (bad advice) unlikely), or programmed to try out its own advice before revealing it to the prompter (and who knows  what kind of latency that might cause), AI users will benefit or be poorly-served in direct proportion to the AI's ability to account for outdated instructions and poor advice.