
I spend a lot of time at the senior center where I live (yes, I am old). Somehow, a buddy and I have been identified as the tech support here for the other seniors.
Today, a gentleman came in and asked us how to print something from an AI chat on a government website from his phone.
Not having a clue how to do this, I asked the AI that I pay for.
But that got me thinking about what happens when AI doesn’t have a clue.
I’m not talking about AI being wrong or making something up. I’m talking about AI simply being unable to answer a question because it has no reference material or data that it can access.
I’m sure you’re asking: how can AI be stumped by a question?
So, let me give you a personal example of where AI could not answer the question.
I have mentioned before that I am developing a personal password vault that has reached the point where I could release it as a very strong beta. It is hosted on GitHub, where I could make it available for download.
The application is very stable, but I have yet to develop the user help documentation.
Even though I used AI extensively during development, AI does not share information between individual users. The AI that helped me develop the application does not somehow pass everything it learned about my application along to every other AI user.
So, suppose I released the application today and someone downloaded it. That user runs into a problem and asks an AI, “How do I do this in MWPV?”
Where does the AI get the answer?
There is no help documentation for it to look up, and the information isn’t already stored in the AI model. Unless the AI has access to some other source containing that information, it simply doesn’t know.
Let me offer another example of this. I am a strong supporter of open-source applications. There are many free alternatives available that provide much of the same functionality as paid commercial software.
GIMP, for example, is an alternative to Adobe Photoshop, while LibreOffice provides an alternative to Microsoft Office. These are well-established applications that have been around for years, and there are countless websites, tutorials, forums, and other sources explaining how to use them. AI has plenty of information available to draw from when answering questions about these applications.
But what about little-known open-source applications that fill a specific niche and have no help websites, online documentation, or YouTube videos about them?They may have help documentation built into the application, but that documentation may not answer the specific question a user is asking. Even worse from the AI’s perspective, that documentation may only exist inside the application, where the AI has no way to access it.
In that situation, AI again would most likely not be able to answer the question.
It might understand what the application is designed to do. It might understand the operating system it runs on. It might even be familiar with similar applications. But none of that means it knows how this particular application works.
The information has to exist somewhere the AI can access. I am sure this can also occur with paid applications, but you get my point.
So, being the problem solver that I am, I began to wonder how I could overcome this problem with my own application.
The developer could create help documentation within the application and also publish that same documentation as a web page that AI could access. The developer could also create help videos on YouTube.
Will AI be able to answer the question now?
Well, that depends.
A lot of developers simply aren’t very good at explaining how to use their applications in a way that is easily understood. Help documentation can become too technical because it is written from the developer’s point of view rather than the user’s.
There is also the problem of familiarity. The developer has spent hundreds or even thousands of hours working with the application. Something that seems completely obvious to the developer may not be obvious at all to someone opening the application for the first time.
If the documentation doesn’t clearly answer the user’s question, giving that documentation to AI doesn’t magically fix the problem. Now, there is another solution to this. The developer could develop an AI specifically for the application.
That sounds great, but let’s come back to reality.
As a lone developer who is seriously strapped for money, I have neither the time nor the money to develop and maintain an AI specifically for my application. I could easily spend more time developing the AI help system than I would spend writing the help documentation in the first place.
And trust me when I say this: I am not the only developer who faces this dilemma.
There are thousands of small developers creating open-source, free, inexpensive, and niche applications. They may have the skills to create some really useful software, but that doesn’t mean they have the resources to build an AI system to support it.Before you ask why I haven’t developed the help documentation yet, my laptop recently went to the great scrap yard in the sky.
So, until I can get another development machine, the password vault will be doing what it was originally developed to do: serving as my personal password vault.
But is what I described above the only situation where AI might not have a clue?
Of course not.
There are probably countless situations where the information needed to answer a question simply isn’t available to the AI. It may be locked away in an application, poorly documented, never published online, or perhaps never written down at all.
This was simply one example that dawned on me today after someone asked me how to do something and, ironically, I turned to AI because I didn’t have a clue.
Sometimes AI doesn’t have a clue either.