Disclaimer: I am not an AI specialist; however, I am an author, and I have some knowledge of computers and the basics of computer programming.

So, with all the hype about AI that has been going on over the last few years, I think that there are some things that people have forgotten, or just don’t realize, about Artificial Intelligence (as it is referred to) that plays a huge role in what is produced by AI.

First of all, we have to remember that AI are computers, and that they run on programs created by humans.  They are not actually sapient life; they are not truly intelligent. Calling them LLMs (Large Language Models) is much more accurate than referring to them with the term “intelligence”.

They are essentially trained on large amounts of text to predict what word follows what word.  They aren’t intelligent or creating new things, they are working off a set of percentages as to what comes next based on the information they have and what has come before.

All they can do is imitate; they can’t actually create.

This has a great many implications that most people ignore.

And that’s where the second issue comes in: GIGO. This is an acronym for “Garbage In, Garbage Out”, and is basically an axiom for computer programmers.  If a computer program receives input that is garbage, all it will output in turn is garbage.

And there is a huge amount of garbage on the internet, which is a lot of what has been used to train LLMs (AKA AI).

Sure, some information sources have been given more weight in terms of what LLMs should pay attention to, rather than others. But that doesn’t change the fact that a hell of a lot of the information that LLMs have been trained on is garbage.

We’ve all heard news stories about people relying on LLMs to do work for them, and then ending up in trouble because the LLM output was garbage, nonsensical, or just plain wrong.

Sure, you can acquire a ChatGPT or other “AI” bot and program it only with your own choice of material. But that still doesn’t necessarily mean that it will behave the way you want it to, or provide you with correct answers.

And it certainly doesn’t mean that it can communicate with humans in your place.

I refer you again to the way in which LLMs work. How many times has the predictive text on your phone been completely wrong for what you were trying to say? I bet it’s nowhere near zero. And while LLMs are (potentially) more sophisticated than the predictive text on your phone (I say “potentially” because both Apple and Google are incorporating LLMs into their phone systems, regardless of what the average user wants), it can still get things wrong far too easily.

We have to learn, as a society, not to expect actual intelligence out of so-called AI. Because it’s not intelligent.

It is a computer program that bases its responses on what information it has access to, and predictions for what word to choose next based on that information and the previous word or sentence.

This can go, and has gone, terribly wrong.

So remember. GIGO still applies to so-called AI, even if responses seem more like natural language. Even if they sound more human. There’s still far too much garbage imbedded in anything it has picked up off the internet, and far too much randomness in its way of answering questions or requests.

😐 tag