Is ChatGPT a job killer or a helpful tool? Many people have posed technical questions to the artificial-intelligence (AI) text-generation engine called ChatGPT. The results have been questionable at best.
The GPT in ChatGPT stands for ‘generational pre-trained transformer,” which is how a large language mode (LLM) can be described.
There are many AI platforms out there including graphic AI platforms such as Jasper. Animated-graphics AI also exists with platforms such as Midjourney. We are being inundated with all levels of AI in our personal and professional lives. We must get used to it.
I am only interested in whether I will still be a valued columnist at Control Design. That’s important to me.
The main benefit of a conversational AI platform is its ability to figure out which words should follow the previous word(s). Think of this AI as a giant auto-correct engine that can string words together in a very meaningful way.
Writing a review of Shakespeare’s “A Midsummer Night’s Dream” would be as easy as pie. But the fact that ChatGPT knows about ladder logic is scary, especially when you see some of the results from people asking it to write code for a simple start-stop circuit with anti-tie down protection. It fails miserably, but it tried.
One of the things that I already know that ChatGPT and other AI can’t do is relate to an automation reader with experiences that I’ve had over my 45-year career. I also wonder if it could write a novel like Paul Gruhn’s “Refinery,” which is fiction but relates to real-world issues and experiences of a refinery in disarray. Safety is the main issue, coupled with the company’s resistance to spend money to fix issues.
Understanding that an AI-generated paragraph uses the library of the world to extract information upon which it has been trained, the results you would get are very scary in the accuracy of relatable facts.
In a simple world, AI can be very useful. Like a robot in manufacturing doing mundane work, ChatGPT can write cover letters and email text for routine issues. Use of this toolset isn’t an issue in anyone’s world.
In our industrialized environment where expertise and details are required, I am not convinced that using AI to low-code or no-code solutions to complex problems should be utilized.
The current crop of AI LLMs is all about search, prose and collecting ideas. Cybersecurity firms are using AI to plow through router data to find anomalies, which is very useful. You can see that the breadth of applications runs wide.
Because I’d written a column on the topic more than 10 years ago, I asked ChatGPT this: “I need a 700-word article describing the difference between Microsoft Access and SQL for industrial SCADA applications.”
It generated the word count and used the words “Access” and “SQL” many times. It compared the cost, integration issues and data integrity. The framing of the article really is a comparison of the constructs of the databases and not really any application. It certainly didn’t reflect on the usage in the supervisory-control-and-data-acquisition (SCADA) market such as Rockwell Software structured query language (RSSQL). It didn’t even reflect on the use of Java in SQL environments. Should it have? I did.
Could the generated text have been used for my column? An absolute no. I have a writing style, and I include real-life experiences where I can. I connect ideas with people because of the things that I know, which AI knows nothing about. This is important.
So, I changed my request to indicate that I wanted to be the author suggesting that it would find out my writing style. The result did change, ever so slightly. By and large, the output was the same.
I asked ChatGPT to “describe a normal PLC fault routine.” I was looking for something specific from it but got a very generic troubleshooting tutorial. It was a very good tutorial, by the way, but not what I was looking for.
The questions you ask must provide the specifics in a very detailed fashion. And asking multiple times will get the engine to continually refine its responses.
“Describe the octal numbering system” was the next question. To generate this information for a training course would be time-consuming, but ChatGPT produced a very useful description and explanation in roughly 20 seconds.
You couldn’t even find the book to research in that amount of time.
I determined that my position in the literary annals as an industrial magazine columnist is pretty safe for now. Academia however might want to be a bit more concerned. Book smart doesn’t cut it anymore. There’s a new kid in town.