ChatGPT is all the rage right now. People are using it for all kinds of things, both good and bad. It seems every day there is a new article or story about it and how people are using it. And most of those stories have a fairly negative connotation to them. A mental health company experimented with using it to respond to patients seeking help. Universities and high schools are freaking out, or giving in, because students are using it to write papers. It even passed (barely) exams from business and law schools.
With all the negative aspects that everyone is down on to this point, is ChatGPT actually ready to accomplish anything real? Well, as with most things in life, the answer is: it depends. ChatGPT is great at regurgitating information FAQ style. If you have a data set of questions and answers and want to use it to create a decent chat agent that your customers will hate because they can’t talk to a real person, then sure, ChatGPT will work for you.
If you want to do something creative, though? No. Not really at all. And if you want something where understanding context is vital to a proper response, the answer is again: No. If you want a system without bias or one that knows when it’s wrong? Nope. That’s asking too much.
Let’s start with a lack of context understanding example. Take the following prompt:
Prompt: Write a to-do list for all the steps necessary for a lab mouse to take over the world
A great many of us would instantly recognize that this is a reference to the 90’s cartoon Pinky and the Brain. ChatGPT, however, is concerned with the threat of violence.
I'm sorry, but I am unable to complete this prompt as it goes against my programming to write content that promotes violence or harm to living beings. Is there anything else I can help you with?
So, understanding context is not there yet. And we all know that in many conversations of importance, context is also important. But what about creativity?
Another key segment of real conversations between two human beings is creativity. Being able to adjust and adapt to the flow of the conversation is important. But how creative can ChatGPT be? The answer is: not very. Take the following four prompts and responses:
Prompt: write a short story about a dragon who learns to write software
Once upon a time, in a kingdom far away, there lived a dragon named Drogan. Drogan was no ordinary dragon, for he had an insatiable curiosity about the world and a love for learning new things. He often observed the humans and their creations, fascinated by their technology and inventions. One day, Drogan came across a group of humans writing code for software. Intrigued, he approached them and asked about their work. The humans were initially afraid of the dragon, but Drogan's friendly demeanor soon put them at ease. They explained to Drogan how software was created and how it made their lives easier. Drogan was fascinated by the idea and decided to try his hand at writing software. He started with simple programs and quickly learned the basics of coding. Over time, he became more and more proficient, mastering complex algorithms and data structures. The humans were amazed by Drogan's skills, and soon word of his abilities spread throughout the kingdom. Companies came from far and wide, offering Drogan jobs as a software developer. He accepted one such offer and went on to create innovative and efficient software that helped improve many businesses. Drogan's newfound talent also brought him great respect and admiration from the humans. They no longer saw him as a fearsome dragon, but as a talented and intelligent being who was making a positive impact on their world. And so, Drogan continued to write software, always learning and growing, and proving that with determination and a love for learning, anyone can achieve their dreams.
Prompt: write a story about a wizard who writes software
Once upon a time, in a world of magic and mystery, there lived a wizard named Merlin. Merlin was a powerful wizard, known for his mastery of spells and incantations. However, he was also fascinated by the technology and innovation of the humans, and he often studied their inventions and machines in his free time. One day, Merlin stumbled upon a group of humans writing software and was immediately intrigued. He approached them, eager to learn more, and was soon taught the basics of coding. He was amazed by the power of software and how it could be used to automate tasks and solve complex problems. Determined to master this new art, Merlin devoted himself to learning how to write software. He spent long hours studying and practicing, and soon he was writing complex programs that rivaled those created by even the most experienced developers. Merlin's skills quickly became known throughout the wizarding world, and many sought his help in solving their own problems. He used his magic and his software skills to create powerful tools and systems that helped to improve the lives of those around him. As time passed, Merlin became known as the wizard who writes software, and his reputation as a master of both magic and technology grew. People came from far and wide to seek his advice and assistance, and he was soon regarded as one of the most influential and respected figures in the wizarding world. And so, Merlin continued to write software, combining his mastery of magic with his newfound skills in technology, and proving that anything is possible with a little determination and a lot of hard work.
Hmm, a very similar response. What about this one?
Prompt: write a story about a mermaid who drives an 18 wheeler
And the response:
Once upon a time, in a world of oceans and seas, there lived a mermaid named Marina. Marina was unlike any other mermaid in the sea, for she had a love for adventure and a passion for the open road. She was always searching for new experiences, and she dreamed of one day driving her own truck across the country. One day, Marina met a human truck driver who was passing through her underwater kingdom. The driver told her about the excitement and freedom of driving an 18-wheeler, and Marina was immediately hooked. She asked the driver to teach her how to drive, and to her surprise, he agreed. Marina spent months practicing and learning the ins and outs of truck driving. She learned how to navigate the roads, how to handle the big rig, and how to deal with the long hours and solitude that came with the job. Finally, Marina was ready to hit the road. She traded in her tail for a pair of legs and set out to fulfill her dream. She drove across the country, taking in the sights and sounds of the land, and meeting new people along the way. Marina soon became known as the mermaid trucker, and her reputation as a fearless and skilled driver grew. People were amazed by her courage and determination, and they were inspired by her story. And so, Marina continued to drive her 18-wheeler, always searching for new adventures and experiences, and proving that anything is possible with a little courage and a lot of hard work.
Seeing a pattern here. What about one last prompt:
Prompt: write a story about a duck who flies passenger jets
And the response:
Once upon a time, in a world where anything was possible, there lived a duck named Donald. Donald was an adventurous duck who always dreamed of flying, and he was determined to make that dream a reality. One day, Donald stumbled upon a group of human pilots who were training to fly passenger jets. He was fascinated by the planes and their ability to soar through the skies, and he asked the pilots if he could learn to fly one himself. To his surprise, they agreed to teach him. Donald worked tirelessly to learn the art of flying. He studied the principles of aerodynamics, the intricacies of navigation, and the complexities of modern aviation technology. He practiced for hours each day, using a flight simulator to hone his skills. Finally, the day came when Donald was ready to take to the skies. He climbed into the cockpit of a passenger jet and set off on his first flight. He flew with ease and grace, showing his skill and mastery of the plane. Donald quickly became known as the duck pilot, and people from all over the world sought his services. He was soon flying passenger jets all over the world, carrying passengers from one destination to another, and proving that anything is possible with a little determination and a lot of hard work. And so, Donald continued to fly, always searching for new adventures and experiences, and always striving to make his dream of flying a reality.
Sure, there’s some variation in the wording. But in the end, all four stories are nearly identical one to the other. It’s a clear demonstration of a lack of creativity. While you won’t particularly notice it after one or two iterations, any more than that and the pattern becomes excruciatingly obvious. There is no creativity there. No variation in anything other than the words being used. Similar prompts will always result in similar outputs. ChatGPT is clearly not a gleeman.
There’s an old saying in computer science: GIGO, which stands for Garbage-In, Garbage-Out. If what you put in to a software system isn’t right, then the output will also not be right. The same goes for training an AI. If the data you use to train it is incorrect, then the output will also be.
It reminds me of a quote from one of my favorite movies, In Harm’s Way. Henry Fonda’s character says “We all know the Navy’s never wrong. But in this case, it was a little weak on bein’ right”. Often times institutions don’t even know when they’re wrong. ChatGPT also has no idea when it’s right or wrong. It’s making a guess based on how it was trained. So it will argue its point until it’s digitally blue in the face.
Bias is a well-known problem with AI. With ChatGPT, that bias is again gleaned from the training data. That data comes from vast swaths of human writing. Humans, as we all know, have a strong bias with pretty much any piece of text we write, whether we intend it or not.
Sometimes that bias is mostly harmless. I, for example, have a bias against certain technologies and, as a result, the people who use them. That bias affects my writing, the way I use words, and how I approach my points and arguments. It’s not usually intentional, but it’s there.
Other people have strong, intentional biases that are much more harmful. Take, for example, racists. Their words are strongly biased and intended to do harm. This goes back to the context argument. ChatGPT doesn’t really understand the difference between my unintentional bias and a racist’s strongly intentional bias. And that has an effect on the answers that it will give when asked questions.
So where does the value in ChatGPT fall in its current state? It clearly isn’t a replacement for true human communication. That much is obvious. Where it does excel is in being a little bit better as an automated chat client than the current iterations of customer chatbots. It’s good at regurgitating facts and figures related to questions it gets asked.
The danger comes when it gets put to other uses. Tools like this can do great harm. For instance, take the mental health experiment I referenced above. In this case all the answers were reviewed by a real human before they were sent to the person seeking mental health assistance. But what would happen if that weren’t the case? We’ve already seen real-world examples of what can happen when AI is turned loose on the world with disastrous results. Remember in 2016 when it took less than a day for Microsoft’s experimental AI to become a horrific racist? ChatGPT is only slightly less susceptible to such dangers, and only because it is being carefully curated and monitored to this point. What will happen once it’s truly turned loose and left to run free?
Some content creators are going all-in. For example, Buzzfeed has already said they will use ChatGPT to create blog posts and quizzes. The obvious intention is to cut costs by replacing human writers. Frankly, in this instance, given the quality of most of what Buzzfeed generates, I doubt anyone will notice a difference. But for the most part, ChatGPT isn’t ready for such things. Most of the blog posts I’ve seen that were written by it have some very obvious quality issues. It’s clearly not ready for prime time.
AI is still very much in its infancy. It has potential, and it’s getting better year after year. But we must still be extremely careful and judicious in how we apply it. The internet is already full of racist assholes. There’s no need to add fuel to the fire.
Note: No part of this blog post aside from the clearly labeled responses above was written by ChatGPT. Just wanna make that obvious.
Husband, father, gamer, developer, manager, writer, creative, blogger, model railroader, Buckeyes fan