How Much of a Threat are the New AI Text Generators?
You’d have to be living under a rock not to know, by now, AI text generators are the talk of the town. Principal among them is a model called ChatGPT from Open AI. Here’s how the developer describes their product:
“We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response.”
You can learn more about the model here:
https://openai.com/blog/chatgpt/
It’s also possible to try it, but more about that below (along with an example).
What appears to have many in the educational and literary fields upset is the possibility that soon (if not now!), we will be reading stories, novels, plays, and other documents that are not generated by human authors, but instead, by computers using these new AI text-generation models. Consider the educator, who now has no idea whether the Senior Theme he or she is reading on a college application was actually written by the student applying for admission . . . or whether the academic or government grant application under review actually was written by the scientist whose name appears on the bottom line. How about the stories we read in our daily newspapers or weekly/monthly magazines? Are they human-generated or the product of a computer? And if the latter, what can be said about the veracity of the claims, facts, and/or assertions made?! Overall, what are we to believe going forward. Who are we to believe?
The field of AI is nothing new. The whole concept revolves around this question: can computers think? Alan Turing, who is credited with breaking the encoded messages enabled by the Nazi’s Enigma Machine in WWII, even proposed what generally is known as The Turing Test. The test, originally called the ‘imitation game’ by Turing in 1950, is a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. You can read more about it here:
https://en.wikipedia.org/wiki/Turing_...
My first encounter with AI was in In the fall of 1977, Radio Shack introduced the TRS-80 for $599 in the summer of that year, and a co-worker, who reviewed PCs for various magazines at the time, was sent one of these systems to tender his opinion. It was a meager system by today’s standards, to be sure: according to Radio Shack, the offering included a BASIC language interpreter, four kilobytes of RAM, a Zilog Z80 processor at 1.77 megahertz, a twelve-inch video monitor, a cassette recorder, a power supply, and a cassette tape containing the games Blackjack and Backgammon.
But it was enough for us to create a small program that produced haikus, a short form of Japanese poetry. For example:
“A Poppy Blooms” by Katsushika Hokusai
I write, erase, rewrite
Erase again, and then
A poppy blooms.
What my co-worker and I did, then, was to create five haikus using the TRS-80 and to mix them with five haikus written by human beings. The ten then were Xeroxed and handed to students in my youngest daughter’s grade school class. The students were asked to decide which ones were real and which ones were computer-generated. None got them all correct. The best score was six out of ten. I have to admit, some of the PC-generated haikus were stunning!
The field of AI advanced rapidly in an extremely short period of time from that point on. In fact, I led a team of programmers who successfully developed the first large-scale AI application for the US Navy as part of our work for the In-Service Engineering Agent-Aegis, Webster, Field, St. Inigoes, Maryland, 40 years ago this year. The program significantly enhanced the ability of US Forces—US Navy, Army, Air Force, Special Forces, and Coast Guard—to create effective communications plans by optimizing interoperability. It did so by taking into account the communication capabilities of the individual units scheduled to participate in any given action.
So, on to something more current: my test of ChatGPT,
Just to put a toe in the water with the model, I decided to “play it” against myself using the latest flash fiction challenge from Indies Unlimited. The photo prompt (copyright by K.S. Brooks) this week is shown below:
“Hide in Plain Sight”
Below is the story I created for submission. I already won a Weekly competition for the month of January, 2023, so the story below was submitted only for the Monthly Editors’ Choice competition.
“Well, Dean certainly has chutzpah, I’ll give him that,” Heather whispered under her breath as she and Sally watched him hug his old flame, Lorraine, outside the Saddle-Up Bar and Grill before the big game was about to begin Sunday afternoon.
“The guy’s got a real ‘set’ on him, all right,” Sally deadpanned.
“Oh, yeah. And to think he had the gall to accuse of her of cheatin’ on him last weekend when he got hammered, stumbled all himself like a stupid oaf, spilled beer on her, and, after passing out in her arms, left her standing alone on the dance floor in front of everyone.”
“What I really loved was the part where he finally sobered a bit later that night and ranted about her letting Rod drive her home.”
“You mean that little scene where he grunted at her: ‘You really shud go home wid da guy who brung you to da dance,’ ” Heather chuckled, aping Brian by swinging her arms from side to side, moving her shoulders up and down, and speaking in a low, street-slang voice.
The women laughed so hard they almost blew beer out their noses.
“It’s not funny,” Sally finally said, wiping tears from her eyes. “But hey, if you can’t have a little fun at a friend’s expense, who can you abuse?”
Now, here’s the story generated by ChatGPT…no fancy instructions, simply a straight-forward description of the photo prompt and a requirement that the ending have a “twist.”
Instructions: Write a short story of not more than 250 words about a man dressed in a gorilla costume embracing a girl in front of a tavern with a crowd around them. There must be a "twist" at the end.
It was a cold winter night and a young man dressed in a gorilla costume had just emerged from a tavern. The crowd around him was large and curious, though none could guess the man's intentions. The man strolled towards a young woman who had been standing in the street, watching the commotion from a distance. When he reached her, he slowly opened his arms and embraced her. The crowd gasped in surprise and some started to whisper about the couple. The man and woman stood there for what felt like an eternity, embraced in a silent hug. The tension in the air was palpable as the crowd watched in awe. When the man finally let go, the woman turned around to face the crowd with tears in her eyes. In a trembling voice, she announced that the man in the gorilla costume was her long lost brother. They had been separated when they were children and this was the first time they had seen each other in years. The crowd was stunned. They had expected a different kind of twist, but this one was even more powerful. Tears of joy and surprise filled the streets as the siblings reunited. The man in the gorilla costume had succeeded in his mission. He had found his sister and brought [to her and the gathered crowd something beyond anything expected during the joyous days leading to Christmas.]
For some reason, ChatGPT did not complete the story (the portion shown in square brackets). Regardless, for an effort that took less than a minute to specify, I posit it certainly would have met the requirements at the 9th grade high school level.
Oh oh…I’m not sure how educators and others are going to handle the problems such AI text generators present. One solution, at least at the grade and high school levels, is to require hand-written essays and themes (vs. computer-generated copy). Beyond that, there is a burgeoning field in the area of determining what is real (versus determining what is fake) … an area that already is showing promise.
Watch this space.
“We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response.”
You can learn more about the model here:
https://openai.com/blog/chatgpt/
It’s also possible to try it, but more about that below (along with an example).
What appears to have many in the educational and literary fields upset is the possibility that soon (if not now!), we will be reading stories, novels, plays, and other documents that are not generated by human authors, but instead, by computers using these new AI text-generation models. Consider the educator, who now has no idea whether the Senior Theme he or she is reading on a college application was actually written by the student applying for admission . . . or whether the academic or government grant application under review actually was written by the scientist whose name appears on the bottom line. How about the stories we read in our daily newspapers or weekly/monthly magazines? Are they human-generated or the product of a computer? And if the latter, what can be said about the veracity of the claims, facts, and/or assertions made?! Overall, what are we to believe going forward. Who are we to believe?
The field of AI is nothing new. The whole concept revolves around this question: can computers think? Alan Turing, who is credited with breaking the encoded messages enabled by the Nazi’s Enigma Machine in WWII, even proposed what generally is known as The Turing Test. The test, originally called the ‘imitation game’ by Turing in 1950, is a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. You can read more about it here:
https://en.wikipedia.org/wiki/Turing_...
My first encounter with AI was in In the fall of 1977, Radio Shack introduced the TRS-80 for $599 in the summer of that year, and a co-worker, who reviewed PCs for various magazines at the time, was sent one of these systems to tender his opinion. It was a meager system by today’s standards, to be sure: according to Radio Shack, the offering included a BASIC language interpreter, four kilobytes of RAM, a Zilog Z80 processor at 1.77 megahertz, a twelve-inch video monitor, a cassette recorder, a power supply, and a cassette tape containing the games Blackjack and Backgammon.
But it was enough for us to create a small program that produced haikus, a short form of Japanese poetry. For example:
“A Poppy Blooms” by Katsushika Hokusai
I write, erase, rewrite
Erase again, and then
A poppy blooms.
What my co-worker and I did, then, was to create five haikus using the TRS-80 and to mix them with five haikus written by human beings. The ten then were Xeroxed and handed to students in my youngest daughter’s grade school class. The students were asked to decide which ones were real and which ones were computer-generated. None got them all correct. The best score was six out of ten. I have to admit, some of the PC-generated haikus were stunning!
The field of AI advanced rapidly in an extremely short period of time from that point on. In fact, I led a team of programmers who successfully developed the first large-scale AI application for the US Navy as part of our work for the In-Service Engineering Agent-Aegis, Webster, Field, St. Inigoes, Maryland, 40 years ago this year. The program significantly enhanced the ability of US Forces—US Navy, Army, Air Force, Special Forces, and Coast Guard—to create effective communications plans by optimizing interoperability. It did so by taking into account the communication capabilities of the individual units scheduled to participate in any given action.
So, on to something more current: my test of ChatGPT,
Just to put a toe in the water with the model, I decided to “play it” against myself using the latest flash fiction challenge from Indies Unlimited. The photo prompt (copyright by K.S. Brooks) this week is shown below:
“Hide in Plain Sight”
Below is the story I created for submission. I already won a Weekly competition for the month of January, 2023, so the story below was submitted only for the Monthly Editors’ Choice competition.
“Well, Dean certainly has chutzpah, I’ll give him that,” Heather whispered under her breath as she and Sally watched him hug his old flame, Lorraine, outside the Saddle-Up Bar and Grill before the big game was about to begin Sunday afternoon.
“The guy’s got a real ‘set’ on him, all right,” Sally deadpanned.
“Oh, yeah. And to think he had the gall to accuse of her of cheatin’ on him last weekend when he got hammered, stumbled all himself like a stupid oaf, spilled beer on her, and, after passing out in her arms, left her standing alone on the dance floor in front of everyone.”
“What I really loved was the part where he finally sobered a bit later that night and ranted about her letting Rod drive her home.”
“You mean that little scene where he grunted at her: ‘You really shud go home wid da guy who brung you to da dance,’ ” Heather chuckled, aping Brian by swinging her arms from side to side, moving her shoulders up and down, and speaking in a low, street-slang voice.
The women laughed so hard they almost blew beer out their noses.
“It’s not funny,” Sally finally said, wiping tears from her eyes. “But hey, if you can’t have a little fun at a friend’s expense, who can you abuse?”
Now, here’s the story generated by ChatGPT…no fancy instructions, simply a straight-forward description of the photo prompt and a requirement that the ending have a “twist.”
Instructions: Write a short story of not more than 250 words about a man dressed in a gorilla costume embracing a girl in front of a tavern with a crowd around them. There must be a "twist" at the end.
It was a cold winter night and a young man dressed in a gorilla costume had just emerged from a tavern. The crowd around him was large and curious, though none could guess the man's intentions. The man strolled towards a young woman who had been standing in the street, watching the commotion from a distance. When he reached her, he slowly opened his arms and embraced her. The crowd gasped in surprise and some started to whisper about the couple. The man and woman stood there for what felt like an eternity, embraced in a silent hug. The tension in the air was palpable as the crowd watched in awe. When the man finally let go, the woman turned around to face the crowd with tears in her eyes. In a trembling voice, she announced that the man in the gorilla costume was her long lost brother. They had been separated when they were children and this was the first time they had seen each other in years. The crowd was stunned. They had expected a different kind of twist, but this one was even more powerful. Tears of joy and surprise filled the streets as the siblings reunited. The man in the gorilla costume had succeeded in his mission. He had found his sister and brought [to her and the gathered crowd something beyond anything expected during the joyous days leading to Christmas.]
For some reason, ChatGPT did not complete the story (the portion shown in square brackets). Regardless, for an effort that took less than a minute to specify, I posit it certainly would have met the requirements at the 9th grade high school level.
Oh oh…I’m not sure how educators and others are going to handle the problems such AI text generators present. One solution, at least at the grade and high school levels, is to require hand-written essays and themes (vs. computer-generated copy). Beyond that, there is a burgeoning field in the area of determining what is real (versus determining what is fake) … an area that already is showing promise.
Watch this space.
No comments have been added yet.
