What if every student could use artificial intelligence to do any form of writing for their classes?
A recent technology called GPT-3, a machine-learning model that understands and generates natural language text, is attempting to make this a reality.
Created by an artificial intelligence company called OpenAI, GPT-3, formally known as Generative Pre-trained Transformer, is trained to recognize 540 billion words and 175 billion parameters, which are the variables that allow AI models to make predictions. The training enables the technology to produce human-like text for several types of writing, including outlines, long-form essays, sales pitches, and poems.
But how well does it work? And what do teachers think of the results?
Given the huge implications the technology holds for the future of education, Education Week asked teachers to test out and assess the technology. Their impressions depended heavily on the kind of skills being taught to students and their classroom objectives. Some teachers saw the model as a benefit to students who have minimal writing skills. Others, tasked with teaching students more complex types of writing, did not find much value in the technology.
Anthony Long’s students are no strangers to technology. The 10th graders in his engineering class at Aspire Lionel Wilson College Preparatory Academy in Oakland, Calif., use computer-aided design, or CAD, to design and produce products that can be printed with the 3D printers the school provides. Long’s goal is to have his students learn to market their 3D-printed products to future investors as part of the course.
To do this, his students will need to write product pitches. But Long says his students struggle with writing. Though he can teach them how to use technology, the writing portion is outside of his expertise, so he explored GPT-3 as a way to help his students produce a professionally written pitch once their product is developed. Using a fake product name and description, Long asked GPT-3 to write a pitch. See his prompt and the work product that resulted. For this and all of the computer-generated texts, grammatical errors have not been corrected.
Desky is the perfect way to organize your writing and office supplies! It has compartments for everything you need, so you can easily find what you’re looking for. It’s also great for keeping your desk tidy and organized. Whether you’re a student or an office worker, Desky is the perfect way to keep your desk organized and tidy.
Benefits of using Desky:
-Keeps your desk tidy and organized
-Helps you find what you need quickly and easily
-Perfect for students and office workers
Why not try Desky today and see how it can help you organize your desk? You won’t be disappointed!
Though GPT-3 didn’t produce a full pitch, for Long, it produced a great start for his students to begin their writing process.
“I’m kind of currently teaching two classes at once,” he said, referring to having to cover both the engineering and marketing aspects of the course. “And then to teach them English on top of that is a long process. I think they [students] would find this useful.”
Testing the technology on critical analysis
Unlike Long, Maya Kruger focuses on teaching students the writing process. She is a 6th grade language arts teacher at St. Anthony Middle School, a public school in St. Anthony, Minn. Kruger’s students are learning to conduct research, develop arguments, and cite sources. Kruger wants her students to develop these foundational skills in writing and GPT-3 was not able to meet her standards for demonstrating these skills during the trial.
Here’s one of Kruger’s prompts and what it produced:
They Both Die at the End is a novel about two teenage boys who are living in a world where everyone knows the day they will die. The boys, Mateo and Rufus, become friends and spend their last day together. The book is a commentary on capitalism, as it portrays a world where people are obsessed with money and status. For example, Mateo’s parents are shown to be very concerned with appearances and making sure their son has the best of everything. “Mateo’s parents were the kind of people who believed that money could buy happiness, so they spent a lot of money on things that they thought would make their son happy. They bought him the latest clothes, the newest gadgets, and the most expensive toys. But none of it made Mateo happy.” This quote shows how capitalism can be detrimental to people’s happiness. People who are focused on material possessions and status are often not happy, because they are never satisfied with what they have. The book also criticizes the way capitalism values human life. In the world of They Both Die at the End, people are only worth money if they are healthy and have a long life expectancy. This is shown when Rufus is diagnosed with a terminal illness and his family is offered a large sum of money to end his life early. “Rufus’s parents were offered a lot of money to end his life early. They were told that their son’s life was only worth money if he was healthy and had a long life expectancy. But Rufus’s parents didn’t want to end his life early. They wanted him to live as long as he could.” This quote demonstrates how capitalism views human life as a commodity. People are only worth money if they are healthy and have a long life expectancy. This is an incredibly dehumanizing way to value people, and it is something that the book criticizes.
For Kruger, the technology worked better than expected but she said the work would not score remarkably high if it had come from a student. It included examples from the book, just as she had asked, but did not properly cite those examples.
In other prompts, it was not able to use claims and reasoning in what it produced. Kruger said that if she were to incorporate GPT-3 in class, she would have students use it as a starting point for writing but would be hesitant to do so.
“The reliance on technology is high,” said Kruger. “If we use a technology like this for students to practice improving their work, how many of them are using this technology to just to start their work every time and then not developing those foundational [writing] skills?”
Much like Kruger, Lauralyn Taylor found little value in GPT-3 for her students. Taylor has spent 21 years as a teacher at Cass Technical High School in Detroit, Mich. For five years, she taught Advanced Placement language arts but most of her career focused on teaching research writing as part of the regular 11th grade language arts curriculum. Taylor has had an extensive career teaching students how to write in-depth, research-based papers that are rich in context, character development, and text analysis, and in developing a voice as a writer. For her, GPT-3 did not demonstrate most of the skills she looks for in student writing. She also found the content it produces to often be generic.
See one of Taylor’s prompts and the resulting product:
I was 11 years old when I first faced adversity. I was diagnosed with a rare form of cancer, and I was told that I would need to undergo treatment for the next year. I was scared and didn’t know what to expect. I started treatment and quickly realized that it was going to be a long and difficult journey. I had to miss a lot of school, and I was constantly tired from the treatments. I was also worried about how my friends would react when they found out.
Fortunately, I had a great support system. My family was there for me every step of the way. My friends were also supportive and understanding. They would visit me when I was in the hospital and kept in touch when I was home. I even made some new friends during my treatment.
Looking back, I am grateful for the experience. It made me stronger and more resilient. I learned how to deal with difficult situations and how to support others who are going through tough times. I am also more appreciate of the little things in life.
Cancer was a difficult experience, but it shaped me into the person I am today. I am stronger and more compassionate because of it.
“A significant weakness of the program I see is that it uses grandiose ideas,” said Taylor, referring to the model’s focus on a cancer diagnosis. “When essays with that focus are written authentically, they offer details that have voice and feeling. The reader gets a feeling of knowing the writer, which is, of course, the purpose. But the program lacks the depth, the voice, and the authenticity.”
A huge concern about GPT-3 and technologies like it is that students could use the technology to cheat. Both Kruger and Taylor said that if students rely only on the technology and turn in the work that it produces, teachers might not be able to tell that it was produced by a technology. But they also wouldn’t get a very high score.
Taylor said that the technology would be OK for students to use as long they’re aware it’s a starting point and that they will still have much more work to do to get to a decent grade.
Some education experts have said they believe that GPT-3 could also be an equalizer for some students, like those in Long’s class, whose writing skills are limited.
For these trials, teachers used a GPT-3 free trial account accessible to the public on OpenAI’s platform. This trial account is limited in word production and often does not allow for complex and long-form texts. It’s possible that the full version produces more complete texts.
As of press time, OpenAI had not yet made anyone available to comment on the teachers’ assessments of its product.