When ChatGPT emerged a year and a half ago, many teachers immediately worried that their students were using it as a substitute for their own written assignments, clicking a button on a chatbot instead of thinking about responding to an essay. incite themselves.
But two English professors at Carnegie Mellon University had a different first reaction: They saw this new technology as a way to show students how to improve their writing skills.
To be clear, these professors, Suguru Ishizaki and David Kaufer, were also concerned that students could easily abuse generative ai tools. And it remains a concern.
However, they had an idea of how they could establish a unique set of guardrails that would create a new type of teaching tool that could help students incorporate more ideas into their assignments and spend less time thinking about formatting assignments. prayers.
“When everyone was afraid that ai would hijack students' writing,” Kaufer recalls, “we said, 'Well, if we can restrict ai, then ai can reduce a lot of the writing retrieval tasks that prevent students from doing so.' students really ) to see what is happening with their writing.”
The professors call their approach “constrained generative ai” and have already built a prototype software tool to test in classrooms, called myScribe, which is being tested in 10 courses at the university this semester.
Kaufer and Ishizaki were in a unique position. They've been creating tools together to help teach writing for decades. A previous system they built, DocuScopeuses algorithms to detect patterns in students' writing and display them visually to students.
A key feature of their new tool is called “Prose Notes,” which can take stray bullet points or stray thoughts written by a student and turn them into sentences or paragraph drafts, thanks to an interface for ChatGPT.
“One obstacle in writing is sentence generation: turning ideas into sentences,” Ishizaki says. “That is a big task. That part is really expensive in terms of cognitive load.”
In other words, especially for beginning writers, it is difficult to think of new ideas and take into account all the rules for writing a sentence at the same time, just as it is difficult for a beginning driver to follow the surroundings of the road and the mechanics. of driving.
“We thought, 'Can we really lighten that load with generative ai?' he says.
Kaufer adds that novice writers often change too early in the writing process and fragment ideas into carefully crafted sentences, when they may later end up deleting those sentences because the ideas may not fit into their final argument or essay.
“They start polishing too early,” says Kaufer. “And so what we're trying to do is with ai, you now have a tool to quickly prototype your language when you're prototyping the quality of your thinking.”
He says the concept is based on writing research from the 1980s that shows that experienced writers spend about 80 percent of their writing time thinking about plans and organization of the entire text rather than sentences.
Tame the chatbot
Teachers say developing their “prose notes” feature took some time.
In their first experiments with ChatGPT, when they put in some fragments and asked it to form sentences, “what we found is that it starts adding a lot of new ideas to the text,” Ishizaki says. In other words, the tool tended to go even further in completing a trial by adding other information from its vast stores of training data.
“So we came up with a very long set of guidelines to make sure there are no new ideas or new concepts,” adds Ishizaki.
The technique differs from other attempts to focus the use of ai in education, in that the only source that the myScribe robot draws from is the student's grades rather than a broader data set.
Stacie Rohrbach, associate professor and director of graduate studies at Carnegie Mellon School of Design, sees potential in tools like those created by her colleagues.
“We've long encouraged students to always make a solid outline and say, 'What are you trying to say in each sentence?' she says, and hopes that “narrow ai” approaches can help in that effort.
And he says he already sees student writers misusing ChatGPT and therefore believes some moderation is needed.
“This is the first year I've seen a lot of ai-generated text,” he says. “And ideas are lost. The sentences are written correctly, but it ends up being gibberish.”
John Warner, an author and educational consultant who is writing a book about ai and writing, says he wonders if the myScribe tool could completely prevent ai chatbot “hallucinations,” or cases where the tools insert information wrong.
“People I talk to think that's probably not possible,” he says. “Hallucination is a feature of how large language models work. The large language model lacks judgment. You may not be able to help but invent something. Because what do you know?
Kaufer says his tests have worked so far. In a follow-up email interview he wrote: “It is important to note that 'prose notes' operate within the confines of a paragraph unit. This means that if you were to exceed the limits of the notes (or “hallucinate”, as you say), it would be readily apparent and easy to identify. “The concern about ai hallucinations would be amplified if we were talking about larger speech units.”
Ishizaki, however, acknowledged that it may not be possible to completely eliminate ai hallucinations in his tool. “But we hope to be able to constrain or guide the ai enough to minimize 'hallucinations' or inaccurate or unintentional information so that writers can correct them during the review process.”
They described their tool as a “vision” of how they expect the technology to develop, not just as a single system. “We're setting the goal of where writing technology should progress,” she says. “In other words, the concept of prose notes is integral to our vision for the future of writing.”
Even as a vision, though, Warner says he has different dreams for the future of writing.
One technology writer, he says, recently pointed out that ChatGPT is like having 1,000 interns.
“On the one hand, 'Awesome,'” Warner says. “On the other hand, 1,000 interns are going to make a lot of mistakes. Interns from the beginning cost more time than they save, but the goal is that over time that person supervises less and less and learns.” But with ai, he says, “monitoring doesn't necessarily improve the underlying product.”
In this way, he maintains, ai chatbots end up being “a very powerful tool that requires enormous human supervision.”
And he argues that converting notes to text is, in fact, the important human writing process that must be preserved.
“Many of these tools aim to make a process efficient that does not need to be efficient,” he says. “Something huge happens when I go from my notes to a draft. It's not just a translation: these are my ideas and I want them on a page. It's more like: these are my ideas, and my ideas take shape as I write.”
Kaufer is sympathetic to this argument. “The thing is, ai is here to stay and it's not going away,” he says. “There will be a battle over how it will be used. “We are fighting for responsible uses.”
