UCI Researchers Win Grant to Help Educators Use AI to Teach K-12 Writing
PapyrusAI leverages emerging technology to help students master the art of writing
By Christine Byrd
March 8, 2024 UCI School of Education researchers recently won a grant to help launch a product aimed at making it easier – and safer – for K-12 students to use artificial intelligence to improve their writing skills. With a $79,500 Proof-of-Product grant from UCI Beall Applied Innovation, researchers begin piloting the program, called PapyrusAI, in middle school classrooms this spring. When OpenAI released ChatGPT to the public in 2022 it was met with strong reactions from educators. Some school districts initially banned the technology, but educators have warmed to it, with more than 60 percent of teachers having used ChatGPT, according to a survey last year. |
“As people have tried ChatGPT and talked more and more as an education community, there’s interest in seeing how to use it well,” said Tamara Tate, associate director of UCI’s Digital Learning Lab who specializes in writing pedagogy.
Generative AI like ChatGPT could solve significant challenges for K-12 writing teachers: with each teacher having 150 or more students in middle or high school, it’s nearly impossible to provide personalized feedback at every stage of the writing process. OpenAI’s newest version, ChatGPT-4, provides high-quality, nearly instantaneous feedback to students working at any stage of the writing process, from planning and outlining to drafting and revising.
PapyrusAI puts ChatGPT-4 into what UCI researchers call a “walled garden,” which limits what type of information students can access from the AI, and protects student privacy by preventing data and text from being transferred back to OpenAI programmers.
A version of PapyrusAI is being used this winter in a writing course for UCI engineering students. That iteration was supported with a $400,000 grant from the National Science Foundation, and data is currently being collected and evaluated. The researchers believe this was one of the first NSF grants to support formal educational uses of AI, although it certainly won’t be the last.
“While we’re excited about what we’re seeing at the university level, we really think the true potential is in K-12 schools,” said Mark Warschauer, professor of education and director of the Digital Learning Lab at UCI. “Teachers and administrators know that students need to be able to use AI for their future, but they also know there’s not an easy, appropriate tool for their students to use to learn about it yet.”
This is the first time Warschauer and colleagues have pursued a commercial strategy for tools they developed. Usually, School of Education researchers make this type of resource available to teachers for free online. But OpenAI charges for ChatGPT-4, so if PapyrusAI were broadly adopted, it would require continual funding. After consulting with school administrators, the team realized that most districts prefer an out-of-the-box program rather than open-source code they have to assemble themselves.
Generative AI like ChatGPT could solve significant challenges for K-12 writing teachers: with each teacher having 150 or more students in middle or high school, it’s nearly impossible to provide personalized feedback at every stage of the writing process. OpenAI’s newest version, ChatGPT-4, provides high-quality, nearly instantaneous feedback to students working at any stage of the writing process, from planning and outlining to drafting and revising.
PapyrusAI puts ChatGPT-4 into what UCI researchers call a “walled garden,” which limits what type of information students can access from the AI, and protects student privacy by preventing data and text from being transferred back to OpenAI programmers.
A version of PapyrusAI is being used this winter in a writing course for UCI engineering students. That iteration was supported with a $400,000 grant from the National Science Foundation, and data is currently being collected and evaluated. The researchers believe this was one of the first NSF grants to support formal educational uses of AI, although it certainly won’t be the last.
“While we’re excited about what we’re seeing at the university level, we really think the true potential is in K-12 schools,” said Mark Warschauer, professor of education and director of the Digital Learning Lab at UCI. “Teachers and administrators know that students need to be able to use AI for their future, but they also know there’s not an easy, appropriate tool for their students to use to learn about it yet.”
This is the first time Warschauer and colleagues have pursued a commercial strategy for tools they developed. Usually, School of Education researchers make this type of resource available to teachers for free online. But OpenAI charges for ChatGPT-4, so if PapyrusAI were broadly adopted, it would require continual funding. After consulting with school administrators, the team realized that most districts prefer an out-of-the-box program rather than open-source code they have to assemble themselves.
The Digital Learning Lab
Founded by Professor Mark Warschauer, the Digital Learning Lab promotes educational achievement and equity through developing, designing, researching and evaluating new technologies. With a cutting-edge approach, the lab seeks to understand digital learning and to develop tools and resources that will improve access and opportunities for learners across pre-school to graduate school. Other current projects include: A five-year, $3 million National Science Foundation grant, “Developing Conversational Videos to Support Children’s STEM Learning and Engagement,” is using AI-based conversational agents to create interactive forms of PBS KIDS STEM videos in which children can learn through dialogue with the main characters. The videos will be distributed to millions of children across the country through PBS’s online platforms. Mark Warschauer (UCI Professor) serves as PI with Andres Bustamante (UCI Assistant Professor) and Ying Xu (University of Michigan Professor; UCI alumna) as co-PIs. A three-year, $850,000 National Science Foundation subaward, the “Building A Teacher-AI Collaborative System for Personalized Instruction and Assessment of Comprehension Skills” builds off of DLL’s partnership with PBS KIDS to create interactive STEM videos, but extends into e-reading material and ways to scale interactive learning materials through AI-teacher collaboration. This grant will leverage generative AI to empower teachers to co-create STEM-focused reading resources. The grant is led by University of Michigan Assistant Professor and UCI alumna Ying Xu. Co-PIs include Shiyu Chan (UC Santa Barbara), Dakuo Wang, (Northeastern University), Mark Warschauer (UCI Professor) and Young-Suk Kim (UCI Professor).
A four-year, $2 million National Science Foundation grant, “Deepening Computational Thinking for English Learners by Integrating Community-Based Environmental Literacy” builds on DLL’s work in developing computer science curriculum targeted at the needs of diverse learners. This project partners with Southern California school districts to develop an elementary school curriculum integrating coding with community-based environmental literacy. Mark Warschauer serves as PI with co-PIs Clare Baek (UCI Postdoctoral Researcher), Symone Gyles (UCI Assistant Professor) and Debra Richardson (UCI Professor and Dean Emeritus of the Bren School of Information and Computer Sciences). |
“The sole goal of commercializing it is to create a sustainable product so that it can be scaled and used for a long time by hundreds and thousands of students in the U.S. and around the world,” said Warschauer.
As part of the PoP grant, the team will participate in NSF’s Innovation Corp, a training program to help researchers turn their innovations into marketable products. The researchers will also have access to resources offered by UCI Beall Applied Innovation to support entrepreneurs in refining and ultimately bringing their product to market. Over the next few weeks, the team will work with educators in El Sol Science and Arts Academy of Santa Ana to refine the writing prompts and the type of feedback provided by the AI, so that the language mirrors the language the teachers use in class. Then, in May, teachers and students will begin using the program to support writing lessons – with UCI researchers observing the results. “The important thing is that students are learning to use AI smartly,” said Tate. “They’re not just relying on AI, but practicing a writing skill with their teacher, then practicing it with the AI, and then reflecting on the feedback – reflection is critical to the learning process.” While the newer version of ChatGPT is more accurate than previous iterations, it’s still an imperfect resource, and that provides important teachable moments about checking sources, verifying information and recognizing bias. “There’s a little less bias now, but AI is still trained on an internet that is disproportionately Western, white and male. It’s going to know more about Taylor Swift than Bessie Smith,” Tate said. “That’s the nature of the tool and a fact that students need to learn how to work with.” Eventually, the team hopes to develop resources to help train teachers to use PapyrusAI, create a more powerful dashboard for teachers, and make it easier to integrate the program with schools’ existing Learning Management Systems. Already, they are applying for additional grants to support the ongoing work. “Educational research organizations are recognizing the need for AI-driven tools in education that are developed based on research, and evaluated for efficacy before moving into commercialization,” said graduate student Daniel Ritchie, who is leading the tool development. This team of UCI School of Education researchers is ideally positioned to provide that, combining their decades of expertise in educational technology with strong connections to local educators willing to engage with and try emerging projects like PapyrusAI. “Papyrus AI offers the best of both worlds,” said Warschauer. “Students can interact with ChatGPT to get feedback on their plans and writing in this safer ‘walled garden,’ while also developing their own AI literacy and understanding of the technology’s limitations.” |