NSF funds U-M initiative leveraging AI to teach students essay writing

The $850k grant will support a multidisciplinary initiative to use large language models as writing assistants for students.
A photo of someone typing on a laptop - a front view of their torso (head is not included in image). In the background is a blurred image of a typed page with red proofreading marks.

The rapid spread of large language models (LLMs) like ChatGPT has fueled broad public anxiety surrounding their use to substitute human input. One arena in which this concern has been particularly salient is education, with students increasingly relying on ChatGPT and the like to complete exams, papers, and other assignments and few mechanisms in place to police this activity.

Prof. Lu Wang
Prof. Lu Wang

But what if LLMs could be used to help students become better writers, instead of writing their essays for them? Researchers at University of Michigan, including Profs. Lu Wang and Xu Wang in Computer Science and Engineering, seek to answer this question by exploring LLMs’ ability to augment students’ argumentative writing skills. Their team has received an $850,000 grant from the National Science Foundation (NSF) to support their project, titled “Argument Graph Supported Multi-Level Approach for Argumentative Writing Assistance.”

“We hope to show that LLMs can be a tool, instead of just a threat,  in the classroom,” said Lu Wang. 

Specifically, the researchers aim to create an LLM-based sandbox for students to practice and receive instant feedback on their writing. 

“Traditionally, it could take days or even weeks for students to receive feedback on their writing, revise, resubmit, and so on,” said Xu Wang. “LLMs such as ChatGPT show the potential of reducing feedback cycles and saving instructors’ time in doing repetitive work so that they can spend more meaningful time with individual students.”

Prof. Xu Wang
Prof. Xu Wang

To test LLMs’ capacity for such assistance, the researchers will build and test a prototype system called ArguAble, which uses AI to help students practice and enhance their argumentative writing. This program will include novel natural language processing (NLP) and machine learning models to ensure adequate understanding and measurement of writing quality. They will also implement a feedback provision and revision suggestion system to help students edit and improve their writing in real time.

“ArguAble provides students with the space to practice their writing with scaffolding activities that build upon each other and help them develop their skills, “ said Lu Wang.

ArguAble will be implemented and evaluated in several classes at U-M, including introductory engineering courses, first-year writing classes, and other classes in which technical or argumentative writing is a focus.

In addition to helping students develop important writing skills that will support their academic success, the researchers hope that their work will shed new light on AI’s ability to serve as a learning tool.

“We aim to show how natural language processing can help us develop efficient and scalable approaches for teaching argumentative writing,” said Lu Wang. “More broadly, we hope our research will demonstrate how AI can help further, instead of hamper, educational goals.”