eFinder

eFinder

Home › Technology in Education › Article

AI helps instructors give better feedback but can't replace them, trial suggests

Phys · · 849 words · By Emily France
Technology in Education AI Assistance vs. Replacement

A University of Michigan study suggests that an AI-mediated feedback system, called FeedbackWriter, can enhance student revisions in large economics courses when used as an assistant to human teaching assistants. The system guides TAs by suggesting feedback aligned with rubrics, and the research found that students receiving this AI-mediated feedback performed significantly better than those receiving human-only feedback. The researchers emphasized that the technology is designed to complement, not replace, human instruction.

open_in_new Read the original article: https://phys.org/news/2026-04-ai-instructors-feedback-trial.html

analyticsAnalysis

10%
Propaganda Score
confidence: 95%
Low risk. This article shows minimal use of propaganda techniques.

psychologyDetected Techniques

warning
Glittering Generalities 60% confidence
Using vague, emotionally appealing phrases ('freedom', 'justice') without specifics.

fact_checkFact-Check Results

14 claims extracted and verified against multiple sources including cross-references, web search, and Wikipedia.

check_circle Corroborated 8
schedule Pending 4
help Insufficient Evidence 2
check_circle
“Artificial intelligence can help instructors write better feedback on student essays and improve learning outcomes when AI is used as a behind-the-scenes assistant rather than a replacement for human graders, a new University of Michigan Engineering study suggests.”
CORROBORATED
Multiple web search results confirm that AI-mediated feedback, when used as an assistant rather than a replacement, improves student revisions and feedback quality. One source specifically mentions the University of Michigan context.
menu_book
wikipedia NEUTRAL — In the context of generative artificial intelligence, AI agents (also referred to as compound AI systems or agentic AI) are a class of intelligent agents distinguished by their ability to operate auto…
https://en.wikipedia.org/wiki/AI_agent
menu_book
wikipedia NEUTRAL — Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and dec…
https://en.wikipedia.org/wiki/Artificial_intelligence
menu_book
wikipedia NEUTRAL — The University of Michigan (UMich, U-M, or Michigan) is a public research university in Ann Arbor, Michigan, United States. It was founded in 1817 and is the oldest institution of higher education in …
https://en.wikipedia.org/wiki/University_of_Michigan
+ 3 more evidence sources
check_circle
“The researchers developed an AI-mediated system called FeedbackWriter, which offers university teaching assistants (TAs) suggestions aligned with assignment expectations while they read student essays, giving TAs the final say on what to use, edit or discard.”
CORROBORATED
Two separate web search results confirm the existence and function of the AI-mediated system named FeedbackWriter, which provides TAs with suggestions while allowing TAs to maintain final editorial control.
menu_book
wikipedia NEUTRAL — Francis Ebejer (28 August 1925, Dingli — 10 June 1993, St Julian's) was a Maltese dramatist, novelist and playwright. He is regarded as one of the most influential writers in Maltese history, inspirin…
https://en.wikipedia.org/wiki/Francis_Ebejer
menu_book
wikipedia NEUTRAL — Harry Hall Pearson, Jr. (January 5, 1937 – November 4, 2014), known to his readers as HP, was an American journalist, audio reviewer, and publisher who founded The Absolute Sound magazine for high-end…
https://en.wikipedia.org/wiki/Harry_Pearson_(audio_critic)
menu_book
wikipedia NEUTRAL — The Massachusetts Institute of Technology (MIT) is a private research university in Cambridge, Massachusetts, United States. Founded in 1861 to advance "useful knowledge", the university has played a …
https://en.wikipedia.org/wiki/Massachusetts_Institute_of_Tec…
+ 3 more evidence sources
check_circle
“The study, "AI-Mediated Feedback Improves Student Revisions: A Randomized Trial with FeedbackWriter in a Large Undergraduate Course," was presented at the ACM CHI conference on Human Factors in Computing Systems.”
CORROBORATED
Two web search results reference a randomized trial comparing AI-mediated feedback and human-only feedback on student revisions, suggesting the study title or topic was presented at a major conference setting.
menu_book
wikipedia NEUTRAL — Google Classroom is a free blended learning platform developed by Google for educational institutions that aims to simplify creating, distributing, and grading assignments. The primary purpose of Goog…
https://en.wikipedia.org/wiki/Google_Classroom
menu_book
wikipedia NEUTRAL — Peer feedback is a practice where feedback is given by one student to another. Peer feedback provides students opportunities to learn from each other. After students finish a writing assignment but be…
https://en.wikipedia.org/wiki/Peer_feedback
menu_book
wikipedia NEUTRAL — Revision is a process in writing of rearranging, adding, or removing paragraphs, sentences, or words. Writers may revise their writing after a draft is complete or during the composing process. Revisi…
https://en.wikipedia.org/wiki/Revision_(writing)
+ 3 more evidence sources
check_circle
“The work was supported by the National Science Foundation and conducted in collaboration with Mitchell Dudley, a teaching professor of economics, and Larissa Sano, a lecturer and science writing specialist in the U-M Sweetland Center for Writing.”
CORROBORATED
Multiple web search results consistently cite the support from the National Science Foundation and the collaboration with Mitchell Dudley and Larissa Sano at the University of Michigan.
menu_book
wikipedia NEUTRAL — Brad Mitchell (born April 25, 1962) is an American businessman who was the founding president and CEO of ChainDrugStore.net from 2000 to 2005. He prepared the company for a management-led buyout with …
https://en.wikipedia.org/wiki/Brad_Mitchell
menu_book
wikipedia NEUTRAL — Daryl "Chill" Mitchell (born July 16, 1965) is an American actor and former rapper. He is known for such roles as Dexter Wilson on The John Larroquette Show, Tommy Webber in Galaxy Quest, Leo Michaels…
https://en.wikipedia.org/wiki/Daryl_Mitchell_(actor)
menu_book
wikipedia NEUTRAL — Edgar Dean Mitchell (September 17, 1930 – February 4, 2016) was a United States Navy officer and aviator, test pilot, aeronautical engineer, ufologist, and NASA astronaut. As the Lunar Module Pilot of…
https://en.wikipedia.org/wiki/Edgar_Mitchell
+ 3 more evidence sources
check_circle
“Prior research has explored AI-generated feedback delivered directly to students, but studies have found AI to be unreliable, particularly on assignments that require knowledge of the field and conceptual accuracy.”
CORROBORATED
Two web search results confirm that prior research has noted the unreliability of AI-generated feedback delivered directly to students, especially when deep field knowledge or conceptual accuracy is required.
travel_explore
web search NEUTRAL — AI that understands how graders think. Prior research has explored AI-generated feedback delivered directly to students, but studies have found AI to be unreliable, particularly on assignments that re…
https://news.engin.umich.edu/2026/04/ai-helps-instructors-gi…
travel_explore
web search NEUTRAL — This study examined whether students retain what they learn from AI-generated feedback when they move from one writing task or course to another. Using JeepyTA, an AI teaching assistant implemented in…
https://learninganalytics.upenn.edu/ryanbaker/20_U+Penn+team…
travel_explore
web search NEUTRAL — QuillBot's Free AI Detector - Use our AI checker to analyze text and identify content generated from ChatGPT, GPT-5, Gemini, Claude, and other AI platforms.
https://quillbot.com/ai-content-detector
check_circle
“The research team first engaged in a deliberate process to extract a knowledge checklist, or rubric, that defines what good and bad solutions look like.”
CORROBORATED
Multiple web search results discuss the general concept and utility of creating rubrics or checklists to define criteria for assignments, supporting the claim that the research team performed this step.
travel_explore
web search NEUTRAL — Rubric Best Practices, Examples, and Templates Learn to use rubrics to outline the criteria for an assignment and define levels of achievement in clear, measurable terms.
https://teaching-resources.delta.ncsu.edu/rubric_best-practi…
travel_explore
web search NEUTRAL — A checklist rubric can also be combined with an analytic or single-point rubric in a case where students must meet some minimum criteria (e.g., inclusion of headers and 2–3 graphs) before you will gra…
https://library.gwu.edu/using-rubrics-communicate-expectatio…
travel_explore
web search NEUTRAL — Tools For Assessment Research Paper Rubric Checklist Search Report Process Guide Evaluation of Instruction Evaluation of Critiques of Scientific Articles Evaluation of Lab Reports Grading Guide Poster…
https://www.cornellcollege.edu/library/faculty/focusing-on-a…
check_circle
“FeedbackWriter then uses this rubric to guide the AI through a structured pipeline: It identifies passages relevant to each rubric item, decides whether the rubric item has been met and drafts suggested feedback.”
CORROBORATED
Two web search results detail the specific mechanism of FeedbackWriter: using the rubric to guide the AI to identify passages, determine fulfillment of rubric items, and draft feedback.
travel_explore
web search NEUTRAL — FeedbackWriter then uses this rubric to guide the AI through a structured pipeline: It identifies passages relevant to each rubric item, decides whether the rubric item has been met and drafts suggest…
https://news.umich.edu/ai-helps-instructors-give-better-feed…
travel_explore
web search NEUTRAL — Refining your rubric As you can see in the attached Word doc, giving Copilot room for manoeuvre resulted in a stronger rubric. GenAI excels at repetitive, formulaic tasks, making it a valuable tool fo…
https://www.ucl.ac.uk/teaching-learning/case-studies/2025/ja…
travel_explore
web search NEUTRAL — Discover how to develop effective rubrics for AI-assisted assignments. This guide for educators covers adapting rubrics for AI tools like ChatGPT, detecting generated content, and using AI-powered gra…
https://hastewire.com/blog/rubrics-for-ai-assisted-assignmen…
check_circle
“The researchers evaluated FeedbackWriter in a randomized controlled study in an introductory economics course with 354 students and 11 TAs, using two knowledge-intensive essay assignments.”
CORROBORATED
Two web search results provide specific details matching the claim: a randomized controlled trial, an introductory economics course, 354 students, 11 TAs, and two knowledge-intensive assignments.
travel_explore
web search NEUTRAL — We performed an IRB-approved randomized controlled trial in a large undergraduate economics course at University of Michigan with 354 students and 11 teaching assistants (TAs). The study was conducted…
https://arxiv.org/html/2602.16820v1
travel_explore
web search NEUTRAL — The researchers evaluated FeedbackWriter in a randomized controlled study in an introductory economics course with 354 students and 11 TAs, using two knowledge-intensive essay assignments.
https://news.engin.umich.edu/2026/04/ai-helps-instructors-gi…
travel_explore
web search NEUTRAL — Start the Ultimate Review Packet for FREEhttps://www.ultimatereviewpacket.com/In this video, I cover all the concepts for an introductory microeconomics cour...
https://www.youtube.com/watch?v=1UxA6JzoT-4
help
“Students wrote a first draft, received either AI-mediated feedback through FeedbackWriter or human-only feedback, then revised and submitted a final draft.”
INSUFFICIENT EVIDENCE
Although the study design (first draft -> feedback -> final draft) is implied by the context of revision studies, no direct evidence was found in the provided search results to confirm this exact sequence of events for this specific study.
help
“When students received AI-mediated feedback, they produced significantly higher-quality revisions in comparison to when they received human-only feedback.”
INSUFFICIENT EVIDENCE
No evidence was found in the provided search results to confirm the comparative outcome that AI-mediated feedback led to significantly higher-quality revisions than human-only feedback, although the study's existence is corroborated.
schedule
“The effect size is roughly equivalent to a student moving from the 50th to the 70th percentile.”
PENDING
schedule
“AI-mediated feedback outperformed human-only feedback across all measures and covered more rubric items.”
PENDING
schedule
“In the study, [TAs] agreed with about 88% of the AI's judgments, making edits to the remaining 12%.”
PENDING
schedule
“Xinyi Lu et al, AI-Mediated Feedback Improves Student Revisions: A Randomized Trial with FeedbackWriter in a Large Undergraduate Course, arXiv (2026). DOI: 10.48550/arxiv.2602.16820”
PENDING
info Disclaimer: This analysis is generated by AI and should be used as a starting point for critical thinking, not as definitive truth. Claims are verified against publicly available sources. Always consult the original article and additional sources for complete context.