A Trial of an AI-Powered Automated Writing Evaluation Platform: Affordances and Limitations

S. K. Lo, S. Y. Ng, Y. M. Wong

Research output: Unpublished conference presentation (presented paper, abstract, poster)Conference presentation (not published in journal/proceeding/book)Teaching and learningpeer-review

Abstract

This presentation reports the findings from a trial of the AI Review Assessment System (AIReAS), an automated writing evaluation platform powered by GPT-4 and developed by the Hong Kong Polytechnic University. AIReAS enables teachers to input assignment details and assessment rubrics, generating a set of rubric-based feedback and a predicted grade for each student submission. The trial was conducted on a research report assignment in a general education course on the history of Hong Kong. Students received feedback from both teachers and the platform and were interviewed for comments on both types of feedback.

Thematic analysis of the focus group interview transcripts showed that students appreciated the AI feedback for its wide coverage, as well as its high efficiency and accessibility. However, they found the feedback too general to be helpful for improving their reports. In contrast, teacher feedback was perceived as more detailed and useful as it included specific examples of issues. Students showed a clear preference for teacher feedback but expressed interest in using both types to maximize their benefits. The results confirmed the value of AI feedback and its potential limitations as identified in research on automated writing evaluation.
Original languageEnglish
Publication statusPublished - 2024
Event4th International Conference on English Across the Curriculum - The Hong Kong Polytechnic University , Hong Kong, Hong Kong
Duration: 2 Dec 20244 Dec 2024
https://www.polyu.edu.hk/en/eacconference/

Conference

Conference4th International Conference on English Across the Curriculum
Country/TerritoryHong Kong
CityHong Kong
Period2/12/244/12/24
Internet address

Fingerprint

Dive into the research topics of 'A Trial of an AI-Powered Automated Writing Evaluation Platform: Affordances and Limitations'. Together they form a unique fingerprint.

Cite this