Generative AI: Implications and Applications for Education

AI-Powered Student Feedback and Assessment

Context: Masters and Doctoral students in Learning Design and Leadership, higher education students, trainee teachers, and young children. This includes online courses, storytelling, and academic writing contexts.

Implementation: Various GPT models (GPT-3, ChatGPT) are used to provide feedback and assess student work. This includes platforms like CGMap integrated with GPT models. The AI reviews are used to provide feedback on student's extended multimodal texts, simulate student discourse in dialogue with trainee teachers, and assess student work.

Outcomes: AI reviews offer more extensive feedback, are considered fast, straightforward, immediate, practical, satisfying, and instant. They can constructively support literacy development, and offer more detailed feedback that fluently and coherently summarizes students’ performance than human instructors. AI has a significant impact in teacher education.

Challenges: AI reviews can be too general, lacking the human touch. They may miss specific areas for improvement and may not be specific enough. AI reviews can have higher variability in results. Challenges include determining the originality of student work, as well as the unreliability of GPT detectors.

Epistemological

C-LLMs are deeply harmful to a social understanding of knowledge and learning in that the machine buries its sources, cannot have a notion of empirical truth, cannot have a conception of a theoretical frame or disciplinary practice, cannot have explicit ethical frames, and cannot engage in critical dialogue.

Proposed Solutions: Recalibration of C-LLMs to include an Epistemic Frame, an Empirical Frame, and an Ontological Frame.

Accuracy/Reliability

GPT detectors are unreliable. The machine has no way of distinguishing fact from fake in its sources. The AI review "missed some smaller points" and "was not specific enough in terms of items that needed to be revised."

Proposed Solutions: Using the generative AI to offer students feedback on the basis of a theory of knowledge applicable to their learning. Require the learners to bring verifiable facts to the machine. Bring the theoretical frames of disciplines to the machine.

Quality of Feedback

The AI reviews were too general, lacking the human touch that the peer reviewers could provide, leading to difficulties in identifying specific areas for improvement. The AI review "missed some smaller points" and "was not specific enough in terms of items that needed to be revised."

Proposed Solutions: Prompt engineering. Recalibration of C-LLMs to include an Epistemic Frame, an Empirical Frame, and an Ontological Frame.

Bias and Ethics

C-LLMs depend on massive textual corpora, and the reality of human legacy text is that the sources are rife with racism, sexism, and homophobia, along with other ideologies and social orientations that are today unacceptable in mainstream public life. C-LLMs require extensive filtering.

Proposed Solutions: Human programmers create the filters to over-ride the "truth" of the source texts.

Project Contact: Dr. Jianhua Yang

LLM Model Version: gemini-2.0-flash-lite