# Practical and Ethical Challenges of Large Language Models in Education: A Systematic Literature Review

**Authors:** Lixiang Yan, Lele Sha, Linxuan Zhao, Yuheng Li, Roberto Martinez-Maldonado, Guanliang Chen, Xinyu Li, Yueqiao Jin, Dragan Gašević  
**arXiv:** [2303.13379](https://arxiv.org/abs/2303.13379)  
**Date:** 2023-03-17  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2303.13379v1/

## Key points

- Systematic literature review conducted to identify the current state of research on using Large Language Models (LLMs) in education
- 118 peer-reviewed papers published since 2017 were analyzed
- Focus on practical and ethical challenges associated with LLMs-based innovations in education
- Concerns identified regarding technological readiness, model performance, replicability, system transparency, privacy, equality and beneficence of these innovations
- Three recommendations proposed: updating existing innovations with state-of-the-art models such as GPT-3; embracing open sourcing models/systems initiative; and adopting a human-centered approach throughout the developmental process
- Contextual information related to educational tasks, stakeholders involved, LLMs used and machine learning tasks performed by each included paper was extracted
- Seven data extraction items were developed to assess technology readiness and ethicality based on Australian Department of Defense's Technology Readiness Levels (TRL)
- Findings provide valuable insights for researchers and practitioners exploring viable directions for developing practical and ethical LLMs-based innovations that benefit students, teachers and institutions while addressing potential challenges.

## Summary

This article presents the findings of a systematic literature review conducted to identify the current state of research on using Large Language Models (LLMs) to automate and support educational tasks. The review analyzed 118 peer-reviewed papers published since 2017, focusing on the practical and ethical challenges associated with LLMs-based innovations in education. The study identified concerns regarding technological readiness, model performance, replicability, system transparency, privacy, equality and beneficence of these innovations. To address these issues and support future research in this area three recommendations were proposed: updating existing innovations with state-of-the-art models such as GPT-3; embracing open sourcing models/systems initiative; and adopting a human-centered approach throughout the developmental process. The study also extracted contextual information related to educational tasks, stakeholders involved, LLMs used and machine learning tasks performed by each included paper. Additionally seven data extraction items were developed to assess technology readiness and ethicality based on Australian Department of Defense's Technology Readiness Levels (TRL). The findings provide valuable insights for researchers and practitioners exploring viable directions for developing practical and ethical LLMs-based innovations that benefit students, teachers and institutions while addressing potential challenges.