Scholarlink

<p>Artificial Intelligence (AI) enables numerous educational tasks such as grading and providing feedback. The recent advancement in Large Language Models (LLMs) served as a catalyst to the adoption of such tools. However, before relying extensively on AI, we should carefully consider its long-term implications and ensure that its use aligns with ethical and educational principles. At present, there is no established framework to guide educators, administrators, and AI developers in assessing the alignment of AI in education with ethical and pedagogical standards. In this article, we bridge this gap by creating an instrument that can be used to evaluate the alignment of AI tools. We did so by, firstly, conducting a systematic search in databases WoS, Scopus, ACM, and IEEE Xplore attempting to find empirical literature and guidelines on AI alignment. The search process identified 409 relevant records, after applying inclusion criteria, 21 articles were chosen. 25 articles were additionally found through snowballing and search in Google Scholar and Google, raising the total to 46 articles. Secondly, utilizing the guidance of 4 comprehensive frameworks and guidelines, we extracted the criteria and concepts related to education from the identified literature to construct an evaluation instrument. Thirdly, we applied the instrument in a case study to evaluate educational recommendations by AI and then conducted an Exploratory Factor Analysis on the results to understand the relation between the criteria. We identified 4 categories to group 11 criteria. Category <i>Clarity</i> which include Accuracy, Presentability, and Justification. Category <i>Feasibility</i> which include Usefulness and Coherence. Category <i>Pedagogical Value</i> which include Implementability and Engagement with Higher-Order Cognitive Skills. Category <i>Learner-Focused</i> which include Learner-Centeredness, Diversity, Equity, and Inclusion. There are two additional criteria that were not identified with any category as their scales are not nominal—Privacy and Ethicality and Alignment with Learning Theories. This instrument can be used by different stakeholders in education to make sure AI usage meets the complex requirements of human needs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Instrument for Human-Based Evaluation of GenAI Educational Alignment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月8日

查看全文
  • Hesham Ahmed,
  • Sonsoles López-Pernas,
  • Leonie V. D. E. Vogelsmeier,
  • Markku Tukiainen,
  • Mohammed Saqr

摘要

Artificial Intelligence (AI) enables numerous educational tasks such as grading and providing feedback. The recent advancement in Large Language Models (LLMs) served as a catalyst to the adoption of such tools. However, before relying extensively on AI, we should carefully consider its long-term implications and ensure that its use aligns with ethical and educational principles. At present, there is no established framework to guide educators, administrators, and AI developers in assessing the alignment of AI in education with ethical and pedagogical standards. In this article, we bridge this gap by creating an instrument that can be used to evaluate the alignment of AI tools. We did so by, firstly, conducting a systematic search in databases WoS, Scopus, ACM, and IEEE Xplore attempting to find empirical literature and guidelines on AI alignment. The search process identified 409 relevant records, after applying inclusion criteria, 21 articles were chosen. 25 articles were additionally found through snowballing and search in Google Scholar and Google, raising the total to 46 articles. Secondly, utilizing the guidance of 4 comprehensive frameworks and guidelines, we extracted the criteria and concepts related to education from the identified literature to construct an evaluation instrument. Thirdly, we applied the instrument in a case study to evaluate educational recommendations by AI and then conducted an Exploratory Factor Analysis on the results to understand the relation between the criteria. We identified 4 categories to group 11 criteria. Category Clarity which include Accuracy, Presentability, and Justification. Category Feasibility which include Usefulness and Coherence. Category Pedagogical Value which include Implementability and Engagement with Higher-Order Cognitive Skills. Category Learner-Focused which include Learner-Centeredness, Diversity, Equity, and Inclusion. There are two additional criteria that were not identified with any category as their scales are not nominal—Privacy and Ethicality and Alignment with Learning Theories. This instrument can be used by different stakeholders in education to make sure AI usage meets the complex requirements of human needs.