Scholarlink

This study explores the application of Google NLM, an AI model uniquely trained on lecture audio, in a specialized engineering course on Production Management. The model was tested under real exam conditions and compared to the performance of 14 students from the 2022–2023 academic year. Results show that the AI consistently passed the exam, achieving an average score of 23.5/30, comparable to the student average of 23/30. While demonstrating strong consistency and factual recall, the AI struggled with numerical reasoning and applied problem-solving, particularly in inventory management and statistical decision-making. Key contributions include the first application of an audio-trained AI in engineering education and an analysis of AI performance in a highly technical domain. While not exceeding top human scores, the AI’s stability suggests potential as a benchmarking tool for exam design and student assessment. Future research should explore multilingual training, hybrid audio-text learning, and domain-specific fine-tuning to enhance AI’s role in academic evaluation.

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

AI vs. Human Performance in University Assessments: A Case Study in Production Management

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Italo Cesidio Fantozzi,
  • Luca Martuscelli,
  • Massimiliano M. Schiraldi

摘要

This study explores the application of Google NLM, an AI model uniquely trained on lecture audio, in a specialized engineering course on Production Management. The model was tested under real exam conditions and compared to the performance of 14 students from the 2022–2023 academic year. Results show that the AI consistently passed the exam, achieving an average score of 23.5/30, comparable to the student average of 23/30. While demonstrating strong consistency and factual recall, the AI struggled with numerical reasoning and applied problem-solving, particularly in inventory management and statistical decision-making. Key contributions include the first application of an audio-trained AI in engineering education and an analysis of AI performance in a highly technical domain. While not exceeding top human scores, the AI’s stability suggests potential as a benchmarking tool for exam design and student assessment. Future research should explore multilingual training, hybrid audio-text learning, and domain-specific fine-tuning to enhance AI’s role in academic evaluation.