Generative AI in University Teaching: Tutor or Shortcut?
Year: 2026
Type: preprint
Abstract: Generative AI can help university students learn faster, but it can also make them learn less. The outcome depends on how lecturers and institutions design its use, not on whether students have access to it. This article reviews randomised controlled trials (RCTs) published between 2025 and 2026, focusing on higher education and drawing on school-level trials where they explain the mechanism. A purpose-built AI tutor helped Harvard physics students learn more than twice as much as an active-learning class, in less time. By contrast, giving 2,379 University of Maryland students a general course assistant lowered final grades by 0.37 standard deviations, with first-generation students losing the most; other university trials found reduced knowledge retention or no effect. The difference lies in design: tools that prompt thinking and are built into teaching help, while tools that hand out answers replace the effort that learning requires. The article derives five design principles for lecturers and recommends how Vietnamese universities can implement Circular 49/2026/TT-BGDĐT, which requires every institution to adopt rules on AI use in teaching and research.
Source: EdArXiv (OSF Preprints)
Author: Pham Viet Hung
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7225594155