Experimental Evidence on the Learning Impact of Generative AI
创建于 更新于
摘要
We report causal evidence from a two-session randomized experiment assessing how access to off-the-shelf generative AI affects undergraduate learning. In Session One, AI access raises immediate knowledge-test scores by about 0.27 standard deviations, with roughly 76% of these gains persisting to Session Two a week later. Essays show modest quality gains during Session One but deliver stronger improvements in writing style and prompt relevance in Session Two when students write unaided, consistent with durable skill gains. Gains are larger for augmentation users (AI used to explain concepts) and fade for automation users (AI drafted text). We identify two mechanisms: AI shifts study time away from drafting toward reading and information searching, and it increases learning enjoyment. [page::2][page::14][page::16][page::17][page::18][page::19]
