HAI Seminar: Learning by Creating – A Human-Centered Vision for AI in Education

The talk presents a compelling and well-articulated argument for a paradigm shift in AI for education, moving from efficiency-driven automation to supporting deeper learning through creation. Subramonyam grounds his vision in established learning sciences theories, such as knowledge building (Bereiter & Scardamalia) and knowledge transformation (Bereiter & Scardamalia), and provides historical context with Lucy Sprague Mitchell’s geography lesson and Elwood Cubberley’s factory model. This historical framing effectively illustrates the tension between two educational philosophies and situates current AI developments within a long-standing debate. The speaker’s critique of current EdTech trends, such as AI that automates essay writing and grading, is timely and relevant, and he correctly identifies the lack of affordances in conversational interfaces for meaningful writing support. The proposed solution, Script and Shift, is a thoughtful design that incorporates layers and ‘writer’s friends’ to scaffold the writing process, aiming to protect learner cognition. However, the talk is primarily an expert opinion and design vision, with limited empirical evidence presented. While the speaker mentions that the tools are co-designed with educators, no data on effectiveness or user studies are shown. The Q&A section likely addresses some of these points, but the transcript provided does not include it. The argument is coherent and persuasive, but the lack of empirical validation weakens the overall scientific rigor. The sources cited are primarily historical and theoretical, and the talk does not reference specific peer-reviewed publications or datasets. The adéquation titre/contenu is excellent, as the title accurately captures the central theme. Overall, this is a valuable contribution to the discourse on AI in education, offering a principled alternative to current trends, but it would benefit from more concrete evidence of the proposed tools’ efficacy.

The title accurately reflects the content, which centers on a human-centered vision for AI in education, emphasizing learning by creating.

8/10

The talk is grounded in established learning sciences research (knowledge building, knowledge transformation) and presents original design work from the speaker's lab, with references to collaborators and funding sources. The argument is coherent and well-supported, though it is primarily an expert opinion with illustrative examples rather than a systematic review or empirical study.

• 00:00Introduction
• 01:32Presentation
• 48:54Q&A

• Stanford HAI — Host institution and seminar series.
• National AI Institute for Exceptional Education — Funding source mentioned by the speaker.
• Stanford Accelerator for Learning — Funding source mentioned by the speaker.

• Bereiter, C., & Scardamalia, M. (1987). The Psychology of Written Composition — Foundational work on knowledge telling vs. knowledge transformation, referenced indirectly.
• Bereiter, C., & Scardamalia, M. (2014). Knowledge building and knowledge creation: One concept, two hills to die on — Further elaboration of knowledge building theory.