From Written Response to Dialogue with AI: How Activity Format, Interaction Modality, and Language Impact Student Learning and Engagement

TEXT START: The widespread availability of LLMs is challenging written learning activities, as students can increasingly generate responses without necessarily engaging with the learning content.

The Dissection

This paper is not rescuing learning. It is optimizing the interface through which students interact with an automated cognitive system. It finds a real but limited effect: dialogue increases time, contributions, interest, and self-efficacy; bilingual voice reduces articulation friction. But fewer students complete dialogic activities, and knowledge gains do not reliably improve. The machine becomes tutor, interlocutor, and activity engine while human participation is redesigned around it.

The Core Fallacy

The paper conflates activity with productive participation. More turns, longer sessions, and greater self-efficacy do not restore the wage-production-consumption circuit. Its own results expose the boundary: engagement rises without a reliable learning advantage. Under the Discontinuity Thesis, modality choice is a lag defense. Voice, text, Kannada, English, and pacing can make dependence on AI smoother; they do not reverse P1 cognitive automation or P3 productive-participation collapse.

Hidden Assumptions

• Engagement is treated as valuable even when it does not produce economically necessary human capability.
• Greater access and learner control are assumed to preserve agency rather than create more efficient dependence on AI.
• Short-term knowledge gains are treated as evidence of learning without establishing durable transfer or independent performance.
• English-linked academic and professional aspirations are assumed to remain reliable pathways after AI weakens the credential-to-employment pipeline.
• Educational redesign is treated as capable of outrunning the displacement of cognitive labor.
• The small field study is implicitly treated as evidence for broad institutional and economic conclusions it cannot establish.

Social Function

Classification: partial truth, transition management, and ideological anesthetic.

The study correctly identifies linguistic and interface friction. Its social function is narrower and more revealing: it converts structural displacement into a design problem—choose the right modality, language, information flow, and pace. That helps institutions keep students engaged while leaving ownership of AI capital, labor displacement, and economic necessity untouched. It makes the transition more tolerable without making humans more sovereign.

The Verdict

This is a competent micro-level adaptation study, not a rebuttal to discontinuity. It shows how to make AI-mediated education less alienating and more inclusive; it does not show that education can preserve human economic necessity. The null knowledge result and lower completion rate are the autopsy findings: the format can increase engagement without increasing learning. Engagement without sovereignty is managed occupancy. At best, this research improves the servitor pipeline—people trained to operate around AI, not owners of the AI capital replacing them.