Generative AI introduces a critical rupture in K-12 education: the separation of performance from learning. When students use LLMs to bypass symbol production, they achieve "unproductive success"—flawless outputs without schema construction. This paper proposes a theory of Tertiary Algorithmicity, arguing that educators must design "pedagogical friction" to calibrate desirable cognitive difficulties while ensuring equity.
An 8th grader submits an essay with a clear thesis, smooth transitions, and solid evidence. By every rubric metric, it is proficient. Yet, the student generated all text via an LLM. Does this constitute learning?
K-12 schools are in a rapid, uncoordinated scramble. While over 60% of school leaders integrate tools, only 31% of public schools have written guidelines (NCES, 2024).
Current institutional responses focus on either detection or integration. Both miss the core problem: generative AI bypasses the cognitive resistance essential for durable learning.
Flawless prose creates a false signal. It deceives teachers and grading rubrics alike into certifying conceptual understanding that was never assembled in the student's brain.
Preserved knowledge is Situational, communal, and mnemonic. Language is transient, and thoughts must be formulated in memorable patterns to endure.
Cognitive operations are externalized. Generative models construct syntax and formulate semantic connections that previously required human internal thought.
The core threat of Generative AI. The output looks flawless, but the student bypassed the desirable difficulties. Retrieval strength and schema consolidation are zero.
Optimal level. Scaffolds cognitive challenge, preserving schema construction while leveraging AI as a critical dialogue partner rather than an outsourcing agent.
Preserves internal cognitive struggle. Requires students to struggle with concepts (e.g., initial drafts, thinking journals) before consulting AI tools.
Friction Calibration: Minimize mechanical language barriers (exclusionary friction) by allowing AI translation or dictionary extensions. However, preserve the core conceptual friction: requires organizing arguments and syntheses internally.
"How can we measure learning in an environment where output is no longer a reliable proxy?"