AI Governance · A Practitioner's Approach
Cognitive Integrity in Practice.
How Banyan thinks about AI in K–12 instruction, with sample rubric rows and our research grounding.
A 12-page preview document covering Banyan's approach to AI cognitive integrity in K–12 classrooms. Includes excerpts from the Cognitive Integrity Rubric (Grade 8), highlights from the Assessment Integrity Companion, and a research-grounding section that names the specific studies our position draws on, including studies that point in different directions.
Section 1
Our position
The Cognitive Integrity Rubric is grounded in two things: Banyan Global Learning's seventeen years of practitioner experience designing and delivering K–12 instruction since 2009, and the cognitive science literature on learning, productive struggle, and skill development.
Banyan has designed creative units of instruction across ELA, math, science, social studies, and world languages, delivered through our own teaching methodology to students in classrooms around the world. The question at the center of the Rubric, what cognitive work must students still perform for learning to occur, predates the LLM era. Banyan teachers have been asking it about every form of cognitive scaffolding since 2009. What's new is applying the same question to a more capable tool.
Section 2
The classification system
Every task in the Rubric is classified into one of four categories. The classification answers a single practical question: who must do the cognitive work for the learning to occur?
This task IS the learning. The cognitive work is inseparable from the skill development. If AI performs it, the learning does not happen.
Examples: argumentative essay drafting, hypothesis formation, geometric proof
Human work comes first; AI enhances afterward. The human cognitive operation is the anchor; AI adds efficiency, perspective, or feedback after the thinking is done.
Examples: grammar editing on a student-written draft, computational practice after conceptual mastery, citation formatting
A new skill students must explicitly learn about AI itself: capabilities, limitations, biases, evaluation criteria, and AI's role in their discipline.
Examples: evaluating AI-generated historical narratives, prompt engineering for research, identifying bias in AI output
A transitional classification. The task is HUMAN-ESSENTIAL during the period when the underlying skill is being built, and moves to AI-AMPLIFIED once the student has demonstrated competency.
Examples: lab report writing, summarization of complex texts, revision and self-editing
Section 3
Sample rubric rows
Three example classifications, one from each major category. The full Preview includes two more rows; the published Rubric contains 48 task classifications across ELA, math, science, and social studies.
Formulating claims and marshaling evidence IS the thinking. The struggle to construct and defend a position builds critical reasoning. AI-assisted drafting bypasses the cognitive operation that the standard exists to develop.
Students should draft without AI assistance. AI may review final drafts for feedback, but argument construction must remain human-driven. Use process artifacts (Google Docs version history) and in-class anchors to make the classification enforceable.
Mechanical editing is not the learning target. After students produce their own draft, AI can assist with grammar, punctuation, and mechanics checking. Practitioner evidence: students who self-edit first and then use a tool to catch missed errors develop editing skill faster than students who self-edit twice.
Students write and self-edit first. Then use AI tools to catch missed errors and polish conventions. Require a one-paragraph disclosure: which tool was used, what was kept, what was rejected.
Students need to learn to critically evaluate AI-generated content in scientific contexts. AI tools produce plausible-sounding scientific claims that may be inaccurate or unsupported. The skill of evaluating those claims is itself a new science literacy.
Provide students with AI-generated scientific claims and ask them to evaluate validity, sourcing, and alignment with established evidence. Teach them to identify confident-sounding fabrication and to ask for citations.
Two additional rows, including a math HUMAN-ESSENTIAL example and a social studies AI-LITERACY example, are in the full PDF. The complete published Rubric contains 48 task classifications across ELA, math, science, and social studies. K–5 and 9–12 grade-band versions are in development.
Download the full PreviewSection 4
What the research says
The Rubric's pedagogical position is informed by three bodies of research: the established cognitive science on learning and skill development, the recent (2023–2025) research on AI's effects on cognition and learning, and Banyan's seventeen years of practitioner observation.
The established science is more settled than the contemporary AI literature. We name where they disagree.
Established cognitive science
The older literature is more robust than the contemporary AI-and-learning research. Our pedagogical position draws primarily from this foundation.
- Robert Bjork — Desirable difficulties.
Productive struggle in foundational cognitive operations builds more durable skill acquisition than frictionless delivery.
Bjork & Bjork, 2011, in Psychology and the Real World, Worth Publishers.
- John Sweller — Cognitive Load Theory.
The framework for understanding when cognitive offloading helps learning and when it harms learning.
Sweller, 1988, Cognitive Science 12(2).
- Daniel Willingham — Classroom-applied cognitive science.
Willingham, 2009; 2nd ed. 2021, Why Don't Students Like School?, Jossey-Bass.
- Roediger & Karpicke — Retrieval practice and the testing effect.
Roediger & Karpicke, 2006, Psychological Science 17(3).
- Risko & Gilbert — Cognitive offloading.
Pre-LLM but conceptually parallel.
Risko & Gilbert, 2016, Trends in Cognitive Sciences 20(9).
The contested AI-and-learning literature
The 2023–2025 research on AI's effects on cognition and learning is moving fast, and the findings frequently contradict each other.
- "Your Brain on ChatGPT: Accumulation of Cognitive Debt."
EEG study finding reduced neural connectivity in LLM-assisted essay writing. Preprint; peer review in progress.
Kosmyna et al., MIT Media Lab (2025).
- "The Impact of Generative AI on Critical Thinking."
Survey of 319 knowledge workers; higher AI confidence correlated with less critical engagement.
Lee, Sarkar et al., Microsoft Research & Carnegie Mellon (2025).
- "Google Effects on Memory."
Pre-LLM but conceptually parallel.
Sparrow, Liu & Wegner, 2011, Science.
- "Assigning AI: Seven Approaches for Students."
Working paper.
Mollick & Mollick, Wharton (2024).
- AI-tutored physics outperforming active learning.
Preprint reporting AI-tutored physics students outperforming active-learning lectures.
Kestin et al., Harvard (2024).
- Khan Academy / Khanmigo research.
Mixed results across implementations.
2024–2025.
Our position
Our classifications lean toward preserving human cognitive work where the evidence is contested. That's a precautionary stance, not neutrality, and it's worth naming as such. As the research evolves, the Rubric will evolve with it.
Section 5
What this Preview is not
This is a starting position, not the final word. The research will continue to evolve and so will our classifications. The full Companion includes a section on the four most contestable rows. Reasonable specialists who read the same evidence may reach different calls.
Districts that adopt the Rubric are expected to adapt it. The Companion provides the structure for that through the quarterly department alignment check and the audit checklist. A row that's wrong for your context should be changed.
Future versions are published as our understanding of the field changes. The public Rubric is updated on banyangloballearning.com.
What's next
If you'd like to discuss this
Banyan Global Learning is a 17-year-old K–12 services company. We help districts make AI decisions deliberately: what to allow, what to teach, what to buy, and how to keep it consistent across schools. The Cognitive Integrity Rubric and Companion are part of our AI Governance service line.
Download the full Preview
Free. 12 pages. Includes all 5 sample rubric rows, Companion highlights, and full research grounding.
Take the AI Readiness Snapshot
Free. 10 minutes. Personalized 2-page maturity report for your district.
Talk to us about an engagement
Readiness Scan, Audit, Blueprint, or Strategic Advisory Retainer. Email Travis directly.
Cognitive Integrity in Practice · v1 · June 2026