Architecture of AI in the Classroom

System key

• Blue: cognitive
• Red: instructional
• Green: ethical
• Purple: institutional

What AI changes and what it doesn't

• AI changes
• First-draft generation
• Information retrieval
• Formatting
• Summarizing
• Basic analysis
• AI doesn't change
• Judgment
• Synthesis
• Original argument
• Ethical reasoning
• Creative risk
• Domain expertise
• Meaning-making
• The question is not whether students use AI. It's what learning looks like when they do.

The cognitive offloading spectrum

• Diagram: a spectrum from full cognitive engagement (student does the thinking) to full offloading (AI does the thinking), with three points along it: productive assistance (AI as scaffold), dependent use (AI as replacement), cognitive bypass (learning doesn't happen).
• Offloading is not cheating. Bypassing is. The line is cognitive engagement.

Academic integrity systems in the AI era

• Old model: detect, then punish
• New model: design, then prevent, then assess authentically
• What integrity-by-design looks like
• Process documentation
• In-class components
• Oral defense
• Iterative drafts
• Personalized prompts
• Detection alone rarely works. Design is the more sustainable strategy.

AI-resistant skill architecture

• Diagram: a pyramid with what AI cannot do at the top, what AI struggles with in the middle, and what AI can do at the base.
• AI cannot do: lived experience, ethical responsibility, creative originality, relational knowing
• AI struggles with: synthesis, nuanced argument, domain judgment
• AI can do: information retrieval, summarization, formatting, basic writing
• Teach what AI cannot replace. Assess what AI cannot produce.

Bloom's taxonomy revised for AI

• Diagram: a table with columns Bloom's level, original, what AI handles, human irreplaceable, instructional implications.
• Create: original: create. What AI handles: co-create, draft. Human irreplaceable: originality, purpose, values, context. Instructional implications: ask for original contributions, real-world creation.
• Evaluate: original: evaluate. What AI handles: analyze criteria. Human irreplaceable: ethical judgment, contextual evaluation. Instructional implications: use judgment-rich assessments.
• Analyze: original: analyze. What AI handles: identify patterns, compare. Human irreplaceable: deep analysis, meaning across domains. Instructional implications: design complex, open problems.
• Apply: original: apply. What AI handles: apply rules, examples. Human irreplaceable: adaptive application in new contexts. Instructional implications: assess transfer and adaptation.
• Understand: original: understand. What AI handles: explain, summarize. Human irreplaceable: integration, explanation in own words. Instructional implications: use exploratory assessments.
• Remember: original: remember. What AI handles: recall facts. Human irreplaceable: not human-unique. Instructional implications: use as baseline, not endpoint.
• AI compresses the bottom of the taxonomy. Faculty must design for the top.

Student AI literacy systems

• Diagram: AI literacy at the center, connected to six parts: understanding how AI works, recognizing AI limitations and bias, prompt crafting as analytical skill, evaluating AI output critically, ethical use frameworks, disciplinary norms around AI.
• Using AI is not AI literacy. Thinking critically about AI is.

Faculty AI anxiety systems

Two typical responses.

• Threat response
• Detect
• Ban
• Punish
• Surveil
• Adaptive response
• Learn
• Redesign
• Integrate
• Upskill
• Drivers of threat response
• Fear of irrelevance
• Loss of control
• Unclear institutional guidance
• Time constraints
• Faculty anxiety about AI is reasonable. Threat responses make it worse.

Assessment redesign for an AI world

• Before
• Take-home essays
• Low-stakes recall tests
• Decontextualized prompts
• Single-shot submissions
• After
• Process portfolios
• In-class writing
• Oral components
• Authentic tasks
• Iterative peer review
• Discipline-specific application
• If AI can do your assignment, the assignment was already doing too little.

AI bias and limitation systems

AI failure modes that matter.

• Hallucination
• Training data bias
• Cultural encoding
• Temporal limitations
• Disciplinary unevenness
• Overconfidence in output
• Student implications
• Verify and fact-check
• Look for bias and blind spots
• Understand context and limits
• Cross-validate with sources
• AI is not neutral. It encodes the data it was trained on.

The academic integrity conversation

A response decision tree for faculty.

• Diagram: What happened? branches into intentional bypass, misuse from ignorance, and productive use gone wrong; these lead to response pathways: punitive, educational, redesign.
• Punitive: penalty, zero, report
• Educational: teach, coach, conference
• Redesign: revise task, clarify norms, improve design
• Many students don't know where the line is. Draw it explicitly.

AI and disciplinary identity

Different fields. Different emphases.

• Diagram: a table with columns discipline, trend, consensus, tensions remain.
• Sciences: trend: tool for analysis & productivity. Consensus: data literacy, accuracy. Tensions remain: lab reports, authorship, reliability.
• Humanities: trend: critical use of texts. Consensus: interpretation is human. Tensions remain: writing norms, AI in argumentation.
• Professional Programs: trend: real-world integration. Consensus: authentic application. Tensions remain: competency assessment, credentialing.
• Arts & Design: trend: creative co-pilot. Consensus: human vision & intent. Tensions remain: originality, style, authorship.
• There is no universal answer. There is disciplinary judgment.

The cognitive development question

Does the struggle still happen?

• Core concern: If AI does the struggling, does the learning happen?
• Protect struggle when...
• Skill development
• Expertise formation
• Deep understanding
• Transfer is the goal
• Remove or reduce struggle when...
• Access & equity barriers
• Routine/low-value tasks
• Scaffolding early learning
• Cognitive overload
• The goal is not effortful learning. The goal is learning that transfers.

The faculty response spectrum

A range of faculty approaches.

• Diagram: a spectrum from full ban to AI-first design.
• Full ban: AI prohibited, detect and punish
• Ambiguity: unclear policy, inconsistent enforcement
• Transparent integration: explicit use norms, literacy instruction, redesigned assessment
• AI-first design: AI as primary tool, human judgment as evaluator
• Illustrative distribution (not measured data): one curve for each approach along the spectrum.
• Policy ambiguity is itself a policy. It favors students who already know the rules.

The AI classroom ecosystem

• Diagram: learning at the center, connected by arrows to six systems.
• Instructor design: clarity, relevance, warmth, responsiveness, high expectations
• Student agency: motivation, self-regulation, ownership, help-seeking, metacognition
• Institutional policy: policy clarity, support, resources, training, academic freedom
• Disciplinary norms: ways of knowing, evidence, methods, standards of practice
• Assessment structure: aligns with learning goals, valid, authentic, ongoing feedback
• Environment: physical space, tools, time demands, outside stress
• AI is not the problem or the solution. It's a force in a system. Design the system.

What remains essential

Teach and assess these, always.

• Critical evaluation of information
• Original synthesis of ideas
• Ethical reasoning and judgment
• Disciplinary judgment and expertise
• Written articulation of complex thought
• Revision as thinking
• Intellectual risk-taking
• Authentic voice and perspective
• These are not soft skills. They are the skills AI makes more important, not less.

The AI learning equation

When systems align, learning transfers.

• Human judgment + AI capability + critical literacy + authentic assessment = learning that transfers
• When the system aligns, AI becomes a tool. When it doesn't, AI becomes a replacement.