Cultivate AI Fluency
The field spent two years asking "how do we stop students from using AI?" The better question is "what does a student who uses AI well actually look like, and how do we teach that?" Those are design questions, not policy questions.
What AI proficiency actually means in practice
AI proficiency is not knowing how to write a prompt. It's knowing what the AI doesn't know that you do — and using that gap productively. A student who is AI-proficient in business research can take the macro trend analysis the LLM produces and combine it with local market intelligence they gathered themselves, producing something neither could produce alone. That's the professional skill. That's what employers are paying for.
The skill looks different by discipline. In design it's knowing when the AI-generated concept is generic and why your judgment overrides it. In nursing it's knowing when the AI-suggested intervention doesn't fit this patient's specific presentation. In research it's knowing which sources the AI missed and why they matter.
What AI proficiency looks like by domain
| Domain | What the AI brings | What the student must bring |
|---|---|---|
| Writing | Fluency, structure, transitions, revision suggestions | Argument, position, specific personal or intellectual experience |
| Research | Source synthesis, framework application, pattern recognition across literature | Research question, theoretical choice, interpretation of what the evidence means |
| Business | Industry trends, framework templates, financial benchmarks | Local or specific data gathered firsthand, judgment on which framework fits this case |
| Design | Concept generation, iteration speed, format suggestions | Design rationale, audience-specific decisions, critique of AI output against brief requirements |
| Science | Background synthesis, hypothesis generation, statistical interpretation templates | Experimental data the student collected, anomaly identification, domain judgment about validity |
| Clinical | General clinical guidelines, standard intervention templates, documentation formats | Patient-specific or scenario-specific reasoning — why this intervention for this presentation |
| Education | Learning objective templates, activity suggestions, general instructional frameworks | Specific learner context, rationale for why this sequence fits this student group, adaptation judgment |
How scaffolding builds the skill progressively
A single scaffolded assignment doesn't build AI proficiency — a semester of scaffolded assignments does. The pattern that works is progressive disclosure of AI access combined with increasing demands on student-specific contribution:
- Early in the course: AI permitted for research and synthesis. Student contribution is framing and judgment. Disclosure required.
- Mid-course: AI permitted for drafting. Student required to document where they diverged from AI output and why.
- Late in the course: AI available as a tool, but the primary deliverable requires knowledge or data only the student could have gathered.
The disclosure requirement at every stage isn't punitive — it's metacognitive. Students who articulate what the AI did and didn't contribute are developing the judgment that makes them useful to an employer.
Work with an Existing Assignment
Most assignment redesign conversations start in the wrong place — asking "what AI risk does this assignment have?" The better starting question is "what does this assignment require that the student can only get by doing the work?"
Where to start
Before redesigning anything, run the assignment through EthosLearn's scan. The scan will classify the domain, score the criticality, identify specific risk flags, and propose scaffolding stages. That gives you a baseline — something concrete to react to rather than starting from scratch.
If the assignment scores READY, the scaffolding is generated automatically. If it scores NEEDS_INPUT, faculty decisions are required before the workspace goes live. If it scores DEEP_REDESIGN, the assignment itself needs structural changes before scaffolding is useful.
→ Go to EthosLearn to scan an assignmentWhat actually changes in a redesigned assignment Key distinction
There are two kinds of redesign. The first kind adds friction — more steps, more documentation, more hoops. Students feel the burden without gaining anything. The second kind adds anchors — requirements that force the student's specific knowledge, experience, or gathered data into the work. That's the kind worth doing.
| Friction-based (avoid) | Anchor-based (aim for) |
|---|---|
| "Submit a draft at each stage" | "At each stage, name the specific source or data that changed your thinking" |
| "Write a reflection on your AI use" | "Identify one conclusion the AI suggested that you rejected, and explain why" |
| "Include a bibliography" | "For each source, note how you found it and what it told you that the AI synthesis missed" |
| "Show your work" | "Document the specific judgment call you made that a student without your context could not have made" |
The business assignment split Important
Business assignments divide into two fundamentally different types, and they require different redesign strategies.
Type A — Applied research: The student brings domain or local knowledge. A market analysis for a specific rural area, an industry analysis in a field where the student has work experience. The AI provides macro pattern recognition; the student provides the intelligence the AI doesn't have. Scaffolding should surface and require that local knowledge explicitly.
Type B — Framework learning: The student is given a case, a company profile, or a constructed scenario. They're being evaluated on how well they apply Porter's Five Forces or a financial ratio framework — not on industry knowledge. The AI is very good at this. Redesign here means requiring judgment documentation: where did you make a decision the framework didn't make for you? What's missing from this case that would change your analysis?
Understand Risk in Your Discipline
AI substitutability isn't uniform across disciplines. The patterns are predictable — and knowing where they concentrate lets you focus redesign effort where it actually matters.
What the criticality score measures
EthosLearn assigns every scanned assignment a criticality score from 0.0 to 5.0. A higher score means the current assignment design provides less resistance to AI substitution — not that the subject matter is more important. A 4.8 business case study and a 4.8 sociology essay have the same problem: a student could submit AI-generated work and no authentic learning would need to have occurred.
The score reflects the assignment as written, not the subject. The same topic, redesigned with specific anchors, can drop from 4.5 to 2.0. The score is a starting point for conversation, not a judgment on the faculty member.
High-risk patterns by discipline cluster
| Discipline cluster | Highest-risk assignment types | What drives the risk |
|---|---|---|
| Business & Management | Generic case studies, industry overviews, business plans | LLMs trained extensively on business writing; standard frameworks well-represented |
| Social & Behavioral Sciences | Theory application papers, literature reviews | Academic writing conventions highly predictable; theory summaries templatable |
| Liberal Arts & Humanities | Analytical essays on canonical texts, historical analysis | Primary sources widely digitized; LLMs have extensive training data on most canonical works |
| Design & Media | Design rationale essays, process documentation without artifact | When the artifact itself isn't submitted, only the written rationale is evaluated |
| Technology | Code documentation, architecture write-ups, project reports | LLMs generate competent technical writing and code explanations |
The five flags that predict high criticality
These flag combinations consistently predict criticality scores above 4.0 regardless of discipline:
- No rubric detected — vague evaluation criteria give students no authenticity signal
- Generic prompt — the assignment could belong to any course in any institution
- Provided data or case — all inputs are pre-packaged; student brings no domain knowledge
- No described process — only the final deliverable is submitted; no evidence of how it was made
- No personal or specific anchor — nothing in the assignment requires the student's particular context
Any two of these together is a signal. All five together means the assignment as written is essentially a prompt for an LLM.
Design a New Project-Based Assignment
The most durable approach to AI-resistant assignment design isn't about restriction — it's about structural properties that make authentic student contribution irreplaceable. These properties can be designed in from the start.
Why project-based learning is the right frame for lower-level undergrad
Introductory and mid-level courses are often where the generic discussion board → essay pattern does the most damage. Students are learning professional workflows for the first time. The assignment that replicates the work they'll do on day one of a job is both more authentic and more AI-resistant than the assignment that asks them to summarize chapter three.
Project-based assignments situate learning in specific contexts. Specific contexts are what the AI lacks. A student analyzing the competitive landscape of a particular rural Ohio market in Q2 of this year is working with information no LLM has — which is exactly the point.
The seven resistance properties Design toolkit
These are structural features of assignment activities that require student-specific knowledge an LLM cannot fabricate. The more of these a project incorporates, the more durable it is against AI substitution — regardless of subject matter.
1. Temporal specificity
Requires knowledge of something that happened at a specific time — an in-class event, a particular moment in the student's own prior work.
2. Artifact specificity
Requires engagement with a specific artifact the AI doesn't have — a classmate's design, a particular dataset the student gathered, a live observation.
3. Process evidence
Requires showing work across multiple states with explanation of what changed — draft → revision → final with documented reasoning.
4. Personalization anchoring
Requires referencing something the student themselves produced in a prior stage — each stage depends on the previous one.
5. Relational specificity
Requires knowledge of specific peer interactions — feedback given to a named classmate's specific work, not generic peer review.
6. Real-time presence
Requires having been at a specific event — in-class session, studio visit, live critique, field observation.
7. Self-referential chain
Each stage requires input only the student has from the previous stage — the chain cannot be reconstructed from scratch.
Matching properties to disciplines
| Discipline | Naturally fitting resistance properties | Example activity |
|---|---|---|
| Design / UX | Artifact specificity, process evidence, relational specificity | Critique a specific classmate's prototype with reference to their stated design rationale |
| Business (applied) | Artifact specificity, temporal specificity, process evidence | Market demand analysis using primary data the student gathered in a specific timeframe |
| Writing / Humanities | Temporal specificity, personalization anchoring, self-referential chain | Revise your Week 2 essay using the theoretical framework introduced in Week 8 — explain what changed |
| Research (social science) | Artifact specificity, process evidence, personalization anchoring | Conduct three primary interviews and analyze them against your Week 3 hypothesis — document where the hypothesis held and where it broke |
Faculty FAQ
Faculty concerns about AI integration in course design are rarely about AI. They're about trust — trust in the approach, trust in the people recommending it, and trust that the institution will use what's collected fairly. Understanding which kind of concern you're facing changes what you do about it.
The three types of trust that matter Framework
Galford & Drapeau's organizational trust framework maps cleanly onto the adoption concern patterns we see in AI integration work. Before deciding how to respond, diagnose which kind of trust is missing.
Strategic trust
"Is the scaffold-over-prohibit approach even the right call?" Faculty who question the strategy need evidence that it works — not reassurance that it's policy.
Personal trust
"Do I trust the ID or department chair recommending this?" Peer credibility matters more than institutional authority here. This is where faculty relationships move things.
Organizational trust
"Will the institution use EthosLearn data to surveil me?" This fear is legitimate and specific. It needs a direct, transparent answer — not a deflection.
Common adoption concerns and what's actually underneath them
| What the faculty member says | What's often underneath it | What actually helps |
|---|---|---|
| "My assignments are fine as they are" | Fear that redesign implies their judgment was wrong | Frame the scan as new information, not a critique. The AI landscape changed; the assignment was designed before that was true. |
| "Students should just not use AI" | Strategic trust gap — they don't believe the scaffold approach is right | Provide evidence from disciplines similar to theirs. Don't argue the philosophy — show what the alternative looks like in practice. |
| "Who sees this data?" | Organizational trust gap — surveillance fear | Answer directly and specifically. Name who has access, what they can see, and what they cannot. Vague reassurance makes it worse. |
| "This is one more thing on my plate" | Capacity constraint, not resistance | Show the actual time cost of running a scan. Offer to run the first one together. Remove the barrier, don't argue past it. |
The "misplaced benevolence" problem
One of Galford & Drapeau's most useful observations: institutions often avoid pushing senior faculty with adoption concerns because of their status — letting them opt out quietly to preserve the relationship. This backfires. Other faculty notice who has to comply and who doesn't, and it signals that the initiative isn't real.
The better approach with senior faculty who have adoption concerns is a direct, respectful conversation that names the pattern: "You have more influence over how this lands in the department than anyone else. That's exactly why your engagement matters." This makes their seniority an argument for participation rather than an implicit exemption from it.