GenAI in Higher Education — Faculty Companion (HKUST)

Helping faculty learn and apply the HKUST CEI Guidelines and Principles on Using Generative AI

⚠ Disclaimer: This companion is based on CEI guidance; it is not official HKUST policy. Always consult the latest HKUST policy pages.

Welcome, Faculty Member

Generative AI (GenAI) is reshaping higher education faster than governance can keep pace. This companion helps you navigate HKUST's guidelines — not just by understanding the rules, but by developing principled judgement about when and how AI should support student learning.

The companion is built around the Snap-to-Solve five-step process — Snapshot, Diagnose, Triage, Determine, Communicate — plus the CRAFT governance framework, AI literacy foundations, readiness reflection, and an assessment strategy library.

GenAI policies should flow from what counts as meaningful evidence of learning, not the other way around. Use this companion as a thinking partner, not a compliance checklist.

Navigate All Modules

🏛️

CRAFT Governance

⚡

Snap-to-Solve Wizard

📊

Diagnostic Grids

🎯

Assessment Strategies

🧠

AI Literacies

📡

AI Risk & Readiness

📚

Resources & Support

Your Learning Progress

0 / 7 modules explored

🏛️ CRAFT Governance Framework

Five interdependent areas for responsible GenAI adoption in higher education

About CRAFT: The CRAFT framework identifies five areas institutions must attend to for responsible GenAI adoption: Culture, Rules, Access, Familiarity, and Trust. Uncoordinated adoption across these areas risks amplifying inequities and entrenching opaque AI use.
Framework by Danny Y.T. Liu and Simon Bates, from APRU Whitepaper (2025)

Interactive CRAFT Wheel

Click a segment to explore each dimension

← Click a segment to explore

All Five CRAFT Dimensions

How CRAFT frames this Faculty Companion: Each module in this companion connects to CRAFT. Snap-to-Solve addresses Rules at the course level. Diagnostic Grids help develop Familiarity. Assessment Strategies strengthen Culture. AI Literacies build Familiarity and Trust. The Readiness Radar reflects institutional Access and Governance.

Credit: CRAFT framework by Danny Y.T. Liu and Simon Bates. APRU Whitepaper "Generative AI in Higher Education: Current Practices and Ways Forward" (2025). Landing page | PDF

⚡ Snap-to-Solve Wizard

A five-step process to review and redesign assessments for a world where students use generative AI

How to use: Work through each step in sequence, or jump to any step. You can start with a single key assessment. Your inputs are saved automatically. Use the export button in Step E to download your policy statement.

Step A — Snapshot: Examine Your Assessment

Before drafting GenAI policies, examine your assessment tasks and their pedagogical rationale. Determine the Intended Learning Outcomes (ILOs), then identify what skills or knowledge your assessment is designed to foster.

Assessment name / title

What are the Intended Learning Outcomes (ILOs)?

What skills or knowledge does this assessment foster?

Values of Inquiry

Select the values that best describe what your assessment is designed to cultivate. Clicking a value shows its definition and example questions.

Core 8 (Ellerton, 2022):

+ Show additional values

GenAI Risk Notes

Based on your ILOs and selected values, which aspects could GenAI easily simulate? Which genuinely require human judgment?

GenAI could easily simulate:

Requires genuine human judgment:

Diagnosing: Assessment not yet named — return to Snapshot

Grid 1: Cognitive Demand × GenAI Resilience

Map your assessment to understand how complex the thinking is and how easily GenAI can complete the task. The top-right quadrant (High Demand + High Resilience) is the most desirable.

Cognitive Demand (Bloom's) →High

⬆ High Demand
Low Resilience

Ideal for practicing critical AI use; GenAI can assist but student must evaluate deeply

★ High Demand
High Resilience

Most desirable: fosters deep learning and critical thinking; resilient to GenAI

Low Demand
Low Resilience

Least desirable; consider repurposing as low-stakes formative only

Low Demand
⬆ High Resilience

Good for verifying actual learning; difficult for GenAI, easy for human

GenAI Resilience → High

Grid 2: AI Leverage Potential × Required Human Agency

Evaluate whether tasks are assigned to GenAI and students appropriately. The top-right quadrant represents the most productive human-AI collaboration.

AI Leverage Potential →High

High AI Leverage
Low Human Agency

GenAI does heavy lifting; student checks output only — limited learning

★ High AI Leverage
High Human Agency

Both AI power and student judgment are substantial — richest outcomes

Low AI Leverage
Low Human Agency

Low-yield automatable activity — consider dropping or replacing

Low AI Leverage
⬆ High Human Agency

Primarily unassisted human expertise — ideal when verifying foundational skills

Required Human Agency → High

Grid 3: Cognitive Offloading Risk × Collaboration Depth

The bottom-right quadrant is ideal: deep human-AI collaboration with low surrender risk. Use structured offloading scaffolds to move tasks toward this quadrant.

Cognitive Offloading Risk →High

High Risk
Low Collaboration

Cognitive surrender risk — AI replaces human effort; passive consumption

High Risk
High Collaboration

Superficial reliance — copying solutions without paraphrasing; illusion of mastery

Low Risk
Low Collaboration

Strategic delegation for routine tasks; acceptable when human skills are primary

★ Low Risk
High Collaboration

Ideal: deep human-AI synergy; human-AI loops; metacognitive engagement

Collaboration Depth → High

Your Diagnostic Profile

Notes from diagnosis (where does your assessment fall, what did you observe?)

Step C — Triage: Categorise Your Assessment

Select the category that best describes your assessment's current situation. This translates your diagnosis into a concrete design decision.

🤖

Automation Alert

The task can be completed almost entirely by AI with little authentic thinking

🚔

Policing Trap

Preventing AI misuse requires burdensome monitoring that adds more busy-work than learning

⚠️

Outcome Mismatch

After applying AI rules, the task no longer produces evidence of the stated learning outcome

💤

Engagement Gap

Sound on paper, but students are likely to engage only superficially

📎

Over-Scaffold

Evidence requirements are so heavy they swamp feedback and motivation

✅

Clear Path

No red flags; keep with minor tweaks and a clear AI-use statement

Values of Inquiry Check

Ask yourself: "Does this assessment, as currently designed, generate evidence of the Values of Inquiry I selected in Snapshot?" If uncertain, treat this as an Outcome Mismatch.

My triage decision and planned actions:

Step D — Determine: Set GenAI Rules with the AIAS

The AI Assessment Scale (AIAS) provides five non-hierarchical levels. Select the level appropriate for your assessment — no level is inherently better than another.

Credit: AI Assessment Scale by Mike Perkins, Jasper Roe, and Leon Furze. aiassessmentscale.com | DOI: 10.53761/rrm4y757 | DOI: 10.37074/jalt.2025.8.2.15

1

No AI

2

AI-Assisted Planning

3

AI-Assisted Task Completion

4

Full AI

5

AI Exploration

AI Usage Map

Break down AI permissions by assignment stage. This prevents the AIAS level from remaining an abstract label and removes ambiguity for students.

Assignment Stage	Permission	Notes

Step E — Communicate: Tell Your Students

Draft your communication statements based on your inputs above. Edit as needed, then export.

Syllabus statement (faculty tone)

Student-friendly explanation (supportive tone)

Acceptable vs. unacceptable examples

Short FAQ

Student Partnership Tip: Invite students to discuss the proposed AIAS level at the start of semester. Explain why certain boundaries exist and ask what they would add or change. Student co-ownership increases compliance and trust.

📖 Example Walkthrough: Redesigning a Literature Review Essay

This is not a workflow to “output” a policy. It is a pedagogical thinking sequence that helps faculty exercise judgement at each step. Click each step to explore Dr. Chen’s reasoning process.

📊 Diagnostic Grids

Three 2×2 visual tools for diagnosing how GenAI interacts with your assessments

How to use: Click any quadrant to reveal detailed meaning, design implications, and redesign moves. These grids are also embedded in the Snap-to-Solve Wizard (Step B). Use the tabs below to switch between grids.

Y-axis: Cognitive Demand (Bloom's Taxonomy — Remember to Create)
X-axis: GenAI Resilience (Low to High)

Cognitive Demand → Higher

High Demand
Low Resilience

Analyse/Evaluate/Create tasks GenAI can handle well

★ High Demand
High Resilience

Ideal zone — deep thinking AND GenAI-resilient

Low Demand
Low Resilience

Lowest value; GenAI can fully automate

Low Demand
High Resilience

Verifies baseline human knowledge

GenAI Resilience → Higher

Note: grid meaning depends on how you define each axis. A task's cognitive demand is measured through Bloom's Taxonomy action verbs. GenAI resilience is assessed by testing the task against current reasoning models.

Click a quadrant to see detailed guidance

Y-axis: AI Leverage Potential (how much can AI genuinely add?)
X-axis: Required Human Agency (how much student thinking is required?)

AI Leverage Potential → Higher

High AI Leverage
Low Human Agency

AI does heavy lifting; minimal student thinking required

★ High AI Leverage
High Human Agency

Best of both: AI augments human judgment

Low AI Leverage
Low Human Agency

Low-yield automatable activity — consider dropping

Low AI Leverage
High Human Agency

Unaided human expertise — verifies foundational competence

Required Human Agency → Higher

Click a quadrant to see detailed guidance

Tri-System Theory (Shaw & Nave, 2026): GenAI functions as System 3 — a third cognitive system. System 1 is fast intuitive thinking; System 2 is slow deliberate reasoning; System 3 is external AI cognition. Cognitive surrender occurs when System 3 overrides both — participants followed AI recommendations ~80% of the time and their accuracy yoked to AI quality rather than their own reasoning.

Y-axis: Cognitive Offloading Risk (High = cognitive surrender risk)
X-axis: Collaboration Depth (Low = transactional, High = rich iterative human-AI loops)

Cognitive Offloading Risk → Higher

High Risk
Low Collaboration

Cognitive surrender — AI replaces thinking entirely

High Risk
High Collaboration

Superficial reliance — copying without paraphrasing

Low Risk
Low Collaboration

Structured offloading for routine tasks; acceptable with acknowledgement

★ Low Risk
High Collaboration

Ideal: human-AI loops, deep collaboration, metacognitive engagement

Collaboration Depth → Higher

Axis warning: grid meaning depends on how you define "offloading risk" and "collaboration depth" in your disciplinary context. These are analytical lenses, not absolute categories.

Click a quadrant to see detailed guidance

⚠ High Risk × Shallow Collaboration

When tasks land in the top-left quadrant (High Risk, Low Collaboration), immediate redesign is needed. Signs: students submit single final products with no intermediate work; no requirement to explain or critique AI outputs; GenAI can complete the entire task end-to-end.

Redesign moves:

Redesign immediately — do not rely on discursive changes alone
Add friction and checkpoints (intermediate submissions, live check-ins)
Shift modality: add oral component, require personal data, use live performance
Require students to annotate and critique AI outputs before incorporating them
Build reflective journaling into each stage of GenAI use
Use peer-AI co-revision workshops

✅ Structured Cognitive Offloading — When Offloading Enhances Learning

Not all offloading is harmful. When deliberately structured — students delegate lower-order tasks while redirecting cognitive resources toward analysis, evaluation, and reflection — offloading can actually enhance higher-order thinking.

Evidence: Research shows that a semester-long "cognitive offload instruction" model can produce significantly greater gains in critical thinking than traditional instruction, with a large effect size, when paired with metacognitive prompts and peer-AI co-revision.

Design strategies for structured offloading:

Require students to annotate and critique AI outputs before incorporating them, keeping System 2 active
Use reflective journaling at each stage of GenAI use
Employ peer-AI co-revision workshops
Add metacognitive prompts: "What did you delegate to AI? Why? What did you learn?"

Key Paradoxes this grid helps address:

Efficiency vs. Engagement: Students can complete tasks faster with GenAI, but this often reduces engagement with content
Assistance vs. Strategy Breadth: AI's persuasive outputs steer users toward first suggestions, narrowing problem-solving approaches
Apparent vs. Actual Higher-Order Thinking: Students may seem to practice higher-order thinking but are actually offloading it to AI — Shaw & Nave (2026) found confidence increased even when AI was systematically wrong

🎯 Assessment Strategies

A unified library of strategies for designing GenAI-aware, learning-centred assessments

How to use: Click any strategy card to expand its full guidance. Each card includes: Focus, When to use, Design moves, Evidence ideas, AI guidance, Watch for, and Faculty prompts. Use the Diagnostic Grids to determine which strategies apply to your assessment.

🧠 AI Literacies

From operational competencies (little-al) to empowerment-oriented sociotechnical practices (Big-AL)

Understanding AI Literacy: little-al vs Big-AL

Most AI literacy frameworks treat literacy as a portable set of individual competencies — prompting, output evaluation, bias awareness, and policy compliance. This is little-al. It is necessary, but not sufficient.

Big-AL (Big Artificial Literacies) goes further: it names the collective practices through which educational communities govern how AI reshapes judgement, authority, and values. Drawing on sociocultural literacy theory (Street, 1984; Gee, 2015) and critical technology scholarship (Franklin, 1990), Big-AL treats AI literacy as participation in sociotechnical Discourses — not just tool competence.

little-al (operational AI competencies)

Understanding what GenAI is and how it works
Using tools effectively: prompting, evaluating outputs
Recognising bias, hallucinations, limitations
Following institutional policies and ethical guidelines
Citing AI-generated material appropriately

Answers: "Can individuals use AI competently?"

Big-AL (sociotechnical empowerment)

Perceiving AI as a sociotechnical system reshaping authority
Monitoring cognitive routes: offloading vs. surrendering
Identifying non-delegable practices and values boundaries
Recognising gradual disempowerment dynamics
Participating in collective governance of AI adoption

Adds: "Can communities govern AI-mediated cognition so agency remains human-governed?"

Key insight: Even well-trained users can drift into routine reliance and cognitive surrender when AI is fast, fluent, and rewarding. Competency frameworks teach people how to obtain high-quality outputs, but do not address how prevalent tool use interacts with cognitive architecture, motivation, or institutional incentives.

The Five Big-AL Literacies

These five literacies are non-linear and interrelated — they should be read as "attention lenses" that co-develop and bleed into one another. Click each dimension to explore.

← Click a dimension to explore

All Five Big-AL Literacies

Cognitive Delegation Matrix

Non-linear interpretation: The same person or course can show different delegation modes across tasks, weeks, or conditions. Treat these as diagnostic patterns, not fixed categories of people. The goal is to identify where practices are drifting and design structures that pull them back toward constructive offloading.

These are best treated as context-sensitive patterns, not a developmental ladder. The same person or course can show different modes across tasks, weeks, or conditions. Click each pattern to explore its signals and faculty response moves.

🔄 Constructive Offloading

▼

Productive support

⚠️ Routine Reliance

▼

Habitual default

🚨 Cognitive Surrender

▼

High concern — judgement outsourced

🚫 Disuse / Algorithm Aversion

▼

Rejection after salient error

⛓️ Abuse / Enforced Automation

▼

System-imposed displacement

Non-linear interpretation: The same person or course can show different modes across tasks, weeks, or conditions. Treat these as diagnostic patterns, not fixed categories of people.

🎯 Design Implications for Faculty

Do not treat output quality checks as sufficient for learning-quality assurance.
Make route-level evidence visible (decisions, rationale, revisions, calibration, what was delegated).
Identify context-specific non-delegable practices (e.g., consequential feedback, mentoring, sensitive advising, high-stakes judgement).
Design for agency: require choices, justification, and AI-independent demonstrations where needed.
Connect classroom observations to institutional governance conversations.

Big-AL adds an empowerment lens to existing frameworks rather than replacing operational AI literacy. The question is not “skills or critique?” but “what kind of practice and judgement do these skills serve?”

⚠️ Counterpoint / Tension

Efficiency trap: A workflow can look successful by speed, satisfaction, or polished outputs while still eroding judgement, agency, or epistemic fluency.

✅ Constructive Stance

Big-AL is not anti-AI. It legitimises constructive offloading when bounded by metacognitive monitoring, values alignment, and governance clarity.

Think like an alien anthropologist: If you only observed prompts and outputs, what essential parts of learning would remain invisible in your course? Those are candidates for process-visible assessment and agency-preserving design.

📡 AI Risk Management & Readiness

Ten-dimension reflective framework for holistic institutional AI readiness

Anti-compliance safeguard: This tool is designed for holistic, reflective exploration — not scoring, benchmarking, or pass/fail assessment. Uneven profiles are normal; institutions need not be at the same level across all dimensions. There is no total readiness score.

About the framework: Ten Dimension AI Readiness Framework (DEC 2025 Artificial Intelligence Working Group). Designed for exclusive use of HKUST. Chaired by Sean McMinn, Director of Center for Education Innovation, HKUST. The framework covers AI broadly — not only GenAI — including learning analytics, adaptive tutoring, AI-driven research, and administrative automation.

AI Readiness Radar

Click a dimension axis to explore reflective prompts. Adjust sliders in the panels below to update the visualization.

← Click a dimension axis to explore

All Ten Dimensions — Reflective Exploration

Systems-thinking feedback: As you explore dimensions, watch for patterns in your profile. For example:

High Operational Readiness + Low Faculty PD may indicate capacity under-utilisation risk — infrastructure exists but practitioners may struggle to leverage it
Strong Stakeholder Engagement tends to act as a catalyst for improvements in Teaching, Learning, and Assessment Strategies
Gaps in Governance + High AI Literacy may create inconsistent practices that undermine trust

Note: These are interpretive patterns that may / can / tend to apply — not prescriptions about what must happen.

Why decision-making under uncertainty? Current AI governance in higher education treats AI challenges as calculable risks — known unknowns that can be predicted and controlled. But AI in education increasingly operates under Knightian uncertainty — unknown unknowns that must be dealt with as they arise. This framework shifts from compliance-based risk management to adaptive, reflective governance.
Framework from McMinn, S. & Lu, Z. (2026). “AI Risk Management in Higher Education.” CEI, HKUST.

Three-Pillar Framework

Click a pillar to reveal what it explains and what institutions often miss.

Five Faculty AI Literacies (Risk- and Empowerment-Oriented)

Interrelated capacities (not sequential skills) supporting reflective, empowered engagement in teaching and learning. Click a literacy to reveal how it connects to the others.

Adaptive Governance Cycle Map

A four-stage cycle: Assess → Engage → Develop → Sustain. Click a stage to explore its focus, output, and example actions.

Faculty Development & Governance Cycle (Assess → Engage → Develop → Sustain)

1 — Assess

Diagnose practices, perceptions, and risks

2 — Engage

Co-design and dialogue across roles

3 — Develop

Targeted capability building and redesign

4 — Sustain

Institutionalise adaptive governance

📚 Resources & Support

Glossary, HKUST links, training resources, references, and feedback

Search the glossary

CEI — Center for Education Innovation

Official HKUST GenAI guidance, policy documents, tools, and workshops.

Visit CEI ↗

HKUST GenAI Policy (2023)

Policy on Generative Artificial Intelligence for Teaching and Learning at HKUST.

CEI Policy Page ↗

HKUST GenAI Platform

Institutional access to GenAI tools for staff and students. Centrally supported and secure.

ITSO ↗

Academic Integrity & Honor Code

HKUST Registry guidance on academic integrity and the Academic Honor Code.

Registry ↗

Note: Links open in a new tab. For the most current information, always check the linked pages directly.

Key Citations

Share Feedback

This companion cannot be perfected through a top-down approach alone. Experiences and feedback from all members of the HKUST community are welcome.

Your name (optional)

Type of feedback

Your feedback

Note: This sends an email via your mail client. For direct contact: cei@ust.hk