TFIS · AI Ethics in Education · Course
The AI Revolution in
Higher Education
Opportunities and Ethical Challenges
Session 1 · Tuesday 1 September 2026 · 2.30–4.30 afternoon
Give me a real assignment question.
From a course you teach this semester. Anyone.
[ live AI chat on screen — P01 · no slides, no introductions yet ]
Cold open · debrief
That took about ninety seconds.
SAY
"Grade it mentally. Most of you just gave it something between a B and an A.
Now here is the only honest question, and it is the question of this entire course:
not how do we stop this — we cannot, and I will show you the evidence —
but what do we do because of this? Selamat datang. Let's begin properly."
Axiom A1
The gap is the curriculum.
- Stanford HAI's 2026 AI Index — the field's independent report card — has one headline finding: the gap between what AI can do and how prepared we are to manage it is widening.
- Capability compounds. Governance, evaluation, and education lag.
- That gap is not a problem to lament. It is the curriculum. Today we understand it. Tomorrow we build inside it. Wednesday we ship policy that closes our institution's share of it.
2026 AI Index Report, Stanford HAI
Baseline · 15 min · private
EAI-CMM — where do you actually stand?
- 20 statements · five pillars: Literacy, Pedagogy, Integrity, Discernment, Governance.
- Score 0–4 each. Score what you do, not what you believe. Inflated baselines steal your own delta.
- You retake this Wednesday morning. The delta is yours to keep.
Show of hands by band afterwards — pillar totals only, anonymous.
Data walk
The state of AI, 2026
All figures: 2026 AI Index Report (Stanford HAI) unless noted
Data walk · 1 of 7
Capability is accelerating, not plateauing
60→~100%
SWE-bench Verified (real software-engineering tasks) in a single year
IMO gold
Frontier models reached gold-medal level at the International Mathematical Olympiad, 2025
2026 AI Index Report
Data walk · 2 of 7
…but the frontier is jagged
50.1%
the same model class reading an analog clock correctly
Benchmark headlines are a poor proxy for behaviour on your tasks. This is why Discernment is a core competency — and why "just trust it" and "just ban it" are both wrong.
2026 AI Index Report
Data walk · 3 of 7
Adoption outran every prior technology
88%
organizational adoption
53%
of the population reached by generative AI — faster than the PC or the internet
$581.7B
global corporate AI investment
2026 AI Index Report
Data walk · 4 of 7
Agents arrived
~12→66.3%
agent accuracy on OSWorld (structured computer-use tasks)
The thing your students use next year won't just answer. It will do.
2026 AI Index Report
Data walk · 5 of 7
Your students are already there
4 in 5
US high-school and college students using AI for schoolwork
~half
of schools have AI policies at all
6%
of teachers say those policies are clear
Assume comparable or higher exposure in your lecture halls.
2026 AI Index Report, Education chapter
Data walk · 6 of 7 · write this down
"We are running a natural experiment on our students — without a control group."
Rigorous evidence on learning outcomes from AI-enhanced education remains limited.
That is itself an ethical fact. It returns tonight, in Session 2.
2026 AI Index Report, Education chapter
Data walk · 7 of 7
What students actually do with AI
- Usage splits ~evenly across four modes — direct problem-solving, direct output creation, collaborative problem-solving, collaborative output creation (23–29% each).
- Nearly half of all use is direct — asking for answers or finished artifacts.
- CS students wildly overrepresented: 36.8% of conversations vs 5.4% of US degrees. Your STEM students moved first; your humanities students are moving now.
Anthropic Education Report — 1M anonymized student conversations, Apr 2025
The other column
The opportunity ledger
Ethics that only inventories harms is theater.
Opportunity · evidence 1
Expertise at marginal cost ≈ zero
+4pp
student topic mastery when human tutors got real-time AI assistance
+9pp
for students of the lowest-rated tutors
The pattern that works: AI amplifying a human educator, not replacing one.
Wang, Demszky et al., Tutor CoPilot (Stanford, 2024)
Opportunity · evidence 2 · full treatment Session 5
Guardrailed tutoring works
+127%
practice performance with a GPT-4 tutor engineered to give hints, not answers — without damaging later unassisted performance
Design determines outcome. Hold that sentence — Tuesday afternoon it becomes the most important study of the three days.
Bastani et al., PNAS (Penn/Wharton, 2025)
Opportunity · deployed at scale
This already exists in production
- Harvard CS50 runs a course-wide AI "duck debugger" deliberately built to guide toward answers rather than hand them over.
- Claude's Learning Mode responds with Socratic questioning instead of direct answers.
Live demo (3 min): ask Learning Mode the cold-open assignment question. Watch it refuse to just answer.
Malan et al. (CS50); Anthropic
Opportunity · your side of the desk
Where faculty get leverage
- Across ~74,000 higher-ed conversations: curriculum design is the top faculty use case.
- Educators lean on AI hardest for building materials — not for grading humans. That's the right instinct.
- Session 4 trains it, hands-on.
Anthropic faculty-usage education report, 2025
Exercise · 30 min · discipline groups
The Threat/Gift Matrix
| Gift | Threat |
|---|
| Teaching | ≥3 items | ≥3 items |
| Assessment | ≥3 items | ≥3 items |
| Campus Life | ≥2 items | ≥2 items |
- Every item concrete enough to name a course.
- Circle the single most urgent threat and the single most undervalued gift.
- Report both in 60 seconds.
Master chart stays on this wall for all three days. Sessions mark items off as they address them.
"Notice how many threats are actually assessment-design problems wearing a technology costume."
Hold that thought until tomorrow morning.
Exit ticket · index card
3 · 2 · 1
- 3 facts from today that survived contact with your skepticism
- 2 things you want to try with AI this week
- 1 question you need answered before Wednesday
Session 3 opens by answering the three most common questions.
Contrarian close
SAY
"The threat to this institution is not that students will cheat with AI. The threat is
institutional denial — running 2019 assessments in a 2026 world and calling the resulting
numbers 'learning.' A student who uses AI on a take-home essay has not defeated your
assessment. They have audited it. Tonight, we do ethics properly. Jumpa lagi at eight."
Appendix · facilitator only
Offline demo fallbacks
- [ insert screenshot: P01 cold-open run — pre-captured on a typical assignment ]
- [ insert screenshot: Learning Mode refusing to answer directly ]
- Two AI accounts pre-logged (primary + backup vendor). Live demos fail; vendors differ.
MIT Report cross-reference
- The "Campus Life" matrix row addresses the MIT Ad Hoc Committee's finding that AI is "increasing isolation, eroding the social contract, undermining study groups and office hours" (MIT AI & Education Report §1.1, Aug 2026).
- Debrief cue if groups raise study-group decline or empty office hours: "MIT saw it too — §3.2.1 asks 'why it matters for students to go to college.' Their answer: structured in-person social learning must be designed, not hoped for."
- Recommended reading: MIT report at aiandeducation.mit.edu/report/ — share the link; participants can explore the 8 principles and 30 sub-recommendations between sessions.