We trace what tuition and four years actually buy — lectures, practice, the degree, jobs and research. How much of it can only be learned at university? No school gets a single score; each function is shown separately, only as far as the evidence goes.
Much lecture content is already reachable through open courses, books and AI, but tuition (avg ₩7.27m a year in 2026), four years of forgone income and degree or licence filters can stand in front of it. First split what truly needs a university.
Neither “university is pointless” nor “university is essential”
Some paths (doctor, nurse, pharmacist, lawyer, architect) legally require a university programme; some goals (general knowledge, basic programming) can be met outside. The same university is worth different things for different goals. Figures come from official statistics, statutes and papers; AMS judgements are labelled; empty cells stay empty.
10-SECOND VIEW
Pick one major
4-year tuition
₩25.7mHumanities & social field avg ₩6.4m/yr × 4 (2026, before aid)
Treated as one product, a university is only “expensive” or “necessary”. Split into lecture, practice, credential, signal, network and research, some parts turn out replaceable and some hard to find elsewhere.
Lecture AMS ANALYSIS
Explaining concepts and theory in sequence
University dependence · Low
The same theory is in textbooks, open courses and AI; what stays university-specific is the researcher’s view and formal assessment.
AI substitutability · High
Explaining, repeating and answering are AI strengths; one RCT found a designed AI tutor beat active-learning classes in physics, while another found unguarded use reduced learning.
Online substitutability · High
MOOCs and recorded lectures cover the same ground, but in a fixed order without instant answers.
Workplace substitutability · Partial
Work pulls in theory as needed but tends to skip systematic foundations.
Practice AMS ANALYSIS
Hands-on work with equipment, patients or sites
University dependence · Depends on field
Clinics, labs and clean rooms are hard to access outside universities, hospitals or institutes; coding and design practice needs only a laptop.
AI substitutability · Low
AI can explain procedures and interpret results, but not manual skill, safety judgement or patient contact.
Online substitutability · Partial
Simulations, virtual labs and digital twins cover part of it, with no evidence they equal real practice.
Workplace substitutability · Depends on field
Where failure is cheap, work is the fastest teacher; where licences or safety apply, entry is closed.
Credential AMS ANALYSIS
A qualification the law or a system recognises
University dependence · Depends on field
Doctors, nurses, pharmacists, lawyers and architects must hold a statutory degree or programme to sit the exam; elsewhere it is the employer’s choice.
AI substitutability · Low
AI learning carries no official degree or licence.
Online substitutability · Partial
Some online routes (credit bank, cyber universities, some MOOC degrees) do lead to formal degrees.
Workplace substitutability · Low
Experience can qualify for some certifications but does not replace a degree.
Signal AMS ANALYSIS
What admission and graduation tell employers
University dependence · High
Getting in and finishing are signals apart from what was learned — the hardest part to replace.
AI substitutability · Low
Learning with AI is not itself a signal; it has to be proven through output.
Online substitutability · Partial
Certificate value varies widely by issuer and industry.
Workplace substitutability · Depends on field
In some fields a portfolio beats a degree; yet a US study found removing degree requirements from ads raised non-degree hiring by only ~3.5%.
Network AMS ANALYSIS
Relationships with people who shared the place and time
University dependence · Partial
Alumni, advisers and classmates are a university strength; we found no official measure of its size.
AI substitutability · Low
AI does not introduce people.
Online substitutability · Partial
Online communities and open source make loose ties; depth depends on the person.
Workplace substitutability · High
People you worked with are the most direct industry network.
Research AMS ANALYSIS
Producing new knowledge and learning how
University dependence · High
Labs, grants, advisers and theses sit in universities and institutes; research funding per full-time faculty was ₩123.6m in 2025.
AI substitutability · Partial
AI helps with literature, code and analysis; framing questions, experiments and accountability remain human.
Online substitutability · Low
Open-data research is possible, without equipment, ethics review or supervision.
Workplace substitutability · Partial
Corporate labs do research, with different publication and teaching aims.
University Varies with the professor AI Re-explains in many ways; can be confidently wrong
Answering questions
University Limited to class and office hours AI Instant, any hour
Personalisation
University One pace for dozens or hundreds AI Adjusts to the learner’s level
Repetition
University Hard to replay without recordings AI Unlimited
Currency
University Strong at the research frontier, slow curriculum change AI May not know, or misstate, recent facts
Pace
University Fixed by semester AI Learner’s pace
Curriculum flexibility
University Prerequisites and degree rules AI Can start from what is needed
Assessment
University Official grades and credits AI Self-checks only, no official standing
Practice
University Labs, clinics, workshops AI Explains procedures, supports simulation
Equipment
University Access to equipment individuals cannot buy AI None
Teamwork
University Team projects and seminars AI Solo by default
Degree
University Formal degree AI None
Social trust
University The school name is a signal AI Must be proven through output
Network
University Alumni, faculty, classmates AI None
Cost
University Avg ₩7.27m a year (2026) plus living costs AI Free to a monthly subscription
Time
University Usually four years on a set schedule AI Depends on goal and level (little measured data)
AI strengths
Questions at any hour · Repeated explanation · Adjustable difficulty · Links to other fields · Only the needed part, condensed · Instant worked examples · Diagnoses current level · Custom curriculum
Faculty strengths
Academic depth · A researcher’s view · Seminars · Assessment · Feedback · Experiments · Mentoring · Team projects · Path to graduate school · Official credit
Evidence — both directions
STUDY
In a crossover RCT in a Harvard physics course (194 students), topics learned with a pedagogically designed AI tutor showed larger gains than active-learning classes, in less time.
Limits: One school, one course, two topics; the authors frame it as freeing class time, not replacing instructors.
In a ~1,000-student high-school maths field experiment, AI raised practice scores (+48% base GPT, +127% tutor GPT), but once removed the base-GPT group scored 17% lower than controls; tutor guardrails largely prevented this.
Limits: High-school students; not directly transferable to university or adults.
There is no basis for “AI beats professors”. The evidence so far points both ways: designed AI tutors can work well in some courses, and answer-giving AI can reduce learning. Item-by-item university-level comparisons (facts, tutoring, writing feedback, problem solving, motivation, retention) will fill this in as they appear.
AI learning risks
Hallucination
Invents plausible facts and sources.
Outdated information
May miss later changes in law, technology or data.
Calculation errors
Gets intermediate steps wrong, confidently.
Shallow understanding
Copying answers leaves only the feeling of understanding.
Over-reliance
Students with unguarded GPT access scored 17% lower once it was removed than those who never had it (PNAS 2025, high-school maths).
No official credential
AI learning is not a degree or licence.
No physical practice
Cannot replace handling equipment, patients or sites.
Recorded lectures vs AI
Recorded
A fixed 60-lecture series
→Watched in order
→Rewind what you missed
AI
Question
→Check current level
→Skip what is known
→Explain only what is needed
→Immediate follow-up
Recorded courses give everyone the same order; AI starts where the learner is — with nobody checking when it is wrong.
LECTURE REPLACEMENT MAP
How much of each major’s lectures AI can cover
No percentage. Theory is mostly AI-assistable; the more a major ties lectures to practice, licences or research supervision, the harder it is to move out. Reasons are listed per major (AMS analysis).
Many labs (machining, materials testing, thermo-fluids) need equipment and safety control.Needs: Machine shop, Laboratory, Expensive equipment, Controlled safety environmentPartly covered by: CFD, Structural analysis (FEA), Digital twin
Chemical engineering · High
Chemistry labs and process plant need hazard control and costly equipment.Needs: Laboratory, Pilot plant, Expensive equipment, Controlled safety environmentPartly covered by: Process simulation, CFD, Digital twin
Electrical & electronic · High
Instrumentation, high voltage and clean rooms; some circuit design can be simulated.Needs: Laboratory, Clean room, Expensive equipment, Controlled safety environmentPartly covered by: Circuit simulation, Digital twin
Architecture · High
Design studio and practical training are central.Needs: StudioPartly covered by: BIM / 3D modelling, Structural analysis (FEA)
Medicine · High
Dissection, clerkships and patient care happen only inside teaching hospitals.Needs: Cadaver, Clinic, Patients, Laboratory, Controlled safety environmentPartly covered by: Medical simulation
Nursing · High
Clinical placement is a large share and needs hospitals.Needs: Clinic, Patients, Controlled safety environmentPartly covered by: Medical simulation
Digital design runs on personal kit; craft/product design needs studios.Needs: StudioPartly covered by: BIM / 3D modelling
Universities stay strong in labs, clinics, machine shops, pilot plants, cadavers, patients, clean rooms, costly equipment and controlled safety. CFD, process and circuit simulation, digital twins, virtual labs and medical simulation cover part of it — with no evidence they equal real practice.
LEARNING BY DOING
Traditional
Learn
→Exam
→Grade
→Graduate
→Work
Alternative
Basic learning
→Real project
→Failure / gap
→AI / mentor learning
→Apply again
Whether learning on the job fits depends on failure cost, safety, licensing, equipment and whether a portfolio can prove it.
Statisticsfailure Low · safety Low · portfolio High
Computer sciencefailure Low · safety Low · portfolio High
Mechanical engineeringfailure Partial · safety High · portfolio Partial
Chemical engineeringfailure High · safety High · portfolio Low
Electrical & electronicfailure Partial · safety High · portfolio Partial
Architecturefailure High · safety Partial · portfolio High
Medicinefailure High · safety High · portfolio Low
Nursingfailure High · safety High · portfolio Low
Life sciencesfailure Partial · safety High · portfolio Low
Designfailure Low · safety Low · portfolio High
TOTAL DEGREE COST
The real cost of a degree
Total = tuition + fees + books + extra housing + commuting + other costs + opportunity cost. Living costs and forgone income vary, so none are prefilled.
Average annual tuition, 2026₩7.3m192 institutions, +2.1% on 2025
Living & study costsBooks, commuting, extra housing, other₩0m
Opportunity cost(forgone monthly income − part-time) × months₩0m
Total economic cost₩29.1m
Payback: Unknown — enter an extra-income estimate
Do not plug in the raw graduate/non-graduate pay gap — it mixes ability, family background, industry, region and major, and is not the causal effect of university.
DEGREE ROI
Does the degree pay back?
Degree ROI = extra lifetime earnings from the degree − total cost. The hard part is that nobody knows the “from the degree” part precisely.
Korea: tertiary vs upper-secondary earnings (25–64)+30%OECD average +54% (EAG 2026); +31% in the 2025 edition
How much more graduates earn on average — +30% in Korea vs +54% OECD.
Causal Education Effect · causal
The difference for the same person with and without university. The observed premium mixes in ability, family, industry, region and major; the causal effect may be smaller, or larger for some majors. No official Korean estimate by major was found (unknown).
Payback Period · payback
Computed above from your own extra-income estimate; official payback by major is unknown.
School Brand Premium · brand
Starting pay can differ by school within a major, but that cannot be credited to the brand — different students enrol in the first place.
Admission Signal · admission
Passing a competitive admission is itself a hiring signal, separate from what the four years teach.
ROI by major
Field tuition and employment rates are official, but there is no official lifetime-earnings estimate by major, so ROI is not computed. Employment rates cover all higher education, 2024 graduates.
Humanitiesemployed 61.1%
Socialemployed 69%
Educationemployed 71.1%
Engineeringemployed 70.4%
Naturalemployed 65.4%
Medicalemployed 79.4%
Artsemployed 66.7%
Graduate employment rate, all higher education (2024)69.5%prior year 70.3%
GPA is not identical to market value. Grades show mastery of a set scope on a set schedule; earnings also depend on scarcity, revenue contribution, negotiation, execution and market size. Not “grades are useless” — they measure different things.
NEED-FIRST LEARNING
Start from the need, not the year
Curriculum-first
Year 1
→Year 2
→Year 3
→Year 4
Need-first
Problem at hand
→Identify needed knowledge
→Remove what is known
→Learn only the gap
→Apply immediately
E.g. a chemical engineer who already knows calculus, physics, thermodynamics and transport, aiming to understand machine design, needs statics, mechanics of materials, dynamics, machine elements and vibration — not four more years.
AI GAP LEARNING PLANNER
Already known (edit freely)
University Path
The full four-year programme for “Understand mechanical design”, including electives and foundations already learned
AI Gap Learning
Statics, Mechanics of materials, Dynamics, Machine elements, Vibration, CAD & drawings6 of 9 topics the goal needs
Skipped
Calculus, General physics, Materials science
Hours depend on your level and target, so none are invented. Hands-on work needing equipment or safety control is not covered by this route.
ADULT RESKILLING
Already a graduate — back to university for a new field?
Adults in non-formal learning (25–44 / 45–64)16% / 15%OECD average 52% / 45%
Korean adults take part in non-formal learning at about a third of the OECD rate. With work changing through the 30s, 40s and 50s, continuous reskilling may matter more than one four-year block.
Eight ways to learn AMS ANALYSIS
University degree
Cost
Highest (avg ₩7.27m/yr)
Time
2–4 years
Personalisation
Low
Practical relevance
Varies
Credential
Degree
Networking
Strong
Practice
Lab access
Feedback
Formal
Schedule flexibility
Low
Continuing education
Cost
Check each course
Time
Weeks to months
Personalisation
Low–medium
Practical relevance
Varies
Credential
Certificate / credit-bank credit
Networking
Medium
Practice
Varies
Feedback
Instructor
Schedule flexibility
Medium
Recorded lectures
Cost
Low
Time
Fixed by series length
Personalisation
Low (fixed order)
Practical relevance
Varies
Credential
Mostly none
Networking
Little
Practice
None
Feedback
Little
Schedule flexibility
High
MOOC
Cost
Free–low (paid certificates)
Time
Set weeks per course
Personalisation
Low
Practical relevance
University-level
Credential
Certificates, some degrees
Networking
Forums
Practice
Assignments, coding
Feedback
Auto/peer grading
Schedule flexibility
High
AI tutor
Cost
Free–monthly fee
Time
Little measured data
Personalisation
High
Practical relevance
High if questions are real
Credential
None
Networking
None
Practice
Simulation support
Feedback
Instant, can be wrong
Schedule flexibility
Very high
Bootcamp
Cost
Check each programme
Time
Months, intensive
Personalisation
Medium
Practical relevance
High (job-focused)
Credential
Certificate
Networking
Cohort, partners
Practice
Projects
Feedback
Mentors
Schedule flexibility
Low
Project learning
Cost
Low
Time
Goal-dependent
Personalisation
High
Practical relevance
Very high
Credential
Portfolio
Networking
Collaborators, users
Practice
Real
Feedback
Outcomes, users
Schedule flexibility
High
Workplace learning
Cost
Earns income
Time
Continuous
Personalisation
Task-driven
Practical relevance
Highest
Credential
Experience
Networking
Strong
Practice
Real
Feedback
Peers, managers
Schedule flexibility
Work schedule
Continuing-ed and bootcamp prices vary by course, so no average is given. No study comparing time to competency across university, recorded courses, AI and projects was found — unknown.
GOAL-BASED EDUCATION
Fitting route
Projects + portfolio + internship — unless postings in the sector filter on degrees.
Check first
Check real postings in the sector first.
MAJOR–JOB MATCH · DEGREE REQUIREMENT
Do graduates use their major, and is the degree really required?
4-year graduates in major-related jobs73.8%KRIVET 2016, GOMS data
Six axes instead of one “degree necessity” score; employer filters and network value have no measured data yet.
Humanities
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
History
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
Economics
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
Business
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
Law
License dependency
High
Equipment dependency
Low
Employer degree filter
High
Credential value
High
Network value
No data
AI lecture substitutability
Partial
Mathematics
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
Statistics
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
Computer science
License dependency
Low
Equipment dependency
Low
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
High
Mechanical engineering
License dependency
Low
Equipment dependency
High
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
Partial
Chemical engineering
License dependency
Low
Equipment dependency
High
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
Partial
Electrical & electronic
License dependency
Partial
Equipment dependency
High
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
Partial
Architecture
License dependency
High
Equipment dependency
Partial
Employer degree filter
No data
Credential value
High
Network value
No data
AI lecture substitutability
Partial
Medicine
License dependency
High
Equipment dependency
High
Employer degree filter
High
Credential value
High
Network value
No data
AI lecture substitutability
Partial
Nursing
License dependency
High
Equipment dependency
Partial
Employer degree filter
High
Credential value
High
Network value
No data
AI lecture substitutability
Partial
Life sciences
License dependency
Low
Equipment dependency
High
Employer degree filter
No data
Credential value
Partial
Network value
No data
AI lecture substitutability
Partial
Design
License dependency
Low
Equipment dependency
Partial
Employer degree filter
No data
Credential value
Low
Network value
No data
AI lecture substitutability
High
Representative jobs
Doctor
Licence Required — medical degree, then national exam University Very high dependence AI Supports theory
Software developer
Licence None — degree demand varies by employer University Employer-dependent AI Large share of teaching replaceable; portfolios can carry high value
Historian / humanities learner
Licence None University Low for knowledge, high for professional research AI Knowledge via AI, books, MOOCs; source criticism and research training favour graduate school
DEGREE INFLATION · EMPLOYMENT FILTER
HBS and the Burning Glass Institute (2024) found US postings dropping degree requirements rose sharply in 2014–2023, yet non-degree hiring rose only ~3.5% on average — ad wording and real filters can differ. Extracting degree/experience/portfolio/certification requirements from Korean postings is not done yet.
Caution: education spending includes government funding, salaries and scholarships paid from non-tuition income, so the difference is not a “student gain”. Different accounting bases — use only for scale.
Faculty salary · Staff salary · Student education · Research · Facilities · Administration · Marketing · Scholarship · Reserve fund · Other
Per-school shares can be computed from each school’s settled accounts on Academyinfo but are not loaded yet; no official national breakdown was found, so none is shown.
How the accounts are split
School account
Runs the school from tuition and grants; split into tuition and non-tuition accounts.
Tuition account
Tuition income and what it pays for.
Non-tuition account
Grants, donations, foundation transfers and their use.
Foundation account
The school foundation; statutory contributions and transfers flow from here.
Industry-cooperation account
Research grants and technology-transfer income.
RESEARCH VALUE · TEACHING vs RESEARCH
Research quality ≠ teaching quality. Faculty work splits into teaching, research and administration; strong publications, citations, patents, grants or tech transfer do not guarantee good teaching. Individuals are not rated — only department-level public data.
No official source publishes facility usage alongside investment (unknown).
UNIVERSITY PROFILE
Per-school profiles will be filled from Academyinfo; no school is filled with invented values: Annual tuition, Scholarships, Students, Full-time faculty, Students per faculty, Education spending per student, Research output, Employment rate, Audit history, Finances (school, foundation, industry-cooperation), Procurement issues.
PROCUREMENT WATCH
Did the university pay a market price?
Contracts for chairs, computers, equipment or construction are compared with same-spec market prices. Only cases with the contract on record are published — none yet — but you can compare one yourself.
Item: Enter both prices
Computed from your inputs only, nothing is stored. AMS publishes procurement cases only with the contract on record.
Single bids or repeated failures may signal tailored terms.
These are risk signals, not findings; only audit dispositions and rulings count as confirmed. Supplier relationships (family, former employment, repeat contracts, affiliates) are shown only from official records.
AUDIT · PRIVATE-SCHOOL MISCONDUCT
Where did the university’s money go?
Allegation, audit finding, investigation, indictment, conviction, acquittal, sanction and closure are kept apart; unconfirmed cases are never written as confirmed, and no generalisation is made about private schools as a whole.
Seonam University and others (founder Lee Hong-ha) CONVICTION
Jan 2007 – Aug 2012 · Misuse of school funds, Embezzlement
The founder was tried for embezzling about ₩100.3bn, mostly school funds, from schools he founded including Seonam and Hanryeo.
Supreme Court confirmed 9 years and a ₩9bn fine in May 2016; Seonam closed in February 2018.
Source priority: Ministry of Education audits → Board of Audit and Inspection → Courts → Prosecutors / police → University internal audit → Press. Types covered: Misuse of school funds, Embezzlement, Breach of trust, Construction costs, Improper appointments, Nepotism, Research-grant fraud, Admissions fraud, Foundation-transfer issues, Improper contracts. Reserve trends and background in Where does tuition go (Korean).
EXAM WASTE · ACADEMIC TIME
How much exam time survives into work?
A frame for splitting four years into lectures, test prep, assignments, required liberal arts, major, labs, projects, internships and job prep is ready; with no school- or major-level time-use data, no estimates are filled in.
Time buckets: Lectures · Test prep · Assignments · Required liberal arts · Major · Labs · Projects · Internships · Job prep
Tracked for work relevance: Lecture time · Assignments · Memorisation · Exam prep · Projects · Labs · Internships
Lifecycle: Admission → Lectures → Exams → Internship → Graduation → First job → 5 years on → 10 years on (1/5/10-year outcomes unknown until official linked data is checked)
COUNTRY COMPARISON
Across countries
US public four-year in-state tuition & fees 2025-26$11,950
US private nonprofit four-year 2025-26$45,000
Government spending per bachelor’s+ student (2023)USD 8,714 (48% of total)OECD average USD 16,405 (68%) — the rest falls mostly on households
US costs add room & board and a large loan system, so tuition alone misleads. Korea’s government covers 48% of bachelor’s+ spending versus 68% OECD, leaving more to households; 14.3% of Korean 4-year students (463,764) used student loans in 2025. Japan/Europe and per-country debt are not added yet.