Adding AI courses is not university reform

Korean universities do not become AI institutions by adding disconnected courses, certificates, or department labels.
A curriculum creates capability only when prerequisites, assessment, feedback, and independent work form a coherent dependency system.
Reform should be judged by what students can integrate and defend rather than by the number of AI offerings.

The Korean Policy Question

The rapid growth of AI-labelled education makes course counts politically attractive. They are easy to fund, announce, and compare. Yet a catalogue can expand while students encounter the same methods repeatedly, miss prerequisites, or complete projects without understanding how the components connect. AI for All: Strategy for Cultivating Artificial Intelligence Talent makes curriculum design a central implementation question for Korea’s talent strategy.

The alternative is to treat curriculum as infrastructure. Mathematics, statistics, computation, domain reasoning, and decision work must be ordered so that later tasks genuinely depend on earlier capability.

The Catalogue Illusion

A program can advertise mathematics, machine learning, programming, databases, ethics, and a capstone. The list appears complete. Yet students may experience six unrelated courses, each beginning with a new vocabulary and ending before its assumptions matter elsewhere.

Coverage is not coherence.

The second term is the difference between collecting topics and building capability.

Curriculum as a Dependency Graph

A course catalogue usually displays nodes. Curriculum design must inspect the edges.

Prerequisites Are More Than Course Names

A formal prerequisite such as “linear algebra” says little about the required capability.

Does the next course require students to multiply matrices, interpret a projection, recognize rank deficiency, derive a quadratic form, or reason about an eigenspace? These are different prerequisites.

Table 1. Curriculum components and their system dependencies

Prerequisite label	Procedural evidence	Conceptual evidence	Applied evidence
Linear algebra	Computes matrix products	Explains projection and rank	Diagnoses collinearity or latent dimension
Probability	Evaluates distributions	Conditions on information	Reconstructs selection and uncertainty
Calculus	Differentiates an objective	Interprets local sensitivity	Explains optimization and marginal effects
Programming	Writes functioning code	Understands state and abstraction	Builds a reproducible analytical pipeline
Statistics	Applies an estimator	States assumptions and target	Designs validation for the DGP

If the later course needs applied evidence but the prerequisite assessed only procedure, the edge exists on paper and fails in practice.

The Vertical Spine

A coherent AI curriculum needs a small set of ideas that recur at increasing depth.

If a program claims that graduates can make decisions under uncertainty but teaching focuses on derivations and assessment rewards code execution, the outcome exists only in marketing language.

Alignment does not require one teaching method. Lectures, problem sets, cases, laboratories, and research can serve different functions. They must converge on the declared graduate capability.

Horizontal Integration

Vertical sequencing is not enough. Students must connect ideas learned at the same stage.

• probability for uncertainty;
• time-series models for dependence;
• programming for data pipelines;
• economics for structural breaks;
• operations for inventory cost; and
• communication for the recommendation.

Horizontal integration can be designed through a shared case. Each course contributes a different object, but faculty agree on data definitions, timing, and the final decision.

Without that agreement, one course may clean away the anomaly another course is meant to explain. One may define the outcome after the decision time. Another may optimize a metric unrelated to the case's cost.

The Curriculum Matrix

A useful design instrument maps courses against capabilities.

Read as a diagnostic rather than a second scorecard, the framework becomes clear: Foundations — Representation: Introduce; DGP and inference: Introduce; Computation: Introduce; Decision: Expose; Communication: Explain results. Core models — Representation: Develop; DGP and inference: Develop; Computation: Develop; Decision: Compare losses; Communication: Defend assumptions. Advanced methods — Representation: Extend; DGP and inference: Stress-test; Computation: Scale; Decision: Optimize under constraints; Communication: Explain uncertainty. Cases — Representation: Integrate; DGP and inference: Diagnose; Computation: Implement; Decision: Recommend; Communication: Write for a decision owner. Dissertation — Representation: Independently formulate; DGP and inference: Defend; Computation: Reproduce; Decision: Bound the claim; Communication: Sustain an argument.

Blank cells are not always defects. A focused course need not do everything. The program, however, should know where each capability is introduced, developed, integrated, and independently demonstrated.

Avoiding the “Methods Parade”

Students learn that progress means moving to the next named method. They do not learn why one representation belongs to one problem.

• Which restrictions does the model impose?
• What structure is learned from data?
• How much data and computation are required?
• What type of uncertainty can be reported?
• How does the method behave under shift?
• What decision does the additional flexibility change?

The sequence becomes conceptual rather than promotional.

Designing Backwards from the Dissertation

The final independent project provides a useful backward-design test.

If graduates must produce a dissertation containing a precise question, DGP analysis, defensible method, reproducible evidence, and bounded conclusion, earlier courses must create those capabilities in stages.

In a fragmented curriculum, mathematics teaches logarithms, statistics teaches regression, economics teaches production, and programming estimates coefficients. The student may never connect them.

1. What institutional process could justify the multiplicative form?
2. Under what conditions can coefficients be interpreted as elasticities?
3. What happens if inputs are chosen in response to productivity shocks?
4. Does the target concern prediction or production causality?
5. How should uncertainty and heterogeneity be represented?

One elementary relationship becomes a thread across the curriculum.

Managing Cognitive Load

Coherence does not mean exposing every connection at once.

The curriculum should make the temporary simplification explicit. Otherwise students mistake a teaching condition for a universal property of the method.

Curriculum Version Control

Curriculum change should be traceable.

Suppose a program replaces one statistical modelling course with a generative-AI applications course. The visible credit total remains unchanged, but the dependency graph may lose edges needed by later causal analysis, validation, or dissertation work.

• the reason for change;
• evidence motivating it;
• prerequisites added or removed;
• assessments affected;
• transition rules for current students; and
• the date of later review.

This prevents curriculum updates from becoming a sequence of isolated reactions to market attention or individual faculty preference.

A Curriculum-System Test

Review should begin with the final work expected of a graduate and trace its dependencies backwards. If students must defend a model under distribution shift, they need prior experience with probability, measurement, validation, domain assumptions, and written argument. A course title alone cannot establish any of these links.

Korean quality assurance can ask for evidence at the dependency boundaries: whether students entering an advanced course can reconstruct prerequisites, whether assessment integrates material across modules, and whether dissertation work demonstrates independent judgment rather than the assembly of familiar templates.

This approach also clarifies where bridge education is legitimate. Additional preparation is not a dilution of standards when it makes prerequisites explicit and testable. The reform failure occurs when institutions conceal missing foundations behind a larger catalogue of fashionable electives.

Publication of curriculum maps would make these claims inspectable. Students could see where a prerequisite is taught, employers could understand what a program title represents, and reviewers could identify gaps that a list of course descriptions conceals. The map should be treated as a maintained academic system rather than a marketing diagram.

A coherent map also gives reform a time dimension. When a new method enters the program, faculty must decide what it replaces, which prerequisite it depends on, and how its value will be assessed. Without that discipline, every innovation increases breadth while silently reducing depth.

From Mechanism to Evidence

For Adding AI courses is not university reform, the policy mechanism should be read as a chain rather than as a headline target. A government changes funding, rules, information, or institutional incentives; organizations respond; workers and students adjust; and only then do productivity or social outcomes change. Each link can weaken the intended effect. A credible evaluation therefore identifies the behavioral response at every stage instead of treating announced spending or participation as proof of success.

The practical counterfactual also matters. The question is not whether the chosen intervention produces some benefit, but whether it performs better than feasible alternatives using the same money, attention, and administrative capacity. A visible national program may be less valuable than less dramatic reforms to incentives, data access, professional mobility, or institutional accountability. Opportunity cost belongs inside the policy argument, even when it is difficult to photograph or count.

Distribution should be examined separately from the average. A reform can raise aggregate capability while widening gaps across regions, firms, occupations, or educational backgrounds. It can also impose transition costs before longer-term gains appear. Policy design should therefore specify who gains first, who bears adjustment costs, and which bridge allows exposed groups to acquire a more durable role rather than simply absorbing the disruption.

Implementation evidence should be chosen before scale. Useful indicators include changes in actual decisions, transfer to unfamiliar tasks, error detection, retention of capable people, and the speed with which institutions correct failure. Enrollment, procurement, and nominal adoption remain relevant, but they are intermediate measures. They become persuasive only when connected to an observable change in capability or performance.

Finally, the argument should survive comparison across settings. A mechanism observed in Seoul may depend on labor mobility, housing, procurement, university incentives, or firm structure that differs elsewhere. Conversely, an imported policy may fail when those supporting conditions are absent in Korea. Naming the boundary conditions makes the recommendation more useful, not less ambitious, because it reveals what must accompany implementation.

Institutional responsibility must be visible as well. A policy can fail because its theory was wrong, because an agency lacked authority, because providers responded strategically, or because implementation was never completed. These explanations imply different remedies. Publishing ownership, milestones, and correction procedures helps prevent every disappointing result from producing another initiative with a new name but the same unresolved constraint.

Conclusion

A curriculum is a system because education is produced through dependencies and transfer.

The course catalogue identifies the parts. The curriculum determines whether the parts form a capable graduate. That requires explicit prerequisites, a recurring conceptual spine, horizontal cases, aligned assessment, and evidence from later independent work.

Have we offered every important topic?

Can students carry the right structure from one problem to another, recognize when it no longer applies, and rebuild it?

That is the difference between curricular coverage and an AI education.

References

John Biggs, “Enhancing Teaching through Constructive Alignment”, Higher Education 32 (1996): 347-364.
National Academies of Sciences, Engineering, and Medicine, Data Science for Undergraduates: Opportunities and Options, National Academies Press, 2018.
Association for Computing Machinery and IEEE Computer Society, Computing Curricula 2020, 2020.
The Economy (2025) ‘Redesigning Education Beyond Procedure in the Age of AI’, The Economy Review, 17 September.
The Economy Editorial Board (2026) ‘Teacher AI Literacy Is the Real Test of AI in Education’, The Economy Review, 22 June.
Swiss Institute of Artificial Intelligence (2026) ‘Cognitive Outsourcing in Education: Why AI’s Real Classroom Crisis Is Verification, Not Cheating’, SIAI Working Papers, 24 July.
Ministry of Education, Republic of Korea (2025) “AI for All: Strategy for Cultivating Artificial Intelligence Talent”.