Universities as productivity infrastructure
Modified
Input
Universities contribute to productivity by developing capability, preserving and extending knowledge, and connecting people to professional communities. Expanding enrollment or funding does not guarantee these outputs; incentives, assessment, faculty time, and knowledge access determine the conversion rate. In an AI-intensive economy, university reform should focus on learning and knowledge diffusion rather than treating degrees as the final product.
More Than a Provider of Degrees
A university is often evaluated through inputs and transactions: admissions, enrollment, tuition, research funding, facilities, publications, and degrees awarded. These quantities are visible. The university's economic contribution is less visible.
The institution can perform at least three productive functions:
- develop human capability;
- preserve, test, and extend knowledge; and
- connect learners and researchers to communities in which knowledge can circulate.
These functions make a university part of a society's productivity infrastructure. A university that issues credentials without performing them may remain administratively active while contributing little to long-run capability.
An Educational Production Function
Let student capability at graduation be
where $H_{i,0}$ is entering capability, $T_i$ is effective study time, $E_i$ is instructional quality, $P_i$ is practice with feedback, $Q_i$ is assessment quality, and $N_i$ is access to knowledge networks.
A multiplicative specification makes the complementarities explicit:
Excellent lectures with no deliberate practice may create recognition rather than capability. Intensive practice with weak feedback can reinforce mistakes. Strong assessment at the end cannot repair a missing sequence.
The institution's output is the change $\Delta H_i$, not the fact of enrollment.
Selection and Value Added
Universities compete partly through the capability of the students they admit. This creates a measurement problem.
Average graduate performance can be decomposed as
A university with highly prepared entrants can produce strong graduates even when its value added is modest. Another university may generate substantial growth from a less prepared cohort while reporting lower final outcomes.
Table 1. University functions that contribute to long-run productivity
| Institutional indicator | What it captures | What it may hide |
|---|---|---|
| Admission selectivity | Demand and entry characteristics | Educational value added |
| Graduation rate | Completion and support | Standard difficulty |
| Research volume | Knowledge production activity | Quality, relevance, and diffusion |
| Graduate income | Market outcomes | Prior ability, field, location, and networks |
| Student satisfaction | Perceived experience | Durable learning |
| Employer reputation | Signal acceptance | Actual capability distribution |
No single indicator resolves the problem. A serious evaluation must examine both entry conditions and learning.
Knowledge Production and Diffusion
Universities do not only teach existing material. They produce and organize knowledge.
Let the stock of useful knowledge be $I_t$. It evolves according to
where $\delta$ is depreciation or obsolescence, $R_t$ is new research, $D_t$ is imported and adapted external knowledge, and $\lambda$ is the institution's capacity to absorb it.
This equation gives diffusion an important role. A university can improve its intellectual capital not only by producing original research but also by understanding, testing, teaching, and adapting work produced elsewhere.
In AI, where methods and tools change quickly, a closed curriculum experiences rapid depreciation. Faculty and students need direct access to research communities, technical documentation, reproducible work, and criticism.
Faculty Time as a Scarce Input
Faculty work is usually divided among teaching, research, administration, advising, and external engagement.
Let a faculty member allocate time
Institutional output is
where teaching quality may depend partly on active research. The institution should not maximize one component mechanically. Excessive administration can reduce both teaching and research; research disconnected from teaching may contribute little to student capability; teaching overload can prevent knowledge renewal.
The optimal allocation depends on mission, but the trade-off cannot be avoided.
Read as a diagnostic rather than a second scorecard, the framework becomes clear: Teaching — Necessary contribution: Explanation, feedback, assessment; Risk when excessive: Knowledge stagnation and faculty burnout. Research — Necessary contribution: New knowledge and intellectual renewal; Risk when excessive: Weak attention to student development. Administration — Necessary contribution: Coordination and accountability; Risk when excessive: Compliance replacing academic work. External engagement — Necessary contribution: Cases, data, and professional relevance; Risk when excessive: Consulting or promotion displacing scholarship.
Reform must address the allocation system, not only individual effort.
Why More Funding May Not Raise Productivity
Funding is necessary for laboratories, faculty, scholarships, data, libraries, and computational resources. Its effect depends on how it enters the production function.
Let university output be
where $H_F$ is faculty and staff capability, $I_U$ is intellectual capital, $P_U$ is physical and financial capital, and $A_U$ is institutional effectiveness.
If additional funding raises $P_U$ while $A_U$ remains weak, the marginal return can be low. New facilities do not automatically change curriculum coherence, feedback quality, hiring incentives, or research standards.
The relevant policy question is therefore not simply how much a university receives. It is which constraint the expenditure relaxes.
The Importance of Assessment
Assessment converts an educational aspiration into a standard.
If students receive a credential after satisfying attendance and submission requirements, the institution may measure compliance rather than capability. A valid assessment asks students to perform the intellectual work the qualification claims.
For AI and data science, that may require:
- reconstructing a model rather than repeating a definition;
- identifying the data-generating process;
- comparing representations;
- defending assumptions;
- reproducing computation;
- diagnosing failure; and
- revising the work after criticism.
Assessment also provides information to the institution. Let the observed performance gap be
where $q_k^{*}$ is the required level on capability $k$ and $\overline{q}_k$ is cohort performance. Persistent gaps should alter prerequisites, teaching, practice, or the standard itself.
Without this feedback loop, curriculum reform becomes a change of course titles.
Universities and Externalities
The return to education is not entirely private. Capable people can raise the productivity of colleagues, create firms, improve public decisions, and transmit knowledge.
Suppose regional productivity is
The term $\overline{H}_r^\phi$ represents a human-capital externality: the average capability of the surrounding population raises the productivity of firms and workers.
This is one reason societies support universities. It is also why low-quality expansion can be costly. Resources devoted to nominal attainment may generate neither private learning nor public spillovers.
What Reform Should Target
University reform is often reduced to mergers, rankings, governance charts, or funding formulas. These may matter, but they are instruments.
A productivity-centered reform agenda asks:
Expressed as a practical comparison rather than another table, the distinctions are clear: Capability formation — Operational question: What can students do at completion that they could not do at entry? Knowledge renewal — Operational question: How does current research enter teaching? Faculty incentives — Operational question: Which activities receive time, resources, and promotion? Assessment integrity — Operational question: Does the qualification verify independent performance? Knowledge access — Operational question: Can students and faculty participate in global scholarly communities? Diffusion — Operational question: How do research and graduates improve organizations beyond the campus?
The objective is not to make every university identical. It is to make the relationship among mission, resources, standards, and outcomes explicit.
AI as Both Tool and Test
AI can lower the cost of tutoring, drafting, coding, translation, and formative feedback. Used well, it can expand practice and make hidden misunderstandings visible.
It can also allow students to submit work they cannot explain. The same technology that raises access can weaken traditional evidence of authorship.
Universities therefore need two responses:
- integrate AI where it improves learning; and
- redesign assessment so that independent capability remains observable.
A student may use AI to explore alternative models, but should still be able to specify the estimand, defend the data, reproduce critical steps, and explain errors. Prohibition alone ignores the productive tool. Unverified use abandons the educational signal.
The Specialized Institution
A specialized institution can concentrate scarce faculty and curriculum design around a defined capability. Its advantage is not automatic. Small scale may limit breadth, resources, and redundancy.
The value arises when specialization improves $A_U$: shared vocabulary, coherent sequencing, integrated assessment, and rapid knowledge renewal.
This is particularly relevant for AI/Data Science, which draws from mathematics, statistics, computing, economics, and domain practice. A fragmented institution can offer all of these subjects without connecting them. A specialized institution must make the connection its central obligation.
This infrastructure has a longer time horizon than most firm-level training. Companies rationally invest in capabilities they expect to use soon, while universities can maintain foundational knowledge whose commercial application is uncertain or distributed across many organizations. They also create shared standards through assessment, publication, and professional communities. These externalities explain why the return to a university cannot be read solely from graduate starting salaries. The institution contributes when it improves the quality of local problem solving, allows knowledge to move between sectors, and preserves the capacity to evaluate new methods before their value is obvious to any single employer. Because these benefits spill across employers and years, underinvestment is predictable. Public and institutional support is justified when it funds transferable knowledge and open standards, not when it merely subsidizes private credentials whose returns remain concentrated with the holder. Evaluation should therefore combine graduate outcomes with evidence of research diffusion, professional standards, and organizational learning beyond the university itself. A productive university leaves capabilities in the surrounding economy that no single graduate or employer can fully appropriate. That spillover is part of its return.
The infrastructure role of universities is broader than degree production. Redesigning Education Beyond Procedure in the Age of AI argues that institutions must redirect teaching toward model formulation and defense. SIAI’s work on lifelong AI learning extends the mandate beyond initial graduation, while Invisible Value places knowledge and organizational assets inside the productive capital stock. Universities raise productivity when they create, test, and diffuse these capabilities across firms and professions—not when they merely increase credential counts.
Conclusion
Universities are productivity infrastructure when they develop capability, maintain knowledge systems, and connect people to communities of inquiry and practice.
Enrollment, funding, buildings, publications, and credentials are inputs or indicators. They become valuable when the educational production function converts them into learning and diffusion.
In an AI-intensive economy, the institutional standard should become more demanding: graduates must be able to use powerful tools without surrendering judgment, and universities must be able to demonstrate how their design produces that capability.
University reform is therefore not principally a campaign for more or less education. It is the disciplined reconstruction of how education becomes productive human and intellectual capital.
References
Eric A. Hanushek and Ludger Woessmann, The Knowledge Capital of Nations: Education and the Economics of Growth, MIT Press, 2015.
Robert E. Lucas Jr., “On the Mechanics of Economic Development”90168-7), Journal of Monetary Economics, 1988.
James J. Heckman, “Skill Formation and the Economics of Investing in Disadvantaged Children”, Science, 2006.
The Economy (2025) ‘Redesigning Education Beyond Procedure in the Age of AI’, The Economy Review, 17 September.
Swiss Institute of Artificial Intelligence (2026) ‘AI Lifelong Learning Must Arrive Before the Skill Gap Hardens’, SIAI Science Review, 25 July.
Swiss Institute of Artificial Intelligence (2026) ‘Invisible Value: Why AI Intangible Assets Make the Economy Look Smaller Than It Is’, SIAI AI Memo, 22 January.
Similar Post
