Credentials, capabilities, and labor-market signals
Modified
Input
A credential can develop capability, certify achievement, and signal unobserved traits, but these functions should not be treated as identical. AI lowers the cost of producing polished applications and portfolios, increasing the value of assessments that reveal reasoning and individual contribution. Educational institutions preserve the value of their qualifications by making the relationship between the signal and the underlying capability inspectable.
What Does a Credential Communicate?
Employers rarely observe a candidate's productivity before hiring. They see a résumé, degree, portfolio, interview, recommendation, and perhaps a short assessment. Each item is evidence, but none is the capability itself.
This creates an information problem.
Let a worker's true productivity be $\theta_i$, which is not directly observable. The employer observes a signal vector
where $d_i$ is a degree, $g_i$ is an assessment result, $p_i$ is a portfolio, $x_i$ is experience, and $r_i$ is a recommendation.
The employer forms an expectation
Hiring and compensation depend on this conditional expectation, not on perfect knowledge of $\theta_i$.
Three Functions of Education
Education can affect the labor market through at least three channels.
First, it can build human capital. The student becomes more capable:
Second, it can certify demonstrated achievement. An institution observes work that an employer cannot easily inspect and verifies that it met a standard.
Third, it can signal persistent traits. If completing a demanding program is less costly for highly capable or highly disciplined students, completion helps employers infer type.
Table 1. What different labor-market signals can and cannot establish
| Function | Core question | Evidence required |
|---|---|---|
| Capability formation | What can the graduate now do? | Change in performance across time |
| Certification | Which standard did the graduate demonstrably meet? | Valid assessment and verified authorship |
| Signaling | What does completion imply about unobserved traits? | A credible relationship between completion cost and type |
These functions may coexist. Problems begin when an institution relies on the signal while neglecting formation and certification.
A Simple Signaling Model
Suppose workers are either high-productivity, $\theta_H$, or low-productivity, $\theta_L$, with $\theta_H>\theta_L$. Education level $e$ has a cost
Education is costly for everyone, but the marginal cost is lower for the high-productivity type.
If employers pay wage $w(e)$ based on observed education, a separating equilibrium can exist when
while
The high type chooses the more demanding signal; the low type does not find imitation worthwhile.
This model explains how a credential may carry labor-market value even when employers cannot observe every skill taught inside the program. It also identifies the source of fragility: if completion becomes weakly related to capability, the signal loses information.
Selectivity Is Not the Same as Educational Value
A credential can appear predictive because capable people entered the institution, because the institution developed them, or because its assessment separated levels of performance.
Let observed graduate productivity be
where $\theta_i^0$ is capability at entry and $\Delta_i^E$ is value added by education.
A highly selective institution may have a large average $\theta_i^G$ even if $\Delta_i^E$ is modest. A developmental institution may create a larger $\Delta_i^E$ while enrolling students with lower $\theta_i^0$.
The two institutions perform different functions. Ranking them only by graduate wages confounds selection, education, geography, networks, occupation, and employer behavior.
Read as a diagnostic rather than a second scorecard, the framework becomes clear: Graduate salary — Useful information: Market valuation after completion; Major confounder: Entry selection, industry, and location. Completion rate — Useful information: Persistence and institutional support; Major confounder: Standard difficulty and admissions. Employer demand — Useful information: Perceived signal value; Major confounder: Brand familiarity and hiring conventions. Pre/post assessment — Useful information: Capability growth; Major confounder: Test validity and practice effects. Independent project — Useful information: Integrated performance; Major confounder: Resources and outside assistance.
Educational quality requires closer inspection than reputation alone.
The Role of Screening
Employers do not merely read credentials. They screen candidates through interviews, technical tasks, probation, reference checks, and work trials.
Suppose an employer uses a test with score $T$. The score distributions for high- and low-capability candidates overlap:
At threshold $\tau$, the employer faces false positives and false negatives:
Raising $\tau$ reduces one error and increases the other. A good screen therefore depends on the cost of each mistake, the base rate of capability, and the relationship between the test and the work.
Generic puzzle interviews may select test preparation rather than job performance. A case that reproduces the actual decision environment can provide a more relevant signal, though it is usually more expensive to evaluate.
AI Changes the Cost of Signaling
Generative AI can improve prose, code, presentation, and visual polish. This is useful for genuine work, but it changes the information content of traditional signals.
If a portfolio once required $k$ hours of individual production and AI reduces the visible production cost to $\lambda k$, where $0<\lambda<1$, weaker candidates can more easily imitate the surface form of strong work.
The employer's signal-to-noise ratio may fall:
When polished artifacts become common, the differentiating evidence moves toward:
- the reasoning behind the artifact;
- the ability to modify it under new constraints;
- a record of intermediate decisions;
- an oral defense;
- reproducibility; and
- verified work completed under controlled conditions.
AI does not make portfolios useless. It makes provenance and adaptive understanding more important.
Capability-Based Evidence
A robust hiring or educational assessment combines several signals with different failure modes.
Expressed as a practical comparison rather than another table, the distinctions are clear: Timed foundational task — What it can reveal: Independent fluency; Main risk: Anxiety and speed bias. Open case — What it can reveal: Problem formulation and judgment; Main risk: Difficult scoring. Code or model review — What it can reveal: Ability to diagnose existing work; Main risk: Domain dependence. Oral defense — What it can reveal: Ownership and conceptual understanding; Main risk: Interviewer variance. Work sample over several days — What it can reveal: Realistic process; Main risk: Unequal outside resources. Probationary project — What it can reveal: Workplace performance; Main risk: High cost to candidate and employer.
The strongest system does not search for one perfect test. It combines evidence and verifies consistency.
For example, a candidate who submits an excellent forecasting model should be able to explain its data-generating assumptions, reconstruct part of the analysis, respond to a distribution shift, and identify situations in which the model should not be deployed.
Recommendations and Social Information
Recommendations add information because the writer may have observed the candidate over time. Their value depends on the writer's access, candor, and calibration.
A useful recommendation reduces uncertainty about a specific dimension:
A generic letter from a prominent person may add less information than a detailed letter from a direct supervisor. The former can function as a social signal; the latter can document behavior.
Institutions should distinguish the two. Prestige is not a substitute for observation.
What a School Owes the Labor Market
A school cannot guarantee employment, but it is responsible for the integrity of its signal.
That responsibility includes:
- defining the capability associated with the qualification;
- assessing that capability rather than attendance alone;
- verifying authorship and independent performance;
- documenting the level and conditions of assessment;
- revising standards when technology changes the cost of imitation; and
- avoiding claims that the evidence cannot support.
This is an institutional commitment, not a marketing claim.
If the qualification says a graduate can perform independent quantitative work, the assessment architecture should make that proposition defensible.
What Employers Owe Candidates
The information problem is not one-sided. Candidates also know little about the actual work, management quality, data access, and development path behind a vacancy.
Employers signal their own type through:
- specificity of the job description;
- relevance of the hiring task;
- access to prospective colleagues;
- clarity about decision rights;
- treatment of candidate time; and
- willingness to explain how performance will be evaluated.
A labor market works better when both parties provide credible information.
Generative AI makes portfolio evidence both easier to produce and harder to interpret. Code, reports, and visualizations can now appear polished before the candidate understands their assumptions. Employers should respond by examining provenance and transfer rather than abandoning work samples. A short reconstruction task, an explanation of why one model was rejected, or a request to adapt the analysis to a changed cost function reveals more than surface quality. Schools can strengthen the signal by preserving intermediate work and requiring oral defense. The resulting credential says not only that a product was submitted, but that the graduate could explain, revise, and take responsibility for the process that created it. This richer signal is more expensive to produce than a certificate, but it is also harder to imitate. Its value comes from the linked evidence: the submitted artifact, the recorded decisions behind it, and the candidate’s performance when one assumption is changed without warning. Employers gain a signal of transfer, while candidates gain a fairer opportunity to demonstrate capability that a familiar institutional name may not reveal. The cost is additional evaluation time, which should be concentrated on roles where mistaken selection is especially expensive. For senior or high-risk work, that additional evaluation can be far cheaper than discovering a false signal after appointment.
AI weakens some familiar signals while increasing the value of demonstrable process. The Quiet Fraud shows why a polished thesis can no longer be treated as conclusive evidence of independent work. The Training Gap Behind the Rise of SuperHuman Labor emphasizes the growing premium on adaptation, and SIAI’s analysis of the AI labor divide shows that access to technology does not equalize capability automatically. Credentials remain useful, but their signal becomes stronger when supported by reproducible work, defense, and evidence of transfer.
Conclusion
Credentials matter because capability is difficult to observe. They can build skill, certify achievement, and signal traits that employers value. These channels are related but not interchangeable.
AI increases the supply of polished artifacts and lowers the cost of surface imitation. The response should not be nostalgia for a time before AI. It should be better evidence: assessments that reveal reasoning, ownership, adaptation, and reliability.
An educational institution protects its qualification by ensuring that the signal remains attached to a real capability. An employer improves hiring by testing the capability relevant to the task. In both cases, the objective is to reduce uncertainty without confusing prestige, presentation, or compliance with productive judgment.
References
Michael Spence, “Job Market Signaling”, Quarterly Journal of Economics, 1973.
Kenneth J. Arrow, “Higher Education as a Filter”90013-3), Journal of Public Economics, 1973.
David J. Deming, “The Growing Importance of Social Skills in the Labor Market”, Quarterly Journal of Economics, 2017.
The Economy Editorial Board (2026) ‘The Training Gap Behind the Rise of SuperHuman Labor’, The Economy Review, 28 July.
The Economy Editorial Board (2026) ‘The Quiet Fraud: Why AI-Assisted Thesis Fraud Is the New Academic Mirage’, The Economy Review, 10 March.
Swiss Institute of Artificial Intelligence (2026) ‘The AI Labor Divide: Who Wins, Who Survives, and Who Falls Away’, SIAI AI Memo, 13 February.
Similar Post
