The Person Is an End, Not an Objective Function

Every institution chooses what it will count as success. A hospital may count readmissions, a school graduation, a platform engagement, and a public agency cases resolved. Numbers help institutions see patterns and coordinate action. The moral danger begins when a person is treated as one more variable inside the measure, valuable only insofar as changing that variable improves the result. A person is not an objective function. A person is the one for whose sake the institution exists and to whom its power must remain answerable.

This claim is more demanding than saying that data can be biased or that some models are inaccurate. A perfectly accurate measure can still be aimed at the wrong end. A school can accurately predict which students will raise its average and then exclude those who need it most. A hospital can accurately reduce length of stay by discharging patients before support is ready. A platform can accurately increase time spent while making attention less free. The question is not only whether the system meets its target, but whether the target respects the dignity and participation of the people affected.

Ends, Means, and Human Dignity

Catholic social thought begins from the person, created in the image of God and called to communion. The Catechism teaches that the dignity of the human person is rooted in creation and vocation (CCC 1700–1706).Gaudium et Spes §26 teaches that the social order and its development must constantly be related to the good of persons, and §27 rejects treating a person as a means to an end. The language overlaps with Kant’s influential principle that humanity must not be treated merely as a means, but the Catholic account adds creation, grace, embodiment, community, and the common good.

To say that a person is an end does not mean that institutions cannot ask for work, follow rules, allocate scarce goods, or impose proportionate restrictions to prevent harm. It means that these activities require a justification in the good of persons, not merely the convenience of the system. A commuter may be routed to reduce congestion, but must not be sacrificed as an invisible cost without public justification. A patient may be triaged, but not reduced to a likelihood of recovery whose lower value makes care optional. An employee may be evaluated, but not treated as disposable output.

Objective Functions and Their Proxies

In machine learning and operations research, an objective function makes a desired result computable. Because many human goods are difficult to observe directly, institutions use proxies: test scores for learning, clicks for interest, profit for commercial health, complaints for harm, or appointments kept for treatment engagement. Proxies are not inherently deceptive. They are often necessary for action. The problem is that the proxy can drift away from the good, especially when people know it will determine rewards or punishment.

The economic statistician Charles Goodhart is associated with the observation, formulated in his 1975 paper “Problems of Monetary Management: The U.K. Experience,” that a regularity used for control can cease to be reliable once it is made a target. Donald Campbell’s 1976 discussion of the corruption of social indicators makes a related point about high-stakes quantitative measures. These are not universal laws that invalidate metrics. They explain why a measure changes when people and institutions adapt to its incentives.

A school that rewards only test scores may narrow learning to tested content. A call center that rewards short calls may make callers repeat themselves or leave unresolved problems. A police department that rewards recorded stops may increase stops without increasing safety. A charity that reports only people reached may count contacts rather than durable help. In each case, the system may be “working” in its narrow computational sense while failing the persons whose lives supply the data. The person has become fuel for the indicator.

Multi-Objective Systems and Irreducible Goods

Human institutions rarely have one legitimate aim. A hospital seeks health, safety, access, dignity, professional integrity, and fair use of resources. A city seeks mobility, environmental quality, housing, economic opportunity, and the safety of children and older people. A business may owe duties to customers, workers, owners, and communities. A single score can be convenient, but convenience does not turn conflicting goods into one commensurable quantity.

Multi-objective optimization can represent several goals through constraints, Pareto frontiers, weighted objectives, or lexicographic priorities. These techniques are useful, but they do not remove moral judgment. A weight assigned to dignity, privacy, or participation is a governance choice, not a fact discovered by the algorithm. A constraint can protect a right, but someone must determine which constraint is non-negotiable and how conflicts are resolved. If the system silently collapses everything into revenue or engagement, one interest governs without being named.

Amartya Sen’s capability approach, developed in economics and political philosophy, is relevant here because it evaluates whether people have real opportunities to achieve valuable ways of living, not only the resources or utilities recorded by a system. Sen is not a Catholic magisterial authority, and his framework is not identical to Catholic social teaching. It nevertheless helps make a shared distinction: outcomes cannot be judged only by aggregate output when people’s agency, health, voice, and ability to participate have been suppressed.

Institutional Incentives and the Disappearing Person

An institution does not need malicious designers to produce dehumanizing results. Incentives narrow attention. A manager judged on quarterly throughput may pass costs to a contractor. A public office judged on clearance rates may decline difficult cases. A social service measured on processing time may discourage conversation with a person whose situation is complex. Each actor can follow the rule and still create a structure in which no one is responsible for the whole person.

Catholic social teaching’s account of the common good is an alternative to both isolated individualism and total system control. The common good includes the conditions that allow people and communities to flourish (Gaudium et Spes §26; CCC 1905–1912). Solidarity insists that burdens borne by the vulnerable count as a claim on everyone, while subsidiarity asks that decisions be made at the appropriate level rather than absorbed by a distant authority (CCC 1883–1885). These principles do not yield a numerical formula. They require institutions to ask who has power, who bears risk, and whether those affected can act together rather than merely being processed.

The principle of participation follows. Mater et Magistraconnects social life with responsible participation, and the Compendium of the Social Doctrine of the Church treats participation as a requirement of human dignity, not a decorative consultation. A design process that invites users to select colors after deciding the objective is not meaningful participation. People must be able to question the purpose, evidence, tradeoffs, and authority of the system. Workers who are measured should have a voice in what is measured and a route to challenge unsafe or unjust consequences.

Design Constraints That Protect the End

Make the purpose and authority explicit

Every consequential system should name its legitimate purpose, owner, affected groups, decision boundary, and prohibited uses. “Improve efficiency” is not enough. Efficiency of what, for whom, and at whose cost? A system built to identify patients needing outreach should not quietly become a reason to deny them service. A workforce forecast should not become a permanent character judgment. Purpose limitation, familiar from data-protection practice, is also a moral discipline: it prevents a record collected for one good from acquiring unreviewed power over a life.

Use constraints, not only scores

Rights and dignity should not be represented solely as weights in a utility calculation. Require hard constraints where appropriate: non-discrimination, privacy, accessibility, safety, due process, and meaningful human review. The United Nations Guiding Principles on Business and Human Rights frame corporate responsibility in relation to human rights, while the OECD AI Principles call for inclusive growth, human-centered values, transparency, robustness, and accountability. These are not substitutes for local law or Catholic doctrine, and they do not solve every conflict. They illustrate a design approach in which some values limit what may be optimized.

Build contestability and participation

Affected people need notice, reasons proportionate to the decision, access to relevant evidence, correction of factual errors, and a human channel with authority to change the result. Appeals cannot be merely a second interface that routes the same decision back to the same score. The institution should record who reviewed the case, what discretion was available, and whether patterns of appeal reveal a flawed objective. Participation also means testing the system with the people most likely to be excluded, compensating their expertise where appropriate, and continuing consultation after deployment.

Audit the incentive, not just the model

A technically fair model can be used in an unfair institution. Evaluation should therefore include downstream effects: who receives an opportunity, who is burdened with verification, who changes behavior to satisfy the metric, and what unmeasured good disappears. Monitor distributional outcomes over time, publish material limitations, and give the system an expiration or review date. Independent audits should have access to enough information to test claims, while respecting privacy and security. If an objective repeatedly produces harmful gaming, revising the objective is more responsible than blaming users for adapting to it.

Strongest Objections

The strongest objection says that institutions cannot function without objectives. Budgets must close, emergencies require prioritization, and public officials need evidence rather than intuition. Replacing metrics with personal judgment can reintroduce favoritism and make decisions impossible to compare. This objection is correct. The person-as-end principle does not prohibit measurement. It asks measurement to remain a means, disclose its limits, and answer to a wider account of the good. Transparent indicators plus review are usually better than unexamined discretion, but neither can escape moral responsibility.

A second objection says that adding constraints sacrifices efficiency and therefore harms the very people the institution serves. A hospital with too many procedural safeguards may treat fewer patients; a small employer may not be able to fund elaborate appeals. Tradeoffs are real, and safeguards have costs. Yet “efficiency” is incomplete when it omits wrongful exclusion, unrecognized labor, privacy loss, or damage to trust. Proportionality matters: a low-stakes recommendation may need fewer protections than a decision about liberty, housing, health, or livelihood. The answer is graduated governance, not unrestricted optimization.

A third objection says that a multi-objective approach merely hides choices behind committees and produces paralysis. There is a risk of vague principles becoming impossible to apply. This is why values should be translated into concrete process requirements, thresholds, review duties, and named authorities. A system can publish which goals are prioritized, which rights are constraints, what evidence is used, and how conflicts are resolved. Clarity about judgment is preferable to the fiction that a single number has made judgment disappear.

Finally, someone may argue that treating persons as ends is merely Kantian language and need not govern secular design. The principle can be defended in several traditions, but its Catholic basis does not depend on universal acceptance of Kant. The person’s dignity is grounded in being created and called by God, not in market value, productivity, or institutional usefulness. In a plural society, that claim can coexist with human-rights language and democratic accountability while offering a deeper reason to refuse systems that make the vulnerable expendable.

A worked boundary: hospital discharge

Consider a discharge model that seeks to free beds while preserving safety. A lawful or moral floor is that the patient receives accessible information, medication continuity, a chance to correct the record, and a safe route to appeal; these are not weights to trade away for throughput. Within those constraints, the institution may optimize expected clinical benefit, available follow-up, and fair access to beds. Residual discretion belongs to a named clinician who must record why the case falls outside the model and who can pause discharge when the patient’s account supplies missing context. Publish the tradeoff and compare readmissions, delayed discharges, disparate burden, reversals, and review cost against a transparent model-only baseline. The essay does not have evidence to claim that hybrid review always wins; its claim is that this comparison makes the moral choice contestable rather than hiding it inside one score.

Practical Implications

Boards and executives should approve objectives as ethical commitments, not leave them to engineering teams alone. Require an impact assessment that describes affected persons, alternatives, distributional effects, foreseeable gaming, and an exit plan. Tie executive incentives to safety, worker conditions, customer outcomes, and repair of harms rather than one growth metric. A system that cannot explain what would count as failure has not been responsibly governed.

Designers should resist the pressure to turn every value into a scalar. Keep a list of non-negotiable constraints; distinguish recommendations from decisions; show uncertainty; and make refusal, correction, and escalation possible. Build accessibility and language inclusion from the beginning. For high-impact uses, use representative testing and post-deployment monitoring. Document changes to objectives and data, so that an institution cannot quietly redefine success after harm appears.

Workers, citizens, and users should have collective as well as individual channels. A person who appeals alone may be labeled difficult; a union, patient council, school board, parish, or community organization can identify a structural pattern. Rerum Novarum and later social teaching affirm associations and worker organization as protections against isolated dependence. Participation is strongest when people can deliberate together about purposes and power, not merely submit feedback to a finished system.

Sources and Qualifications

The Catholic framework draws on the Catechism of the Catholic Church, especially CCC 1700–1706, 1881–1885, and 1905–1912; Gaudium et Spes §§12, 26–27, and 63; Dignitatis Humanae §3; Leo XIII’s Rerum Novarum; John XXIII’s Mater et Magistra§§65–67; and the Compendium of the Social Doctrine of the Church on dignity, participation, solidarity, and subsidiarity. These sources do not provide an algorithm or imply that every institutional decision can satisfy every person. They set moral limits and orient institutions toward the common good.

The technical and economic discussion refers to Charles Goodhart, “Problems of Monetary Management: The U.K. Experience” (1975), Donald Campbell’s 1976 discussion of social indicators, and Amartya Sen’s capability approach. Governance references include the OECD AI Principles, the United Nations Guiding Principles on Business and Human Rights, and established practices in multi-objective optimization and data protection. These authorities address different questions. Legal duties vary by jurisdiction; this essay is ethical and architectural guidance, not legal, compliance, or mathematical advice.

Synthesis: Design for Persons, Not Merely Results

An objective function can organize a computation, but it cannot bear dignity, responsibility, or the pain of being excluded. Proxies are useful signs, yet they become dangerous when a target replaces the good it was meant to indicate. Goodhart-like failures are not only technical bugs. They reveal institutions that reward what is easy to count while shifting costs onto people with the least power.

To treat the person as an end is therefore a design constraint and a governance practice. Name the purpose, preserve plural goods, protect rights, involve affected communities, audit incentives, and keep a path for correction. Let systems support judgment without making judgment disappear. Catholic social teaching supplies the horizon: technology, markets, and institutions are ordered to the person and the common good, not the other way around. The best system is not the one that extracts the highest score from human lives. It is the one whose power remains answerable to the people it is meant to serve.