Was den Menschen zum Menschen macht

AI Ethics

10 Min. Lesezeit

🇩🇪 Deutsche Version: KI-Ethik

The ethical evaluation of Artificial Intelligence from the perspective of personal ontology poses a fundamental question: How is one to deal with a technology that can simulate person-behavior without itself being a person?

The Personal-Ontological Perspective

The personalist norm — that the person is to be affirmed for its own sake — holds for persons, not for machines. An AI ethics grounded in the ontology of personhood draws a clear distinction:

  • The person is someone — an AI is something. This distinction is not gradual but fundamental.
  • Dignity belongs to the person because it is a person — not because it exercises certain functions. An AI has no dignity, even when it appears “intelligent.”
  • Responsibility presupposes freedom and self-consciousness. An AI cannot act responsibly — responsibility lies with the persons who develop and deploy it.

Dangers of the Oblivion of the Person

The greatest danger in dealing with AI is a new form of the oblivion of the person:

  1. Humanization of the machine: When AI systems are ascribed personhood, the concept of person is devalued.
  2. Mechanization of the human being: When the human being is understood as a mere information-processing system, its spiritual being is denied.
  3. Instrumentalization: When AI is used to manipulate, surveil, or replace persons, the personalist norm is violated.

Guidelines from Personal Ontology

An AI ethics oriented to the reality of the person demands:

  • AI must not infringe upon the dignity of the person
  • AI decisions about persons require personal responsibility
  • The distinction between person and machine must be preserved
  • AI should serve the unfolding of personhood, not replace it
  • AI is not capable of truth and cannot render ethical judgments — it has no conscience

Why AI Cannot of Itself Render Genuine Ethical Decisions

Large language models and other AI systems are statistical-probabilistic models. When an AI “answers” an ethical question, it reproduces the statistically most probable answer that follows from its training data — a weighted average of human opinions, not a reasoned judgment. This is not ethical competence but an echo.

Genuine ethical decisions presuppose what personal ontology identifies as the essential characteristics of the person: reason, which weighs grounds; conscience, which distinguishes between good and evil; freedom, which can decide for the good; and responsibility, which answers for one’s own action before oneself and others.

An AI possesses none of these essential characteristics. It has no interiority, no self-consciousness, no free will — and therefore no conscience either. What looks like an “ethical judgment” of the AI is deutera energeia without prote energeia. It is the simulation of an action without a self that acts — a statistical ethics simulation.

This deficiency shows itself especially where ethical dilemmas arise. The impossibility theorem (Chouldechova 2017) proves mathematically that not all statistical fairness criteria can be satisfied simultaneously. When one criterion is optimized, another is violated. A statistical system can recognize that a bias is present. But it cannot decide which fairness criterion takes precedence. For that it requires a normative justification. And normative justifications presuppose an ontological foundation that goes beyond statistics.

The Growing Urgency of a Grounded AI Ethics

The question of an ontologically founded AI ethics is not academic but of immediate practical relevance. Autonomous and semi-autonomous AI systems already render decisions today that directly concern persons — in ever more domains of life:

  • Autonomous driving: A vehicle must decide in fractions of a second how to react in a hazardous situation. Such decisions potentially concern the life of persons — and are currently made on the basis of statistical optimization, not on the basis of an ethics that respects the dignity of every affected person.
  • Medicine: AI systems support diagnoses, prioritize patients, and recommend treatments. Where an algorithm decides who is treated first, the personalist norm is at stake: the person must not be reduced to a data point.
  • Human resources (HR): Algorithms filter applications, assess performance, and influence promotion decisions. An algorithmic assessment that is not measured against the dignity of the person threatens to become instrumentalization.
  • Justice and administration: Predictive policing, risk assessments in criminal law, automated benefit decisions — everywhere that algorithms judge over persons without a conscience at work.

In all these domains it holds: a statistical system can recognize patterns, but it cannot justify why a decision is morally right or wrong. It cannot answer the “why” question that the EU AI Act implicitly poses with its demands for “human-centric AI” (Art. 1) and the protection of “fundamental rights” — but does not itself formally answer.

An AI ethics that goes beyond statistical fairness metrics and can ground ethical decisions ontologically therefore becomes more urgent with each new application. The more autonomously AI systems operate, the less it suffices to leave responsibility with the human being alone, without giving the machine itself a normative orientation. This orientation is no substitute for the human conscience, but a formalized justificatory structure by which the machine orients itself.

From a non-personal-ontological side, Luciano Floridi has formulated the most influential contemporary counter-proposal with his Information Ethics: he seeks to derive moral status not from personhood but from the informational being of entities, and ascribes a graded intrinsic worth (entropic harm) even to non-personal informational entities. Personal ontology contradicts this extension: what is due to the someone cannot accrue to the something. Floridi’s approach nonetheless remains a serious point of reference, because it articulates the need for a normative structure addressed to the AI itself.

AI and Art — from the Ontological to the Ethical Question

The question whether AI can create art is at first not an ethical but an ontological one: What is art, and what must a being be in order to bring it forth? Personal ontology gives a clear answer: what AI produces is deutera energeia (the exercise of activities) without prote energeia (substantial being). It is output without a heart that has been touched by values. Without an individuality that expresses itself. Without a self that gives itself over in self-transcendence. AI simulates the form of art, not its origin.

This ontological insight becomes ethical where it is socially ignored. When AI-generated images, texts, and music can no longer be distinguished from human art — and when this distinction is no longer held to be important — then a new form of the oblivion of the person is taking place. Personhood is reduced to observable functions. This is precisely the empirical-functionalist fallacy that personal ontology rejects.

Beyond this, AI-generated art raises concrete ethical questions: the instrumentalization of artists whose works serve as training data without consent. The deception when AI output is passed off as a human work. And the devaluation of personal creative achievement in a society that increasingly levels the difference between someone and something.

Algorithmic Assessment

An algorithmic assessment is an ascription about a person generated by an artificial agent. This assessment is temporally limited and not essence-determining. The person is not its assessment.

This is decisive in personal-ontological terms: the dignity of the person is grounded in its being, not in an algorithmically generated metric. Algorithmic assessments must therefore always be measured against the personalist norm within an ethical review.

Ontological classification:

  • Subclass of: Temporal Attribution

Automated Decision Process

An automated decision process is a process in which an artificial agent makes or prepares decisions that concern persons. It must be measured against the personalist norm. The person has the right not to be treated exclusively on the basis of automated assessment.

No decision about persons may be made without ultimate personal responsibility. Every automated decision process is therefore subject to the control of a responsible person. This protects the dignity of the person against reduction to data points and algorithms.

Ontological classification:

  • Subclass of: Process
  • Relations: is assessed by (person to process), is subject to human control (process to person)

The free, enlightened assent of a person to a medical intervention. Informed consent (informed consent) is an expression of respect for the free will and the rationality of the person — essential characteristics of personhood (cf. Bexten 2017, pp. 322 ff.).

From a personal-ontological perspective, the demand for informed consent is grounded in the dignity of the person: the person must not be made a mere object of medical action, but must be respected as a subject — as a someone with its own interiority, freedom, and cognition. The personalist norm forbids using the person merely as a means; informed consent is the practical consequence of this prohibition in the medical context.

Two elements are constitutive of a valid informed consent. First, the capacity for truth: the person must be able to understand the relevant information and must be informed truthfully. Second, freedom: the assent must occur without coercion, deception, or manipulation. Where one of these elements is lacking — for instance in an embryo, an unconscious patient, or under coercion — no valid consent can be present. Special protective measures are then required.

Informed consent is to be classified as a personal act: it is a conscious, free act of the person in which the person disposes over itself. At the same time, informed consent shows the limits of self-disposal: the person cannot consent to everything. To an intrinsically evil act — for instance to one’s own killing (euthanasia) — the person cannot give morally valid consent, because the dignity of the person is inviolable, even for the person itself.

In the context of reproductive medicine, the question of informed consent arises with particular sharpness: in artificial fertilization or surrogacy, the affected child — the person whose existence is at stake — can give no consent.

Ontological classification:

Chapter assignment: Chapter 5: Oblivion of the Person

Sources: Generated by querying the Personhood ontology.

Further sources:

  • Turing, Alan (1950): “Computing Machinery and Intelligence.” In: Mind 59, pp. 433–460 (behavioral equivalence as a criterion — rejected by personal ontology).
  • Spaemann, Robert: Persons: The Difference between ‘Someone’ and ‘Something’, transl. Oliver O’Donovan. Oxford: Oxford University Press, 2006 (the someone/something distinction as the basis of AI ethics).
  • Wojtyła, Karol: Love and Responsibility, transl. H. T. Willetts (the personalist norm).
  • Singer, Peter (1979/1993): Practical Ethics. Cambridge: Cambridge University Press (defends the functionalist concept of person — critically discussed in the dissertation).
  • Kant, Immanuel (1785): Groundwork of the Metaphysics of Morals. Akademie edition, vol. IV, p. 429 (the formula of humanity as an end in itself: the person never merely as a means).
  • Chouldechova, Alexandra (2017): “Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments.” Big Data 5(2), pp. 153–163 (the impossibility theorem of statistical fairness).
  • European Union (2024): Regulation on Artificial Intelligence (AI Act). Regulation (EU) 2024/1689.

See also

Graphansicht

Backlinks