AI and Human Dignity

What "dignity" is doing in moral thought

Before surveying particular accounts, it is worth being clear about the work the concept of dignity does in moral philosophy.

Dignity is the answer to a particular question: what is it that makes human beings morally important in the first place? The question is foundational. Without an answer, all subsequent moral claims hang in the air. We say that people have rights, that institutions should serve people, that certain kinds of treatment are wrong, that certain protections are required. All of these claims presuppose that human beings are the kind of thing whose treatment matters in a distinctive way. Dignity is the term for that distinctive moral importance.

Different philosophical traditions answer the underlying question differently, and the differences in their answers cash out in differences in what they think dignity requires. But the role of the concept is shared: dignity is the ground from which moral and political obligations to persons are derived.

Three structural features of dignity are common to all the major accounts:

Inherence. Dignity is something persons have, not something they earn. It is not a reward for good behavior or successful achievement. It is intrinsic to being a person.

Universality. Dignity belongs to all persons, not only to some. The accounts differ on what counts as a person, but within the boundary they draw, dignity is universal.

Inviolability. Dignity cannot be legitimately removed or canceled by any human authority. It can be respected or violated, but a violation does not extinguish it. Even the worst criminal retains dignity, even when their other rights have been forfeited.

These three features are the spine of every serious dignity account. They are what distinguish dignity from related concepts like merit, status, or worth-by-achievement. With those features in mind, we can turn to the specific accounts.

The Kantian account

Immanuel Kant's account of dignity is the most influential secular account in modern moral philosophy. It runs through European Enlightenment thought, into the human rights tradition, into the structure of liberal democratic constitutions, and into contemporary AI ethics. If you read an AI ethics paper that grounds its claims on dignity without specifying further, it is usually Kantian dignity in the background.

Kant grounded dignity in rational autonomy: the capacity to give oneself moral law and to act on principle rather than mere impulse. On Kant's view, rational beings are ends in themselves, never mere means to other ends. The famous formulation of the categorical imperative captures this: "Act so as to treat humanity, whether in your own person or in the person of any other, always at the same time as an end and never merely as a means." The wrong of treating a person as a mere means is, in Kantian terms, a failure to respect their rationality.

The strengths of the Kantian account are real. It is rigorous, it is non-religious, it is universally accessible to anyone capable of philosophical reflection, and it generates strong protections against using persons instrumentally. The Universal Declaration of Human Rights and most twentieth-century human rights documents are Kantian in their underlying structure, even when they don't credit Kant explicitly.

The Kantian account also has weaknesses that become especially visible in the AI age. The most important is that dignity is tied to rational capacity. Beings that are rational have dignity; beings that are not rational, or not rational enough, are in a different category. This creates two problems.

First, it creates pressure to extend dignity to non-human rational agents if any are discovered. If AI exhibits the kind of rational autonomy that grounds dignity, the Kantian framework cannot in principle exclude AI from moral consideration. Some philosophers have followed this logic and argued that sufficiently advanced AI would have to be granted dignity. The Kantian account does not give us a clear way to say no.

Second, it creates pressure to deny dignity to humans whose rational capacities are limited. Infants are not yet rational in the relevant sense. People with severe cognitive disabilities may never be rational in the Kantian sense. People in vegetative states are not currently rational. People with advanced dementia are losing the relevant capacities. On a strict Kantian account, the dignity of these humans is at best derivative, grounded in their potential or former rationality rather than their current state. Kant himself was clear that animals lack dignity because they lack rationality. The logic is uncomfortable when extended to non-paradigm humans.

Sophisticated Kantians have ways to address these problems (extended notions of rationality, dignity as belonging to the human species rather than individual humans, moral standing through relational connections). But the basic structural feature, that dignity tracks capacity, remains. This is the feature that becomes most strained as AI capacities advance.

The capability approach

The capability approach, developed by Martha Nussbaum and Amartya Sen and increasingly influential in international development and AI policy, takes a different starting point. Rather than grounding dignity in a single feature (rationality, ensoulment, etc.), the capability approach focuses on what human beings need in order to live a fully human life. Dignity, on this view, is realized in the conditions that enable persons to develop and exercise their distinctive capabilities: bodily health, bodily integrity, senses, imagination, thought, emotions, practical reason, affiliation, relationships with other species, play, control over one's environment.

The capability approach has been influential in shaping how international bodies think about human flourishing, including in the AI context. UNESCO's Recommendation on the Ethics of AI draws heavily on capabilities-style thinking. The EU AI Act's risk-based framework can be read as protecting capabilities from AI-driven erosion.

The strength of the capability approach is its concreteness. It directs attention to specific dimensions of human life and asks how technologies, institutions, and policies affect each. It avoids the abstraction of more metaphysical accounts and produces actionable evaluation criteria. For AI ethics, the question becomes: which human capabilities does this AI system serve, and which does it diminish?

The weakness is that the capability approach is less clear on what grounds the list of capabilities itself. Why these capabilities and not others? Different versions of the capability approach answer differently, and the question is harder than it sounds. Without a deeper account of what makes capabilities central, the approach can drift toward consensus-list-of-the-moment rather than principled moral commitment. The capabilities are also vulnerable to redefinition: if AI expands what human beings can do, do we now have more capabilities? Fewer? Different ones?

The capability approach is most powerful when it is paired with a deeper account of dignity that grounds the list of capabilities. Without such grounding, it gives us a useful checklist but not the foundation of one.

The human rights tradition

The Universal Declaration of Human Rights opens with: "All human beings are born free and equal in dignity and rights." The Declaration places dignity at the foundation of modern international human rights law, but it is studiously non-committal about what grounds dignity. The drafters in 1948 deliberately avoided specifying a metaphysical or religious foundation, because the document had to be acceptable to representatives from many traditions.

The result is a foundational concept that does enormous practical work without any clear philosophical grounding. The human rights tradition has been spectacularly effective: it has shaped law, policy, and moral discourse for nearly eighty years. Most contemporary AI governance documents (the OECD principles, the EU AI Act, UNESCO's Recommendation) draw on the human rights framework.

The weakness is the same as the strength. The non-committal grounding means that the framework is vulnerable to drift. If different parties interpret dignity differently, the framework can mean different things to different people while using the same words. This is fine when the parties broadly agree in their underlying commitments. It becomes a problem when they don't, or when new questions arise (like AI) where the underlying commitments matter for what dignity actually requires.

The human rights tradition is best understood as a practical political achievement built on top of deeper philosophical and religious traditions, not as a complete moral framework of its own. The OECD's AI Principles, when they talk about dignity, are doing the work of a tradition without naming the tradition. This works most of the time. It strains in the cases AI is now producing.

The Catholic account

Catholic teaching grounds dignity in a particular metaphysical and theological claim: every human being is made in the image and likeness of God (imago Dei). The claim comes from Genesis 1:27, has been developed across the Catholic tradition, and is given systematic expression in modern Catholic Social Teaching. The Second Vatican Council's Gaudium et Spes develops the modern Catholic account most fully, particularly paragraphs 12-22.

What does imago Dei actually claim? Several things, none of which reduce to the others.

It claims that human beings are ordered toward God. Their nature is to be in relationship with their Creator, and this orientation belongs to their being, not their choice.

It claims that human beings have a rational nature, but it does not reduce dignity to current rational performance. The rational nature is what kind of being a human is, not what cognitive level a particular human happens to exhibit. Aquinas distinguishes between the rational nature of human beings (which is a metaphysical claim about kind) and the exercise of rationality (which is variable across individuals and times). Dignity tracks the first, not the second.

It claims that human beings are persons in a particular and irreducible sense. Personhood is a metaphysical category that captures the relational, intelligent, self-conscious, and self-determining nature of the human being as a unity of body and soul. The classical definition (Boethius) is "an individual substance of a rational nature."

It claims that dignity is universal: every human being, from conception to natural death, has it. No exception, no degree, no condition.

It claims that dignity is unconditional: it does not depend on any quality the person possesses, any function they fulfill, any social role they hold, or any capacity they exhibit. A human in a vegetative state has full dignity. A human with severe cognitive disability has full dignity. A human criminal has full dignity. A human enemy combatant has full dignity. Dignity is not earned and cannot be lost.

It claims that dignity is inviolable: no human authority can legitimately remove it. Dignity can be respected or violated; it cannot be canceled.

These claims are not minor variations on Kantian or capability accounts. They are a different structure of grounding. The work that capacity does in Kantian and capability accounts is done by nature in the Catholic account. Dignity belongs to a being because of what kind of being it is, not because of what that being can do.

This is the philosophical move that gives the Catholic account its distinctive shape. It is also what makes it uniquely well-suited to AI.

Why the Catholic account holds up in the AI age

The hardest test for any account of dignity is whether it generates stable moral conclusions across the cases the world produces. Accounts that work in easy cases but break in hard ones are accounts in trouble.

AI is producing hard cases. The question is whether the major dignity accounts hold up under the pressure.

Hard case 1: AI exhibits increasingly impressive cognitive capacities. Some current AI systems perform better than most humans on certain reasoning tasks. The trajectory points to more, not less, of this. On accounts that ground dignity in cognitive capacity (strong Kantianism, especially), there is no principled way to deny dignity to AI as its capacities approach or exceed human performance. The account either has to extend dignity to AI (a significant moral claim with practical consequences) or has to articulate why human-level cognitive performance in AI doesn't ground dignity even though human-level cognitive performance in humans does. The second move is hard to make consistently.

The Catholic account does not face this problem. AI's cognitive capacities are not the basis on which humans have dignity, so impressive AI capacities don't generate any pressure on human dignity. They also don't generate pressure to extend dignity to AI, because dignity in the Catholic account isn't about capacity at all. AI can be as impressive as it wants; the structure of moral standing doesn't shift.

Hard case 2: humans with limited cognitive capacities. The case is older than AI but is sharpened by AI. If dignity tracks cognitive capacity, then humans with severe cognitive disabilities, humans in vegetative states, humans with advanced dementia, and human infants are in difficulty. Their dignity is either derivative (grounded in the dignity of the species or of their previous selves) or attenuated. This is uncomfortable but follows from the capacity-based grounding.

The Catholic account treats these humans no differently from any other. Dignity belongs to the human person because of the kind of being a human person is, not because of the cognitive capacities a particular human happens to exhibit. A newborn and a Nobel laureate have the same dignity. A person with Alzheimer's disease in advanced stages has the same dignity she had at age 25.

This is not a marginal feature. It is one of the most consequential things the Catholic account does. In an age when AI's cognitive performance is being used (implicitly or explicitly) to redefine what counts as morally important, an account that refuses to make dignity a function of capacity has a structural advantage.

Hard case 3: AI-mediated decisions about humans. Hiring algorithms, lending algorithms, healthcare triage algorithms, parole risk-assessment algorithms. These systems make consequential decisions about persons by processing them as data points. The wrong here is not just that the algorithms might be biased (though they often are); the deeper wrong is that the algorithmic process treats the person as something other than what they are. It treats them as a problem to be optimized, a risk to be scored, a probability to be calculated.

All the dignity traditions agree that this is wrong, but they ground the wrong differently. The Kantian grounds it in the failure to treat the person as an end in themselves. The capability theorist grounds it in the erosion of practical reason. The Catholic grounds it in the failure to recognize the person as imago Dei: a being whose moral status is unconditional and inviolable.

The Catholic grounding has a particular bite here. Algorithmic decision-making is in a sense the perfect technology for capacity-based moral thinking: it processes inputs and produces outputs based on the features the system has decided are relevant. If dignity were a function of features, the algorithm could in principle calculate it. The Catholic insistence that dignity is not a function of features is what makes it resistant to algorithmic capture. There is nothing in the person an algorithm can compute that exhausts what makes them morally important.

Hard case 4: AI relationships and the substitution of presence. AI companions, chatbot therapists, simulated friends. The marketing implies that these systems can stand in for relationships with persons. Most dignity traditions are uncomfortable with this but struggle to articulate the wrong precisely.

The Catholic account articulates it cleanly. The wrong is not that AI is bad at simulating relationship; some AI systems are quite good at it. The wrong is that the user is being invited to treat a non-person as a person, which fails to honor what the person actually is (a being whose nature is to be in relationship with other persons) and what the AI actually is (a tool that simulates without being). Antiqua et Nova calls this a "grave ethical violation" (paragraphs 60-62). The grounding is the dignity of the human person, who deserves real relationship with real persons rather than the simulation of relationship with non-persons.

The convergence in practice

It is worth saying clearly that the four accounts of dignity surveyed above converge in many practical conclusions. They all agree that:

AI must not be used to manipulate persons against their interests. AI must not be deployed in ways that reduce persons to data points. AI must not concentrate power in ways that diminish human agency. AI must respect human privacy and security. AI must be accountable through human responsibility. AI must protect vulnerable groups disproportionately affected by AI's failures. Lethal autonomous weapons systems require special scrutiny and probably prohibition. AI's environmental costs must be honestly accounted for.

All four traditions support these conclusions. The disagreements are about grounding, not (mostly) about what dignity requires in practice.

Why does the grounding matter, then, if the conclusions converge? Three reasons.

First, hard cases reveal where the conclusions don't converge. AI's effect on humans with limited cognitive capacities, AI's standing if it becomes increasingly capable, the moral evaluation of AI relationships: these are cases where the accounts can pull apart. The grounding matters when the practical paths fork.

Second, the grounding shapes which considerations get weight. A framework that grounds dignity in rational capacity gives capacity-related considerations more weight in close cases. A framework that grounds dignity in imago Dei gives different considerations more weight. Both can converge on most cases while differing on how they handle exceptions.

Third, the grounding shapes long-term stability. As AI advances, frameworks that ground dignity in features AI exhibits will face pressure their foundations may not bear. Frameworks that ground dignity in something AI cannot exhibit by its nature are more stable. The Catholic account is in this second category.

What this means for AI ethics

Several practical implications follow.

Use the word "dignity" with awareness of what tradition you're using. The phrase is doing different work in different documents. When the EU AI Act invokes dignity, it is invoking a hybrid Kantian-human-rights tradition. When Magnifica Humanitas invokes dignity, it is invoking the Catholic imago Dei tradition. When UNESCO invokes dignity, it is doing capabilities-tinged human rights work. The convergences are real and important; the differences become visible in the hard cases.

Pay attention to what your dignity claim does in cases where it gets stressed. Does it commit you to extending dignity to AI as AI's capacities grow? Does it commit you to diminishing the dignity of humans whose capacities are limited? Does it commit you to treating dignity as a feature that can be computed or as something that resists computation? These are real choices, and they have real consequences for what AI you build and what AI you accept.

Take seriously that capacity-based dignity accounts are under pressure. The pressure is not malicious; it is the result of taking the dignity claim seriously and noticing what it implies. But the pressure is real, and AI is intensifying it. Frameworks that have a different grounding for dignity may be more robust as the AI moment deepens.

Notice the Catholic alternative even if you can't follow its theological commitments. The structural feature of the Catholic account, that dignity is grounded in kind rather than capacity, can be appreciated philosophically without requiring agreement on the theological foundations. Several recent secular philosophers have argued for similar structural features on different grounds. The Catholic account is not the only account with this structure, but it is the most developed and the most consistently maintained over time.

Do not let the convergence of practical conclusions hide the depth of the disagreement on grounding. The agreements about what AI should and should not do are real and important. The disagreements about why are also real and become important as the cases multiply.

The Magnifica Humanitas frame

The page on Pope Leo XIV's first encyclical, Magnifica Humanitas, notes that the encyclical's full subtitle is "On the Protection of Human Dignity in the Age of Artificial Intelligence." This is not a slogan. It is a precise philosophical claim. The encyclical, which releases May 25, 2026, is positioning dignity not as one principle among many in AI ethics but as the central principle on which everything else turns.

The encyclical will build on Antiqua et Nova's metaphysical foundations and on a century of Catholic Social Teaching's development of dignity as the foundational moral concept. It will also speak to the broader AI ethics conversation, which is using the word "dignity" constantly without always grounding it carefully.

For readers who want to understand what Pope Leo XIV is doing, this is the conceptual frame: he is articulating, with the magisterial weight of an encyclical, a particular account of human dignity that he believes is uniquely well-suited to the AI age. The account is not novel; it is the Catholic imago Dei tradition. What is new is its application, in this comprehensive form, to AI. The encyclical's success will be measured in part by whether the dignity account it articulates can do the work the AI age requires of it.

This page is the conceptual primer for that work.