This is the seventh and final post in the “AI and the Future of Work” series hosted on the Notre Dame-IBM Technology Ethics Lab’s blog. In this series, we examine the implications of AI for jobs and society: how tasks and roles might change, how organizations can adapt, how workers are affected, and what governance frameworks are being considered to create and support beneficial and responsible outcomes.
Throughout this series, we have examined how AI reshapes tasks, reconfigures skills, redistributes costs, and challenges existing governance structures in the workplace. Each of these dimensions raises important questions. But they share a deeper foundation that this final blog addresses directly: the relationship between work and human dignity. The core concern is not whether AI systems perform efficiently — they often do — but whether efficiency, treated as the primary or sole criterion for deployment, can erode the conditions under which work respects the inherent worth of the people who perform it. This blog develops a normative framework for evaluating workplace AI through the lens of dignity, drawing on established philosophical traditions, institutional commitments, and emerging evidence about how workers experience AI-mediated environments.
What dignity at work means
Immanuel Kant, the Enlightenment philosopher, grounds dignity in rational autonomy. In the Groundwork of the Metaphysics of Morals, Kant distinguishes between what has a price—and can therefore be exchanged for an equivalent—and what has dignity, which is "raised above all price and therefore admits of no equivalent." One Kantian formulation, the formula of humanity, requires treating persons "always at the same time as an end, never merely as a means." Applied to workplace AI, the operative word is “merely.” The violation occurs not when an AI system supports or coordinates work, but when it reduces workers to optimizable inputs in a grand system or treats them as nothing more than pathways to other ends, bypassing their rational consent through opacity, deception, or coercion.
The Catholic social tradition, rooted in the concept of the human person as created in the image of God, offers a thicker and more relational account. John Paul II's encyclical Laborem Exercens distinguishes between the objective dimension of work (tools, inputs, outputs, processes) and the subjective dimension (the worker as a conscious, purposive individual) and insists that the subjective dimension is primary: "Man is treated as an instrument of production, whereas he — he alone, independently of the work he does — ought to be treated as the effective subject of work and its true maker and creator." The Vatican's 2025 Antiqua et Nova extends this analysis to AI directly, noting that current approaches to the technology "can paradoxically deskill workers, subject them to automated surveillance, and relegate them to rigid and repetitive tasks." Pope Francis's 2024 World Day of Peace message similarly urged that "respect for the dignity of laborers and the importance of employment for the economic well-being of individuals, families, and societies" be treated as a high priority as AI systems penetrate workplaces. And Pope Leo XIV, in his first address to the College of Cardinals in 2025, framed AI explicitly in the tradition of Rerum Novarum — the 1891 encyclical written in response to the Industrial Revolution — calling for Catholic social teaching to respond to "another industrial revolution" that "pose[s] new challenges for the defense of human dignity, justice and labor."
The capabilities approach, developed by Amartya Sen and Martha Nussbaum, provides a third lens. Rather than grounding dignity in metaphysical claims, it evaluates whether people possess the real freedoms to live lives worthy of human dignity. Nussbaum's tenth central capability explicitly includes "being able to work as a human being, exercising practical reason and entering into meaningful relationships of mutual recognition with other workers." This reframes the evaluative question: the issue is not only whether a given AI deployment raises productivity, but whether it preserves workers' real freedom to exercise judgment, form collegial relationships, and exert meaningful control over their material and professional environment.
These three traditions differ in their foundations of rational autonomy, theological anthropology, and substantive freedom, but they converge on a shared prohibition. Systems that treat workers merely as inputs to be optimized, monitored, and managed without regard for their status as purposive agents violate the conditions of dignified work, regardless of the efficiency gains they produce. Importantly, purposive agency in this context should not be understood narrowly as limited to creativity, self-actualization, or highly skilled “knowledge work.” Human purpose in work also encompasses more basic but no less dignified aims: supporting one’s family, securing stability, maintaining independence, participating in social life, and sustaining the conditions necessary for survival. Accordingly, the erosion of dignified labor through automation is not confined to elite professional occupations. The displacement, fragmentation, or hyper-surveillance of lower-wage, manual, and service-sector work may equally undermine workers’ ability to exercise agency, maintain economic security, and participate in work as persons rather than as interchangeable units of production.
What makes work meaningful
Dignity sets a floor below which work becomes degrading. But the ethical analysis of workplace AI also requires attention to what makes work meaningful, because AI systems can satisfy minimum dignity thresholds while still hollowing out the dimensions of work that give it value to the people performing it. The concern is therefore not only whether workers are treated humanely, but whether they retain opportunities for agency, growth, recognition, and purpose through their labor. What follows are several complementary accounts of how meaning is constituted through work and why AI-driven transformations may threaten those conditions.
Hannah Arendt's The Human Condition offers a foundational distinction. She distinguishes labor (cyclical biological maintenance), work (the fabrication of durable objects and artifacts), and action (the disclosure of one’s inmost self through initiative and interaction with others). Arendt's concern was that modernity progressively collapses these tiers: work is reduced to labor, and action is displaced by process and consumption. In contemporary societies, this erosion is often obscured by narratives that frame work primarily through the language of careerism, productivity, self-optimization, or personal fulfillment. Workers are encouraged to “do what they love” or align identity entirely with professional achievement, even as many forms of labor become increasingly standardized, monitored, and subordinated to systems of efficiency and consumption. Generative AI raises a version of this concern in concentrated form. Generative AI raises a version of this concern in concentrated form. When AI systems absorb not only routine labor but also significant portions of work such as drafting, analysis, design, and interpretation, the question is whether workers retain a domain of action, a sphere in which they appear to colleagues and clients as distinct persons exercising initiative, or whether they become components or supervisors of processes they did not shape and outputs they did not generate. This concern about the erosion of human agency connects closely to Alasdair MacIntyre's concept of practices and their internal goods.
Alasdair MacIntyre's concept of practices and their internal goods sharpens this point. A practice, in MacIntyre's sense, is a coherent activity through which goods internal to that activity are realized. The practice of medicine, teaching, engineering, or legal analysis each produces internal goods (diagnostic skill, pedagogical judgment, structural insight, argumentative precision, corporeal/material/or interpersonal care) that are distinct from external goods like revenue or status. Internal goods are developed through apprenticeship, failure, and iterative refinement. AI tools can deepen practices when they free practitioners for the judgment and engagement through which internal goods are cultivated. But they can also bypass the apprenticeship process entirely, delivering outputs that resemble the artifacts of skilled practice without developing the capacities that make such practice meaningful.
Building on this idea, Andrea Veltman's philosophy of meaningful work identifies several dimensions through which work contributes to human flourishing:
1) Developing and exercising human capabilities
2) Supporting virtues such as pride, integrity, and self-discipline
3) Providing a sense of purpose
4) Integrating work with broader elements of one's life including relationships and identity.
Veltman’s framework helps explain why concerns about AI and work extend beyond economic displacement alone. Even where employment formally remains intact, systems that strip workers of discretion, skill development, or opportunities for contribution may diminish the conditions under which work becomes meaningful.
Richard Sennett captures a related idea in his account of craftsmanship as "an enduring, basic human impulse, the desire to do a job well for its own sake." His emphasis on craftsmanship reinforces the importance of engagement, mastery, and pride in competent performance. At the same time, neither Sennett nor Veltman romanticizes labor as such. Veltman explicitly endorses the automation of genuine drudgery. The point is not that all human labor must be preserved, but that meaningful work involves irreducible dimensions of skill development, moral formation, social contribution, and purposeful engagement that deserve protection even in technologically advanced workplaces.
Finally, recognition theory, particularly as developed by Axel Honneth, adds an explicitly social dimension to these accounts. The workplace is a primary site where individuals receive recognition for their contributions that sustains self-esteem and social standing. When algorithmic evaluation systems reduce contributions to quantified metrics, they risk producing what Honneth would call systematic misrecognition: the invisibilization of care work, mentoring, informal problem-solving, and collaborative judgment that sustain organizations but resist measurement. In this sense, the problem with workplace AI is not merely that it can automate tasks, but that it can narrow the forms of contribution organizations are capable of seeing and valuing.
How efficiency-driven AI deployment may erode both dignity and meaning
This kind of philosophical analysis becomes practically urgent when set against emerging evidence of how AI systems are actually experienced in workplaces. The concern is not with AI in the abstract but with a specific pattern: deployments organized primarily around efficiency maximization, cost reduction, and managerial control.
Scholars have documented the mechanisms through which algorithmic systems restructure workplace power. For example, Kellogg, Valentine, and Christin's influential typology identifies six functions of algorithmic control: systems restrict worker options, recommend actions, record behavior, rate performance, replace those who underperform, and reward compliance. These systems are "more comprehensive, instantaneous, and opaque than traditional control mechanisms." In a separate but related piece, Aloisi and De Stefano describe the result as an intensification of managerial power that produces an “assault on dignity and the destruction of individual and collective worker privacy." The concern is that algorithmic management can operate with a scope, granularity, and opacity that forecloses the contestation and dialogue that is required to make authority accountable.
Even in augmentation contexts, where AI tools are designed to support rather than replace, the evidence is mixed. A landmark study of AI-assisted customer service agents found average productivity gains of fourteen percent and significantly larger benefits for less-experienced workers. But research on similar deployments suggests that AI copilots can also undermine intrinsic motivation and erode workers' sense of control over their own performance; another study recently showed use of AI can reduce workers’ self-reported confidence in their own knowledge and skills. Efficiency gains and dignity harms are thus not mutually exclusive; they can coexist in the same system, which is precisely why efficiency alone is an insufficient evaluative criterion.
Erik Brynjolfsson's concept of the "Turing trap" captures this structural dynamic: "As machines become better substitutes for human labor, workers lose economic and political bargaining power… In contrast, when AI is focused on augmenting humans… humans retain the power to insist on a share of the value created." The trap is not a property of the technology but of the incentive structures surrounding it. When organizations pursue automation as a default and treat augmentation as an afterthought, they systematically erode the conditions under which workers can negotiate, contest, and shape how AI changes their roles.
Madeleine Clare Elish's concept of the "moral crumple zone" identifies a further dignity harm: in human-in-the-loop designs, human operators can become the point at which accountability collapses, absorbing blame for complex, blackbox system failures they lacked the authority or information to prevent. This creates a structural injustice: responsibility without power, accountability without agency.
What institutions already affirm
The normative intuition that efficiency must be bounded by dignity is increasingly embedded in international institutional frameworks.
The ILO's Centenary Declaration of 2019 commits to a "human-centred approach to the future of work, which puts workers' rights and the needs, aspirations and rights of all people at the heart of economic, social and environmental policies." The ILO's Global Commission on the Future of Work calls for a future "that affords dignity, security and equal opportunity, expanding human freedoms" and proposes a "human-in-command" approach to algorithmic accountability.
The OECD AI Principles, adopted by forty-seven governments and updated in 2024, commit signatories to respecting "human rights, democratic and human-centred values throughout the AI system lifecycle," including "dignity" and "internationally recognised labor rights." The EU's High-Level Expert Group Ethics Guidelines for Trustworthy AI are explicit: AI systems "should not unjustifiably subordinate, coerce, deceive, manipulate, condition or herd humans" and should "augment, complement, and empower human cognitive, social, and cultural skills," with particular attention to "asymmetries of power or information, such as between employers and employees."
UNESCO's Recommendation on the Ethics of AI, adopted by 193 member states, establishes dignity as its first value and calls on governments to ensure that workers' rights are respected and that transitions are managed fairly. The Rome Call for AI Ethics commits signatories to AI "that respects the dignity of the human person" and "does not have as its sole goal greater profit or the gradual replacement of people in the workplace." The NIST AI Risk Management Framework, while voluntary, explicitly names "dignitary harm" as a risk category.
Taken together, these frameworks do not constitute binding law in most jurisdictions, but they represent an unusually broad normative consensus: across secular and religious traditions, across advanced and developing economies, and across labor, technology, and human rights institutions, the principle that workplace AI must respect human dignity, and that efficiency is an insufficient criterion, is increasingly treated as foundational.
Preserving domains of irreducibly human work
A stronger claim emerges from the philosophical and institutional analysis that certain domains of work should be preserved for humans not because AI cannot yet perform them, but because the humanness of the person performing them is constitutive of the good the work produces.
Shannon Vallor's work on technology and the virtues argues that moral formation depends on "intimate and repeated exposure to our mutual dependence, vulnerability, weakness, concern, and gratitude for one another." When technological surrogates replace human agents in caregiving, teaching, or counseling, what is lost is the relational conditions through which moral development occurs for both the provider and the recipient. Brian Cantwell Smith draws a related distinction between reckoning (calculative operations) and judgment (deliberative thought grounded in ethical commitment and responsible action), warning that societies may come to "rely on reckoning systems in situations that require genuine judgment" and recalibrate expectations for human mental activity accordingly. Frank Pasquale's proposed "new laws of robotics" codify the principle, suggesting that AI should complement professionals rather than replace them and should not "counterfeit humanity" by simulating relational engagement that only persons can authentically provide.
The argument here is not protectionist. It is a judgment about the nature of specific goods. Education, as Gert Biesta argues, "always involves a risk" precisely because students are not objects to be molded but subjects whose responses are unpredictable and whose agency is to be cultivated. Caregiving, adjudication, pastoral work, and democratic deliberation, for example, similarly depend on the presence of a responsible human agent whose engagement is not merely functional but constitutive. Preserving space for these domains is a commitment to the conditions under which certain distinctively human goods remain possible and respected.
Toward dignity-centered design and governance
If dignity sets ethical limits on workplace AI deployment, the question becomes how those limits might be operationalized. Three levels of response are proposed, each building on the others.
At the design level, frameworks such as Ben Shneiderman's human-centered AI and Batya Friedman and David Hendry's value sensitive design offer structured approaches for embedding human values into system requirements from the outset, rather than treating them as constraints to be satisfied after optimization goals are set.
At the organizational level, however, design-level interventions are necessary but insufficient. Ben Green's review of forty-one human oversight policies found that many provide "a false sense of security in adopting algorithms and enable vendors and agencies to shirk accountability for algorithmic harms." Individual human-in-the-loop oversight, when unsupported by institutional structures, training, and genuine authority, becomes a legitimation device rather than a safeguard. Effective oversight usually requires collective and institutional mechanisms: for example, work councils with codetermination authority over monitoring technology, algorithmic impact assessments, and escalation pathways that are resourced and enforceable.
And at the governance level, the principle of worker voice (the ability of workers and their representatives to participate in decisions about how AI is introduced and managed) is not merely an instrumental tool for better outcomes. As Lisa Herzog argues, participation in the rules that govern one's conditions of work is constitutive of self-respect; being "recognised and affirmed as equals" in shaping those conditions is part of what it means to be treated as a person rather than an input. The ILO's tripartite model, the OECD's emphasis on social dialogue, and the EU's requirements for worker notification all reflect this understanding. Together, they argue that governance strategies which exclude worker voice in decisions about AI deployment reproduce the very asymmetry of power that dignity norms are meant to constrain.
Final thoughts
This series began with a question about what AI is and what it is not, moved through the design choices that shape how AI transforms tasks and skills, and examined who bears the costs when those transformations are left unmanaged. This final blog post has argued that beneath these practical questions lies an important normative one: what kind of work, and what kind of relationship between persons and their work, do we owe each other?
The answer offered here is that efficiency is an instrumental value rather than a terminal one. It serves human purposes when it reduces drudgery, expands capability, and creates the material conditions for flourishing. The philosophical traditions examined in this post disagree on foundational premises but converge on the idea that systems that work to or in effect reduce persons to optimizable variables, foreclose their agency, and/or render their contributions invisible critically violate conditions that any adequate account of dignified work must protect.
The institutional consensus is striking precisely because it bridges traditions and geographies: the ILO, OECD, EU, UNESCO, the Vatican, and major technology companies have all affirmed that workplace AI must be bounded by dignity. The challenge, as this series has explored throughout, is translating that consensus into the design choices, governance structures, and organizational practices that determine how AI is actually experienced by the people whose work it transforms. The same AI capability can expand what workers are able to do or contract it; can deepen their engagement or hollow it out; can distribute gains widely or concentrate them narrowly. What determines the outcome is not the technology itself, but the values embedded in its creation and deployment and whether the institutions responsible for that deployment treat human dignity as a constraint to be satisfied or as the criterion by which their choices are ultimately judged.
To the readers who have followed this series from the beginning and to those joining at the end: thank you. It has been an honor to work through these ideas with you, and I hope they serve as useful starting points for the harder, more particular decisions that lie ahead in your own institutions and communities. I look forward to the conversations ahead.
Perla Khattar is a J.S.D. candidate at the University of Notre Dame Law School and a Research Fellow at the Notre Dame-IBM Tech Ethics Lab. In her research, she analyzes emerging technologies such as synthetic data, neurodata, quantum computing, and AI through a privacy lens, focusing on how law and policy can anticipate technological disruption while safeguarding individual rights.
]]>
