1. Introduction
Artificial intelligence (AI) is increasingly embedded across sociotechnical domains, raising questions about how these systems shape human experience, redistribute influence, and mediate consequential decisions. In response, AI ethics frameworks have converged around principles such as fairness, transparency, and accountability (Reference Jobin, Ienca and VayenaJobin et al., 2019), establishing an important foundation for responsible AI development. Yet even when rigorously implemented, these principles offer limited guidance on how systems structure relational respect, negotiate power dynamics, or shape the lived experience of those subject to algorithmic decision-making. A system may be technically transparent and protect privacy, yet still be experienced as constraining, paternalistic, or dehumanizing (Reference Mittelstadt, Allo, Taddeo, Wachter and FloridiMittelstadt et al., 2016; Reference Yang, Steinfeld, Rosé and ZimmermanYang et al., 2020). These limitations become clearer when AI is understood not as a standalone tool but as part of broader institutional arrangements. Here, AI systems are treated as sociotechnical infrastructure embedded within workflows, professional norms, and governance structures. Design choices therefore shape who has decision authority, how much room actors have to exercise judgment, and how individuals are positioned within institutional processes. In healthcare, for example, diagnostic or decision-support tools may reshape clinician autonomy, patient agency, and organizational expectations in ways that alter professional identity and care experience (Reference Dean, Talbot and DeanDean et al., 2019). Such tensions between efficiency and relational quality, or between safety and autonomy, point beyond these principles alone toward questions about how systems structure the treatment and standing of those affected.
One way to engage these questions is through the concept of dignity. Long central to human rights, ethics, and political philosophy (United Nations General Assembly, 1948; Reference McCruddenMcCrudden, 2008), dignity has been variously described as inherent worth, relational recognition, moral status, and the capacity for agency or flourishing (Reference NordenfeltNordenfelt, 2004; Reference NussbaumNussbaum, 2011; Reference Edlund, Lindwall, Post and LindströmEdlund et al., 2013). In complex sociotechnical environments, dignity can be difficult to assimilate: preserving it for one group may constrain it for another, and expanding agency in one dimension may introduce new risks elsewhere (Reference Friedman, Kahn, Borning and HuldtgrenFriedman et al., 2013; Reference ShiltonShilton, 2018). Rather than treating it as a fixed moral conclusion, this paper approaches dignity as a design-relevant construct that becomes more visible when technologies mediate power, authority, and consequential choice (Reference Latour, Bijker and LawLatour et al., 1992).
This approach is referred to herein as dignity-centered design (DCD) and is framed as a sociotechnical design framework for reasoning about these tensions in AI-mediated systems. Drawing on philosophical and healthcare traditions and extending the operational logic of Project UNITE, Inc.’s (“UNITE”) Dignity Index® (Reference Ball, Byrne, Meppen and GrahamBall et al., 2023) from discourse into system design, it offers a structured way to compare dignity-relevant design choices. The Dignity Index® demonstrates how an abstract normative concept can be translated into observable criteria and comparative judgments (Reference Ball, Byrne, Meppen and GrahamBall et al., 2023); this logic is adapted to AI by shifting the focus from speech acts onto AI-system features.
The framework presented here offers a theoretically grounded lens intended to inform reflective design practice and future research. Three interrelated dimensions of dignity (individual, relational, and systemic) are articulated alongside five core design principles, and a Dignity Spectrum in AI Design is introduced to support comparative assessment of how particular design decisions may preserve or undermine dignity. The central question guiding this argument is how dignity can function as a complementary design lens within responsible AI, enabling evaluation of AI systems not only in terms of outcomes, but in terms of how they shape lived experience and distribute influence across institutional contexts.
2. Background: Dignity in sociotechnical and AI contexts
2.1. Current AI ethics frameworks
Global AI ethics guidelines from governmental bodies, professional associations, and technology companies have gathered emphasis around five key principles: accountability, beneficence (do no harm), equity and fairness, transparency, and privacy protection (Reference Jobin, Ienca and VayenaJobin et al., 2019). While mentioned, notably underemphasized are concepts like dignity, solidarity, and sustainability. The “ethics washing” phenomenon has also been critiqued, where organizations adopt ethical principles rhetorically without embedding them in design practice (Reference BiettiBietti, 2020). Principles for ethical AI have been widely discussed, yet a gap persists between high-level commitments and implementation (Reference Alahmed, Abadla, Ameen and ShteiwiAlahmed et al., 2023). While these principles are necessary, translating them into concrete design criteria, measurable practices, and enforceable governance structures remains difficult. Questions about how systems shape quality of human treatment and relational positioning are not easily operationalized in ways that meaningfully guide everyday design decisions. For example, fairness frameworks, while essential, also reveal some inherent limitations. In machine learning and algorithmic fairness literature, multiple mathematical definitions of fairness exist and often conflict (Reference Corbett-Davies, Gaebler, Nilforoshan, Shroff and GoelCorbett-Davies et al., 2023). A system fair by one definition may be demonstrably unfair by another. Moreover, fairness frameworks primarily address distributional outcomes across groups but do not directly evaluate how specific system configurations structure opportunities for discretion, appeal or override, and recognition within institutional workflows. A patient may receive statistically fair allocation of diagnostic resources while experiencing profound indignity in how they experience care mediated by the AI system. Explainability and transparency similarly provide necessary but insufficient guidance. Explaining how an algorithm works does not ensure it respects human agency. Transparent processes can still be dehumanizing. These gaps motivate this search for a complementary principle that addresses dimensions other frameworks leave unaddressed.
At the same time, responsible AI frameworks remain indispensable components of AI design, and the limitations discussed here do not render these frameworks obsolete; rather, they highlight that current standard metrics alone may not capture experiential dimensions of system impact. DCD is therefore proposed as a complementary analytic lens that examines how authority, discretion, and recognition are mediated within AI-mediated systems.
2.2. Dignity theory: Philosophical and applied foundations
Dignity’s conceptual foundations span many domains such as international human rights law, philosophical ethics, and healthcare contexts. The Universal Declaration of Human Rights (1948) asserts that “all human beings are born free and equal in dignity and rights,” positioning dignity as foundational to modern human rights discourse. The capability approach furthers this foundation by emphasizing that dignity entails not only protection from harm but also the presence of substantive freedoms that enable agency and flourishing (Reference SenSen, 1999). Within this tradition, Reference NussbaumNussbaum (2011) argues that a life with dignity requires securing central capabilities, such as practical reason, affiliation, and control over one’s environment, which together create conditions for full human functioning. Kantian moral philosophy frames dignity as treating persons as ends in themselves rather than merely as means (Reference Gregor and WoodGregor & Wood, 1998), a principle with clear implications for technology-mediated systems where individuals risk reduction to optimization targets. Legal scholarship further demonstrates that dignity is interpreted variably across jurisdictions, reflecting an understanding of dignity as inherent human worth as well as a relational understanding concerned with treatment and recognition (Reference McCruddenMcCrudden, 2008). These perspectives collectively underscore that dignity operates simultaneously at individual, relational, and institutional levels.
In healthcare and bioethics, dignity has been articulated in more granular terms. Reference NordenfeltNordenfelt (2004) distinguishes among several forms of dignity, including dignity of merit, moral stature, identity, and universal human dignity. Empirical work on dignity-conserving care identifies specific clinical behaviors and contextual factors that either support or compromise patient dignity (Reference ChochinovChochinov, 2002). Violations of dignity in care settings have been associated with psychological distress and diminished well-being, demonstrating that dignity is not solely an abstract philosophical concept but a lived and measurable dimension of care experience (Reference Edlund, Lindwall, Post and LindströmEdlund et al., 2013). These applied traditions show that dignity can be operationalized in context-sensitive ways without collapsing into a single definition.
The Dignity Index® represents a further step in this trajectory by translating the concept of dignity into an evaluative framework for public discourse (Reference Ball, Byrne, Meppen and GrahamBall et al., 2023). Developed through nonprofit UNITE’s collaboration with the University of Utah, the Dignity Index® operationalizes dignity along an eight-point spectrum ranging from dehumanization to full recognition of inherent worth (Reference Ball, Byrne, Meppen and GrahamBall et al., 2023). Rather than treating dignity as binary, it is framed as a matter of degree grounded in observable communicative features. Pilot validation research demonstrates reliable intercoder agreement, indicating that structured dignity evaluation can be made analytically tractable (Reference Ball, Byrne, Meppen and GrahamBall et al., 2023). Its significance for design methodology lies in its demonstration that an abstract normative concept can be translated into identifiable characteristics and systematic reflection. Although developed for discourse analysis, its graded structure can be adapted to other domains where human treatment is mediated. Applied to AI systems, the unit of analysis shifts into system features, interaction patterns, and governance structures. As a technology system, AI does not merely process information; it can allocate authority, constrain or enable discretion, structure opportunities for override, and shape how individuals are recognized within institutional processes (Reference Latour, Bijker and LawLatour et al., 1992). Adapting the operational logic of the Dignity Index® to AI design therefore involves identifying how specific design choices map onto degrees of dignity preservation or erosion across these mediated interactions. This adaptation does not assume equivalence between public discourse and technology or treat dignity as a fixed metric. Rather, it treats dignity as an evaluative dimension that becomes visible when examining how systems structure agency, recognition, and power.
2.3. Human-centered and participatory design in AI
Human-centered AI (HCAI) promotes systems that enhance human capabilities and maintain meaningful human control (Reference ShneidermanShneiderman, 2020). Similarly, participatory design emphasizes stakeholder involvement in shaping technologies across development stages (Reference Muller and KuhnMuller & Kuhn, 1993). These traditions provide important foundations for collaborative and user-oriented design. However, both approaches may prioritize usability, performance, or procedural inclusion without systematically interrogating how authority, recognition, and discretion are structured within sociotechnical systems (Reference Yang, Steinfeld, Rosé and ZimmermanYang et al., 2020). Participation alone does not guarantee dignity; engagement processes may remain tokenistic, and power asymmetries may persist despite formal inclusion. DCD builds on the HCAI tradition by providing an explicit evaluative lens for examining how systems redistribute power, mediate relationships, and shape the conditions under which stakeholders are recognized and able to exercise agency.
3. Conceptual foundations: Dignity as a design construct
3.1. Defining dignity in design-science context
DCD is presented here as an organizing framework that positions human worth, agency, and respect as explicit considerations throughout the design lifecycle. This provides a structured lens for examining how design decisions allocate decision-making power, constrain or enable professional judgment, and frame the status and standing of individuals within institutional processes. Consistent with prior scholarship demonstrating dignity operates across dimensions, DCD is articulated through three interrelated levels of analysis: individual, relational, and systemic.
Individual dignity concerns recognition of persons as more than data subjects, risk scores, or optimization variables. In design terms, this dimension examines how systems frame users: whether they are treated as interchangeable units or as agents with meaningful identities, expertise, and valuable context-specific knowledge. Questions at this level include whether design choices preserve material opportunities for self-direction and whether individuals retain the ability to understand, contest, or influence decisions that affect them. Relational dignity addresses how systems mediate interactions among stakeholders. Sociotechnical systems shape relational power dynamics, communication flows, and the distribution of discretion. This dimension examines whether system design facilitates mutual recognition and justified authority, or instead introduces asymmetries that render one party opaque, overruled, or systematically monitored. Systemic dignity attends to how technologies interact with institutional norms, governance structures, and broader patterns of equality and inequality. AI systems may expand access and capability, or they can reinforce existing hierarchies or standardize decision-making in ways that obscure opportunities for contextual nuance. This dimension examines how design choices redistribute power and opportunity across populations.
These three dimensions are analytically distinct but are entangled together in practical application. Enhancing agency at the individual level may constrain efficiency at the systemic level; standardization intended to improve safety may reduce discretionary autonomy in relational contexts. DCD does not resolve such tensions. Instead, it provides a framework for making them visible and subject to deliberation during design and evaluation.
In practice, DCD operates across the development lifecycle, with four general (model agnostic) stages: (1) problem framing, where dignity-relevant stakes, affected stakeholders, and potential tensions are identified; (2) contextual analysis, in which existing workflows, authority structures, and experiential impacts are examined; (3) design synthesis, where configurations are generated and compared using dignity-informed criteria; and (4) iterative evaluation, where deployed systems are revisited to assess dignity effects and adjust accordingly. These stages are not necessarily linear but present as recurring opportunities to surface and deliberate dignity implications throughout system development.
3.2. Dignity and fairness: Complementary evaluative dimensions
To clarify how dignity relates to existing responsible AI approaches, fairness is examined here as a representative and widely operationalized principle. Dignity can be understood as both complementing and extending fairness frameworks. Fairness typically focuses on distributional outcomes: whether resources, opportunities, or system performance are equitably allocated across groups. These questions remain central, as a system cannot reasonably be considered as dignity-supporting if it systematically disadvantages particular populations. At the same time, dignity draws attention beyond distribution into dimensions of lived experience. Two sepsis alert systems may demonstrate comparable sensitivity and equitable detection rates across patient groups while differing in how they communicate uncertainty, allow clinicians to dismiss alerts, and record override patterns for review. In one case, recommendations may function as effectively authoritative, with deviations flagged for review; in another, recommendations may be accompanied by uncertainty information and framed as inputs to professional judgment. Both may satisfy fairness criteria yet position clinicians and patients differently within the decision process, illustrating that fairness addresses outcome distribution while dignity attends to how decisions are enacted and experienced. It asks how individuals are situated within institutional processes, including whether they retain meaningful scope for judgment, influence, and contestation.
Dignity may also serve as a helpful perspective when fairness definitions conflict. As noted in the algorithmic fairness literature, multiple mathematical formulations of fairness can be mutually incompatible (Reference Kleinberg, Mullainathan and RaghavanKleinberg et al., 2016). In such cases, dignity offers an additional lens for reflecting on which approach better supports agency and respectful treatment in context. Building on these foundations, the following section translates dignity into a structured design framework for AI-mediated systems.
4. Design methodology: Operationalizing dignity
4.1. Core principles in designing for dignity in AI
The following principles articulate commitments that guide DCD and identify recurring domains of concern where dignity is implicated:
-
1. Primacy of human agency: Design systems to preserve, and where appropriate enhance, human capacity for self-directed action. Automation should support professional or personal judgment and should not displace decision authority without clear, contextually justified reasons.
-
2. Participatory governance: Design systems so that users and affected stakeholders have meaningful influence over problem framing, design, implementation, and revision. Where users bear responsibility for system outcomes, they should have corresponding authority to shape system rules and modification processes.
-
3. Recognize inherent worth: Design systems to account for the inherent dignity of affected individuals alongside performance objectives. When tradeoffs arise, design decisions should explicitly consider impacts on role integrity, lived experience, and relational respect, and should make clear how competing interests are balanced.
-
4. Relational transparency: Design systems to make decision authority, accountability structures, and avenues for contestation visible. Users should be able to identify who holds authority, how decisions are made, and how recommendations can be questioned or overridden.
-
5. Accountability to affected parties: Design systems so that decision impacts and underlying data practices remain traceable to identifiable actors or institutions and provide accessible mechanisms for review, correction, and redress. Those who design, deploy, or mandate system use should remain answerable to individuals and communities affected by system decisions and data use.
These principles describe what designers can attend to, while the Dignity Spectrum in AI Design introduced in Section 4.2 provides a structure for assessing how design choices represent or impair these commitments in practice.
4.2. Creating a Dignity Spectrum in AI System Design
The Dignity Spectrum in AI System Design (Table 1) translates these principles into a comparative structure for examining how design choices shape human treatment within sociotechnical systems. Each level describes a recognizable pattern of features that influence agency, role positioning, and power distribution across individual, relational, and institutional dimensions. The spectrum does not function as a scoring instrument or moral ranking. Instead, it provides a structured means of comparing how systems that achieve similar technical outcomes may differ in how they mediate the lived experience of its stakeholders. Importantly, the spectrum is not intended to assign a single global level to an entire system. In practice, different components, such as override mechanisms, data governance practices, personalization features, or performance monitoring structures, may align with different levels simultaneously. The spectrum therefore supports feature-level analysis rather than categorical labeling. The core principles and the spectrum can be applied iteratively across the development stages outlined in Section 3, serving as structured points of reflection during problem framing, contextual analysis, design synthesis, and evaluation.
Level 8 represents a mostly aspirational benchmark rather than a currently standardized state. It envisions systems that adapt to user expertise, support meaningful work, and structure transparent value exchange around data contributions. The emphasis shifts from harm prevention to fostering capability and recognition within interconnected systems. Levels 6 and 7, achievable in current practice, differ in how human–algorithm collaboration is structured. A Level 7 system presents recommendations with clear acknowledgment of uncertainty, allows seamless override, and treats human input as learning rather than deviation. A Level 6 system provides explanation and override but lacks adaptive reciprocity. Both may be accurate, but are distinguished in how authority and discretion are distributed. The transition from Level 5 to Level 4 marks the point where systems begin to harm. Level 4 treats users paternalistically, presuming algorithmic authority over human judgment. Systems that require extensive documentation to reject recommendations or flag deviations for review create pressure to comply rather than decide. Stakeholders remain accountable or liable for outcomes while lacking meaningful influence over system rules or design. Override exists but is cumbersome, discouraging professional judgment and increasing moral strain. This contrasts with Level 7, where override is seamless and valued as input. Levels 3 through 1 represent escalating dignity violations. Level 3 imposes surveillance that treats decisions as errors, eroding confidence and autonomy. Level 2 removes agency through opaque, unappealable rules, such as automated coverage denials or eligibility determinations issued without explanation or meaningful review. Level 1 automates life-altering decisions without explanation or recourse, denying that people deserve dignity in consequential choices. The framework demonstrates that systems can differ profoundly in dignity. DCD highlights these distinctions and centers human treatment as a primary design objective rather than a secondary concern.
Dignity Spectrum in AI system design

Table 1 Long description
A table with eight rows and two columns. The first column lists dignity levels from Level 8 to Level 1, and the second column describes key characteristics and design indicators for each level. Row 1: Level 8: Full dignity, Calibrated collaboration that strengthens human judgment and preserves meaningful choice. Adapts to user-stated expertise, values, and context with explicit control over personalization. Automates routine work while protecting time for relational care and high-judgment tasks. Credits human contributions (including overrides) and makes decision pathways legible to affected parties. Equitable value exchange for data contributions, with transparent governance and user control. Row 2: Level 7: Strong autonomy, Respects role-based expertise; preserves meaningful human decision authority in consequential contexts. Provides clear rationale and known limitations, including calibrated confidence/uncertainty information. Enables frictionless override and escalation pathways (no punitive flags). Makes accountability structures and appeal mechanisms visible and usable. Offers opt-in/out choices and acknowledges contributions within clear governance constraints. Row 3: Level 6: Generally respectful, Supports autonomy within explicit, justified parameters that are visible to affected parties. Provides understandable explanations and appropriate options for questioning or appeal. Allows reasonable override without punitive consequences. Acknowledges expertise and context within constrained workflows. Uses basic transparency and meaningful consent, with defined responsibilities and points of contact. Row 4: Level 5: Neutral, Functional but impersonal; minimal adaptation to context or expertise. Generic explanations; limited support for deliberation or value-sensitive choices. Override may exist but is inconsistently supported or poorly integrated into workflow. Governance and data practices are described in generic terms with limited practical relevance. Efficiency-oriented defaults with limited attention to quality of treatment of people. Row 5: Level 4: Constraining & paternalistic, Restricts discretionary judgment without adequate justification tied to safety or risk. Rigid workflows prioritize compliance over context-sensitive decision-making. Override exists but is burdensome or socially costly (documentation friction, implied blame). Limited visibility into who is accountable, how decisions can be contested, or what happens after contestation. Responsibility for outcomes placed on users without corresponding influence over system design or rules. Row 6: Level 3: Actively undermining, Treats deviations from algorithmic output as error requiring correction. Imposes surveillance or performance flagging that pressures conformity. Frames human judgment as bias to be managed rather than expertise to be engaged. Provides nominal contestation pathways that are procedurally burdensome or stigmatized. Uses data collection to monitor, rank, or discipline stakeholders. Row 7: Level 2: Eliminating agency, Eliminates meaningful opportunity for influence or contestation. Imposes unappealable decisions without transparency. Treats individuals as passive inputs to institutional objectives. Extracts data, labor, or behavioral signals without meaningful consent, representation, or influence over their use. Assigns responsibility to individuals without granting corresponding authority. Row 8: Level 1: Complete denial of worth, Automates life-altering decisions without explanation or recognition of moral status. Treats individuals as fully interchangeable or expendable. Removes any possibility of recognition, recourse, or redress. System logics convert data, labor, or expertise into value streams while disregarding the moral standing of those from whom they originate. Uses dehumanizing classifications or language. Deploys system logics that explicitly or implicitly regard certain roles or groups as disposable.
Note. Conceptually informed by the Dignity Index® framework and referenced with written permission from UNITE. The Dignity Index® and its associated proprietary framework, methodologies, implementations and materials remain the intellectual property of UNITE. Separate written authorization is required for commercial or applied uses beyond this scholarly work.
5. Application context: Healthcare AI systems
Healthcare offers a useful example context for DCD, as medical encounters can involve vulnerability, asymmetries in authority, and questions of human worth, while AI systems are increasingly shaping clinician–patient interactions, diagnostic reasoning, and treatment decisions (Reference ChochinovChochinov, 2002; Reference Edlund, Lindwall, Post and LindströmEdlund et al., 2013; Reference Mittelstadt, Allo, Taddeo, Wachter and FloridiMittelstadt et al., 2016). In this setting, design choices may influence professional identity, patient agency, and perceptions of moral responsibility in addition to technical performance. Three potential tensions help illustrate the relevance of DCD in this domain: clinician autonomy, patient agency, and moral injury. The comparative scenarios that follow draw on recurring issues described in clinical decision support, algorithmic governance, and EHR workflow research.
5.1. Clinician autonomy example: Emergency department triage system
Scenario A (Lower dignity; consistent with Level 3–4): The system displays a recommended triage level prominently and pre-populates associated orders. The explanation panel displays a probability score (i.e., “Predicted acuity: Level 2, 78% confidence”) without key contributing inputs, uncertainty bounds, or known blind spots. If a clinician selects a different acuity level, a mandatory pop-up requires completion of multiple justification fields before the order can be signed. Monthly reports include clinician-level override rates that are routed to supervisors. Over time, clinicians may experience their judgment as provisionally subordinate to the algorithm, affecting their sense of recognition as accountable decision-makers rather than monitored executors.
Scenario B (Higher dignity; consistent with Level 6–7): The system displays the recommended triage level alongside a structured explanation panel listing key contributing inputs (i.e., abnormal vital signs, prior admissions, presenting complaint), model confidence intervals, and a brief statement of limitations (such as limited visibility into nonverbal distress or caregiver concern). If a clinician selects a different acuity level, the interface allows progression without a blocking pop-up. An optional note field is available but not required. Override data are aggregated at the system level for model evaluation. Governance language, accessible within the interface, specifies that algorithmic outputs are advisory and that clinicians retain decision authority. In this configuration, professional discretion is operationalized as a legitimate component of the decision pathway. The design affirms clinicians as accountable agents whose contextual judgment is expected to complement algorithmic output.
5.2. Patient agency example: Cancer care treatment decisions
Scenario A (Lower dignity; consistent with Level 4–5): The interface displays a single treatment regimen labeled “Recommendation.” Alternative regimens are available under a secondary tab labeled “Other Options.” The primary screen shows median survival benefit and guideline concordance but not comparative quality-of-life trade-offs unless additional screens are opened. Patient preferences are documented in a free-text field that does not alter the visual ranking of options. While this configuration may streamline information display, it can narrow the perceived space of legitimate choice. When personal values are recorded but not visibly integrated, patients may feel that their priorities are acknowledged procedurally but not substantively, attenuating their sense of agency in a decision that carries significant life consequences.
Scenario B (Higher dignity; consistent with Level 6–7): The interface displays multiple guideline-concordant treatment options side by side, alongside expected survival range, common adverse effects, visit frequency, estimated out-of-pocket cost range, and uncertainty intervals. Before ranking options, the system prompts structured input regarding patient priorities (for example: maximizing longevity, minimizing hospitalization, preserving cognitive function, limiting financial burden). The comparative display dynamically adjusts based on those inputs. The patient experiences the decision process as responsive to their stated priorities.
5.3. Moral injury example: Workflow scheduling system
Scenario A (Lower dignity; consistent with Level 3–4): The system assigns standardized visit lengths based on diagnosis codes. If visit duration exceeds preset thresholds, the encounter is flagged as an “efficiency variance” in monthly productivity summaries. Performance dashboards display individual throughput metrics relative to departmental averages. Quality-of-care indicators are tracked separately, not integrated into the primary efficiency score. Although intended to improve efficiency, this configuration can create sustained tension when clinicians determine that relational or clinical complexity warrants additional time. This misalignment may contribute to moral distress and, in some cases, moral injury, particularly when clinicians feel responsible for outcomes but lack authority to recalibrate the constraints (Reference Dean, Talbot and DeanDean et al., 2019).
Scenario B (Higher dignity; consistent with Level 6–7): The scheduling system assigns baseline visit durations but with a clinician-controlled adjustment option within the interface. Clinicians can document the reason for extending visit times under predefined categories such as clinical complexity, patient comprehension needs, or care coordination. Efficiency metrics are displayed alongside care-quality indicators, including follow-up rates and patient-reported understanding. Governance language specifies that documented clinical adjustments are not penalized in productivity evaluations. Periodic review processes assess whether standardized durations remain appropriate across patient populations. In this configuration, operational goals remain but are mediated through structures that acknowledge professional judgment. By reducing friction between ethical obligation and workflow constraint, the system lowers the likelihood that clinicians experience policy in conflict with professional identity.
6. Discussion and implications
DCD offers a structured way of examining how AI systems shape institutional practice and everyday experience by making visible how design decisions position individuals within system processes. It integrates three analytic dimensions (individual, relational, systemic) with five core design principles, while the Dignity Spectrum in AI Design functions as a comparative tool for assessing how concrete system configurations align with those commitments. The framework thereby broadens the criteria in responsible AI approaches available to designers and researchers, adding a dimension of analysis concerned with how systems shape human treatment and institutional positioning. At the same time, dignity should not be understood as something that can be maximized in simple ways in all situations. In some cases, constrained discretion may be justified. For example, a clinical decision-support system might temporarily block a medication order when a high-probability interaction risks serious patient harm. In such circumstances, the relevant dignity question is not whether authority is constrained, but whether the intervention is justified, proportionate to the level of risk, and aligned with clear pathways for review and professional input. A system that communicates the basis for its recommendation and acknowledges professional judgment may preserve dignity even while limiting immediate discretion. DCD is intended to clarify these trade-offs rather than to dictate how they should be settled across all contexts.
This framework also opens avenues for further research. The Dignity Spectrum in AI Design has not yet been evaluated for reliability among design teams or examined systematically across sectors. Future studies could investigate how consistently practitioners interpret and apply the categories, including inter-rater reliability among designers, and how dignity-based assessments interact with established evaluation frameworks. Additional case studies beyond healthcare may surface tensions that require refinement of the model or reveal context-specific variations in how dignity is understood.
As AI systems become more deeply embedded across institutional settings, careful examination of how they structure influence and responsibility becomes increasingly important. Considering dignity as a sociotechnical concern invites closer attention to how design choices shape lived experience within organizations. In doing so, it expands ongoing conversations about responsible AI to include not only what systems achieve, but also how they reorganize relationships and decision environments in practice, and what that means for the people who live and work within them.
