The Ethical Knowledge Gap: Dispersed Knowledge, Sensemaking Failures, and Epistemic Dependence

Abstract: Ethical software development remains stubbornly difficult despite two decades of normative frameworks, professional codes, and participatory methodologies. This paper offers a diagnostic rather than prescriptive contribution: it argues that the persistent gap between ethical intention and ethical implementation is a structural epistemic condition, not primarily a failure of will, education, or normative guidance. Three independently sufficient mechanisms interact to produce what I call the ethical knowledge gap -- a condition in which the knowledge required for ethically informed decision-making is systematically unavailable at the point of decision, even when the organization as a whole possesses it. First, drawing on Hayek's (1945) analysis of dispersed knowledge and its organizational extensions, the paper establishes that ethically relevant knowledge in software development is constitutively distributed across roles, largely tacit, and -- unlike efficiency-related knowledge -- unsupported by any spontaneous aggregation mechanism analogous to the price system. Second, an interpretive deficit, analyzed through Weick's sensemaking framework and the literature on framing and epistemic cultures, renders developers unable to recognize the ethical significance of what they know: the sensemaking apparatus of engineering culture makes technical decisions intelligible while systematically obscuring their ethical dimensions. Third, a credibility attenuation, analyzed through the social epistemology of testimony and epistemic dependence, discounts developers' observations as they cross organizational role boundaries, so that hybrid judgments combining technical detail with ethical assessment lose their epistemic force.

• The paper introduces a formal model that quantifies the ethical gap by identifying dispersed knowledge, sensemaking deficits, and credibility attenuation as key mechanisms.
• The paper critiques established frameworks like principal-agent theory and value-sensitive design, showing they fail to address systemic epistemic barriers in software development.
• The paper advocates for a translation layer mechanism to restructure organizational design, enabling effective aggregation of tacit ethical insights into decision-making.

Structural Origins and Dynamics of the Ethical Knowledge Gap in Software Development

Diagnostic Framework: Three Mechanisms Underlying the Ethical Gap

The paper "The Ethical Knowledge Gap: Dispersed Knowledge, Sensemaking Failures, and Epistemic Dependence" (2604.24160) critically reframes the persistent difficulty of ethically aligned software development as an epistemic, rather than merely motivational or normative, problem. It identifies three independently sufficient and interacting mechanisms—organisational dispersal of knowledge, systemic interpretive deficit, and credibility attenuation—that together constitute what is termed the "ethical knowledge gap." This gap is defined as the structural, cumulative deficit that keeps ethically relevant knowledge systematically unavailable at the point of decision, even when present somewhere within the organization.

Dispersed Knowledge: Drawing on Hayek's theory of dispersed knowledge, the paper demonstrates that ethically relevant information is distributed across organizational roles and is often tacit and context-dependent. Developers possess granular implementation knowledge; managers, product owners, and domain experts hold contextual knowledge—user needs, regulatory environments, sector-specific vulnerabilities. Ethical judgments typically require access to both, but aggregation fails because no spontaneous organizational mechanism exists, analogous to price signals in markets, to bring these fragments together.

Sensemaking Deficits: The interpretive apparatus governing engineering culture constrains the categories through which technical work is rendered intelligible. Technical decisions are understood as engineering trade-offs and are routinely detached from their ethical significance, since ethical framing is not embedded in the sensemaking categories available to practitioners. The paper formalizes the hermeneutical deficit as the lack of pathways linking implementation details to ethical salience, which persists regardless of individual awareness or virtue.

Credibility Attenuation: The social epistemology of organizational testimony reveals that hybrid ethical-technical insights are systematically disaggregated, reframed, or dismissed as they cross domain boundaries. Developers’ technical testimony is authoritative only in its technical domain, while ethical implications are often discounted unless voiced by recognized ethical authorities. Testimonial efficacy weakens through role-based epistemic dependence, reducing organizational uptake of cross-domain ethical insights.

Critique of Existing Institutional Frameworks

The paper provides a rigorous critique of established interventions and frameworks for ethical software development, including principal-agent theory, knowledge-based theories of the firm, and value-sensitive design (VSD).

Principal-Agent Theory: The unidirectional information asymmetry and incentive misalignment presupposed by principal-agent models are ill-suited to the structural epistemic conditions found in software development. The paper argues that ethically relevant knowledge is not simply hidden by agents but is genuinely fragmented, partly tacit, and resistant to centralization or contract-based aggregation.

Knowledge-Based Theory of the Firm: While Grant, Tsoukas, and Foss's distributed-knowledge frameworks capture the dispersal mechanism, ethical knowledge introduces additional interpretive and epistemic authority challenges absent from efficiency-oriented knowledge coordination. The paper highlights that ethical knowledge is not only dispersed but structurally unsupported by aggregation mechanisms, and further compounded by sensemaking and credibility failures.

Value-Sensitive Design: VSD’s top-down, participatory methodologies operate at the abstraction level of values hierarchies but fail to penetrate the implementation-level emergence of ethical concerns. Participatory approaches are necessary for aggregation but insufficient without mechanisms for ethical legibility and testimonial authority at the local decision-making level.

Formal Sketch of the Ethical Knowledge Gap

The paper introduces a formal model to make explicit the cumulative structure of the ethical gap. Let $D$ denote development decisions, $I(d)$ and $X(d)$ denote sets of implementation and contextual facts, and $R(d) \subseteq I(d) \times X(d)$ represent ethically relevant relations. Distributed access among agents, interpretive limitations, and attenuated testimonial uptake sequentially lower the proportion of actionable ethical relations $U(d)$ compared to the total $R(d)$. The ethical gap $G(d) = 1 - \frac{|U(d)|}{|R(d)|}$ quantifies the shortfall and underscores its cumulative, structural character.

Implications for Institutional Design and Practice

Translation Layer as Structural Solution

The paper posits that only institutional mechanisms targeting all three epistemic blockages can alleviate the gap. It specifies the need for a "translation layer"—an institutional practice explicitly designed to (i) enrich developers’ sensemaking categories, (ii) restructure epistemic-dependence architectures to grant credible cross-domain uptake, and (iii) enable targeted, bidirectional knowledge aggregation rather than comprehensive centralization.

Organizational design must integrate:

• Bidirectional knowledge flow: Aggregating local implementation and contextual knowledge at decisive risk points.
• Sensemaking enrichment: Changing the interpretive categories through which technical work is understood, not merely informational supplementation.
• Trust over reliance: Institutional restructuring to treat technical actors as credible ethical speakers.
• Targeted aggregation: Distributed, context-sensitive mechanisms to collect and integrate tacit knowledge.

Value Type Differentiation

The gap operates differently across techno-generic and domain-specific values. For techno-generic values (e.g., fairness in ML), sensemaking interventions suffice; for domain-specific values, knowledge integration dominates. Cross-product cases—where both types interact—demand full translation-layer implementation.

Agile Methods and Limitations

Agile development methods partially approach the translation-layer vision via flat hierarchies and collaborative ceremonies. However, they void full structural correction by failing to address interpretive deficits and relying on procedural reflection without sensemaking enrichment. Product Owners function as translation points but face single-point-of-failure risks and cannot bridge the full epistemic gap for hybrid ethical decisions.

Numerical and Structural Claims

The formalization supports a strong claim: single-vector interventions—ethics training, reporting channels, participatory formats—address, at most, one mechanism, leaving the structural gap intact. The gap is measurable: even in scenarios where the organization possesses six of ten ethically relevant relations, only one may be recognized and acted upon, leaving a 90% gap.

The paper also asserts a symmetry: the same epistemic structures that obscure ethical harms prevent the realization of ethical opportunities. Ethical underperformance is as much a function of the gap as harm production.

Theoretical and Practical Implications

This framework fundamentally alters the locus of ethical responsibility from individual actors’ virtue or awareness to institutional epistemic design. It suggests future progress in ethical AI and software development must focus on reorganizing sensemaking categories, authority structures, and aggregation mechanisms, rather than adding more normative guidance or procedural compliance. It provides a foundation for evaluating the sufficiency of interventions and for anticipating new organizational challenges as AI systems increase in complexity and ethical stakes.

Conclusion

The paper provides a rigorous diagnostic of the ethical knowledge gap as a structural epistemic feature of software development organizations. By situating ethical implementation failures in the interaction of dispersed knowledge, sensemaking limitations, and credibility attenuation, it rejects motivational or normative deficiency explanations. It offers institutionally actionable design principles and a formal framework for measurement, demonstrating why single-vector interventions fail. The ethical gap is not contingent, but constitutive, and closing it requires comprehensive epistemic restructuring.