Openness is not intrinsically ethical: a formal relational ethics framework shows whether open data and models spur generativity or entrench monopoly depends on historical power relations, roles and trajectories. Policymakers should assess relational histories and dependencies—not apply blanket openness or closure rules.

Main Finding

The paper proposes a Generative Relational Ethics framework for evaluating expansion of public knowledge: formalize the relational structures and historical trajectories that make knowledge generative or extractive, and ground ethical interpretation in those historically-situated, role-dependent relational conditions rather than in fixed priors (e.g., “openness is always better”). Openness and restriction are neither intrinsically virtuous nor vicious — their ethical significance depends on the relational system, its history, and the roles and dependencies produced.

Key Points

• Distinction between relational syntax and normative interpretation:
  • The paper separates a formal description of how knowledge actors, relations, events and access change over time from the moral judgments applied to those changes.
• Formal representation:
  • Relational states are modeled as typed graph configurations (actors, roles, ties, permissions).
  • Histories of interaction and transformation are modeled as labelled 2-complexes linking successive relational boundaries (capturing trajectories, events, changes in control/dependency).
• Political‑economy interpretive resources:
  • Uses concepts like accumulation, appropriation, concentration, enclosure, dispossession, dependency, extraction, rent formation, institutional capture, and re‑enclosure to identify consequential transformations in public knowledge.
  • These categories do not carry invariant moral valence in the model, but they supply lenses for interpretation.
• Generative Relational Ethics:
  • Normative interpretation is emergent, historically generated, role-dependent, and revisable.
  • Ethical judgment is responsive to reasons furnished by the evolving relational context, not reducible to a fixed rule set or optimization procedure.
• Practical consequence:
  • Openness has no fixed ethical priority; under some historical-relational conditions restrictions protect generativity, in others restrictions enable dependency and enclosure.
  • Ethical evaluation must therefore attend to actors’ roles, power asymmetries, temporal trajectories, and institutional contexts.

Data & Methods

• Formal methods:
  • Typed graphs to represent relational states (nodes = actors/roles/knowledge objects; typed edges = relations such as access, control, reuse).
  • Labelled 2-complexes to represent histories and transformations (cells/labels encode events, transactions, changes in rights or dependencies, and their sequencing).
  • The formal representation preserves granularity — actors, roles, relations, events, and changes in access/control — suitable for mapping complex knowledge ecosystems.
• Interpretive/analytical methods:
  • Political economy analysis to classify and interpret transformations using categories like extraction, rent formation, and institutional capture.
  • Ethical theory concentrated on hermeneutic, contextualized judgment rather than algorithmic rule application.
• Empirical stance:
  • The paper is primarily conceptual/formal and analytic rather than empirical; it supplies a representational toolkit and an interpretive normative account rather than reporting measurement or case‑study data.

Implications for AI Economics

• Rethink “openness vs. closure” policies for AI:
  • Blanket prescriptions (e.g., fully open datasets/models as always ethical) are inadequate. Policy should assess how openness will interact with existing power relations, incentives, and institutional structures — sometimes targeted restrictions protect generativity (e.g., safeguard commons), other times they entrench monopolies.
• Measurement and monitoring:
  • Use relational representations (graphs, histories) to operationalize and measure knowledge capital flows: who provides data, who gains access, who extracts value, and how dependencies form (e.g., data-owner → platform → model → downstream lock‑in).
  • Develop metrics for concentration, dependency, and enclosure (e.g., centralities in access graphs, persistence of asymmetric control across historical steps).
• Anti‑trust and governance:
  • Antitrust, data-governance, and public‑infrastructure policies should consider dynamic, historical trajectories — not merely current market shares — because past transactions can create long-lived dependency and enclosure.
  • Remedies might include interventions that change relational trajectories (mandated interoperability, data portability with provenance, non‑exclusive licenses for foundational data/models, public funding of commons).
• Platform and model design:
  • Architects of data/model-sharing mechanisms should design for generativity: preserve provenance, enable recombination without enabling rent extraction, guard against re‑enclosure (e.g., by limiting transformative exclusivity).
  • Conditional openness regimes (time‑limited exclusives, staged releases, share-alike licensing) can be ethically preferable in some contexts.
• Labor and extraction:
  • Recognize how seemingly “public” data and annotation labor may be appropriated into private rent streams; policy and contracts should address compensation, attribution, and collective ownership where appropriate.
• Research directions for AI economics:
  • Empirically map knowledge ecosystems with the paper’s formal tools to identify where enclosure and dependency arise.
  • Simulate counterfactual institutional interventions (e.g., different licensing regimes) on generativity and distribution of rents.
  • Develop operational indices for generativity vs. enclosure usable in policy evaluation and regulatory impact analysis.

Limitations and cautions - The framework is a conceptual/formal proposal; operationalizing it requires empirical work to map real-world relations and calibrate measures. - Ethical judgment is intentionally left as context-sensitive and deliberative — the framework guides what to look at and how to interpret, but does not output fixed normative answers.