Conflicting incentives between local courts and central authorities fragment knowledge and block the scaling of innovations across Italy's judiciary, undermining potential productivity and equity gains from shared data and AI tools. The KADVM mapping points to targeted policy levers—aligned metrics, standardized data, and formal sharing channels—to unlock national returns from AI investments.

Main Finding

The study finds that knowledge dynamics within the Italian judicial organization are shaped by persistent, dysfunctional feedback loops between two competing purposes—local court efficiency and national-level uniformity of practice. These tensions, driven by conflicting goals, rules, and incentives, distort knowledge flows and undermine the development, sharing, and deployment of critical knowledge assets across the judiciary.

Key Points

• Two primary and conflicting knowledge purposes:
  • Local efficiency: courts optimize processes to meet immediate productivity/performance targets.
  • National uniformity: central authorities aim to harmonize practices and standards across courts.
• Dysfunctional feedback loops arise when local incentives and informal competition (between courts) inhibit horizontal knowledge sharing and vertical dissemination of innovations.
• Organizational fragmentation and formalization coexist: formal rules exist but are implemented in a fragmented, decentralized environment, producing counterproductive interactions among local innovation, vertical control, and horizontal diffusion.
• Unbalanced incentives (reward structures, performance metrics) encourage local optimizations that can be at odds with system-wide goals, leading to underdevelopment or misdeployment of shared knowledge assets.
• The study demonstrates the practical usefulness of the Knowledge Asset Dynamic Value Map (KADVM) for mapping and analyzing complex knowledge systems in public-sector, knowledge-intensive organizations.

Data & Methods

• Empirical basis: a comprehensive body of qualitative data collected across the Italian judicial organization (coding and triangulation of organizational data sources).
• Analytical approach:
  • Construction of a hierarchical objective tree to map critical knowledge assets and associated performance objectives across organizational levels.
  • Development of a feedback-loop systemic model using the Knowledge Asset Dynamic Value Map (KADVM) method to reveal dynamic interactions, reinforcing/balancing loops, and failure points.
• Outcomes: identification of specific feedback loops and leverage points where interventions could realign knowledge dynamics with intended performance outcomes.

Implications for AI Economics

• Diffusion and scaling of AI in public-sector systems require aligned incentives and clear governance; misaligned local vs. national objectives can block effective AI adoption and reduce aggregate returns to AI investments.
• Fragmentation and informal competition impede creation and sharing of high-quality training data, labeled datasets, and best-practice models—key knowledge assets for AI—raising costs and increasing heterogeneity in AI performance across jurisdictions.
• Feedback-loop distortions can produce adverse selection and local optimization of AI tools (e.g., narrowly tuned automated decision aids) that increase systemic risk, bias, or inefficiency when generalized across the national system.
• KADVM-style systemic mapping is valuable to AI policymakers and implementers: it helps identify where to invest (data infrastructure, standards, shared ontologies), which incentives to redesign (performance metrics, rewards for sharing), and how to stage diffusion (pilot–scale–national rollout) to maximize social returns.
• Recommended policy levers informed by the study:
  • Align performance metrics and incentives to reward both local efficiency and contribution to shared knowledge assets.
  • Create formalized channels and resources for horizontal knowledge sharing (platforms, curated datasets, model registries, federated learning frameworks).
  • Standardize data, annotation, and evaluation protocols to reduce integration costs and improve model portability.
  • Use systemic models (like KADVM) ex ante to anticipate unintended feedbacks and to design phased AI deployment strategies that mitigate risks from decentralization.
• Economic consequences: better-aligned knowledge dynamics can increase productivity gains from AI, reduce duplication of effort, and improve equity in service quality across regions; failure to address these dynamics will lower realized returns on public AI investments and risk uneven distribution of benefits.