Institution profile

Naval Postgraduate School

Academic institutionnorthamerica · us
Official website
Research library36linked papers
Opportunities0open roles
Selected work

Representative Papers

Unsafe at any AUC: Unlearned Lessons from Sociotechnical Disasters for Responsible AI

Jul 15, 2026

This study addresses a critical gap in AI safety research, which has predominantly emphasized the reliability of technical components while overlooking systemic risks inherent in sociotechnical systems. Drawing lessons from historical large-scale human-made disasters, this work positions social and organizational dynamics as first-order engineering considerations in AI safety design, thereby challenging conventional evaluation paradigms centered on technical metrics such as AUC. By integrating sociotechnical systems analysis frameworks, interdisciplinary theories, and in-depth case studies, the research uncovers recurrent mechanisms through which AI systems replicate past failures. It further offers actionable recommendations to advance responsible AI from a component-level focus toward a holistic, system-level approach, enhancing capabilities in risk awareness, accountability tracing, and organizational coordination.

0 citationsRead paper

Markov Decision Process Approximation Methods for Water Distribution Network Inspection and Maintenance: A Case Study of the U.S. Virgin Islands

Jul 05, 2026

This study addresses the challenge of insufficient subsurface pipeline condition awareness in data-scarce regions, such as the U.S. Virgin Islands, which hinders effective inspection and maintenance decisions. The authors propose a repair-oriented decision-making framework for water distribution network maintenance, formulating the problem as a discounted Markov decision process coupled with high-fidelity hydraulic simulation. Relying solely on readily available system-level observations, the framework infers latent pipe conditions by establishing a unique mapping between observable system dynamics and failures in specific pipe segments, thereby enabling virtual sensing without segment-level instrumentation. The approach explicitly captures heterogeneous failure characteristics across pipe segments and generates state-dependent optimal maintenance policies, demonstrating the feasibility of dynamic-system-based, resource-efficient inspection planning under constrained conditions.

0 citationsRead paper
Recent publications

Latest Papers

Unsafe at any AUC: Unlearned Lessons from Sociotechnical Disasters for Responsible AI

Jul 15, 2026

This study addresses a critical gap in AI safety research, which has predominantly emphasized the reliability of technical components while overlooking systemic risks inherent in sociotechnical systems. Drawing lessons from historical large-scale human-made disasters, this work positions social and organizational dynamics as first-order engineering considerations in AI safety design, thereby challenging conventional evaluation paradigms centered on technical metrics such as AUC. By integrating sociotechnical systems analysis frameworks, interdisciplinary theories, and in-depth case studies, the research uncovers recurrent mechanisms through which AI systems replicate past failures. It further offers actionable recommendations to advance responsible AI from a component-level focus toward a holistic, system-level approach, enhancing capabilities in risk awareness, accountability tracing, and organizational coordination.

0 citationsRead paper

Markov Decision Process Approximation Methods for Water Distribution Network Inspection and Maintenance: A Case Study of the U.S. Virgin Islands

Jul 05, 2026

This study addresses the challenge of insufficient subsurface pipeline condition awareness in data-scarce regions, such as the U.S. Virgin Islands, which hinders effective inspection and maintenance decisions. The authors propose a repair-oriented decision-making framework for water distribution network maintenance, formulating the problem as a discounted Markov decision process coupled with high-fidelity hydraulic simulation. Relying solely on readily available system-level observations, the framework infers latent pipe conditions by establishing a unique mapping between observable system dynamics and failures in specific pipe segments, thereby enabling virtual sensing without segment-level instrumentation. The approach explicitly captures heterogeneous failure characteristics across pipe segments and generates state-dependent optimal maintenance policies, demonstrating the feasibility of dynamic-system-based, resource-efficient inspection planning under constrained conditions.

0 citationsRead paper