The Digital Fish Simulation Project is a machine learning-driven innovation initiative aimed at optimizing fish production processes by developing simulation-based tools for digital modeling of fish behavior. Set within the context of the seafood industry's increasing demand for sustainability, precision, and efficiency, the project operates in a domain characterized by limited prior expertise and significant biological variability. As such, knowledge creation must be dynamic, interdisciplinary, and continuously evolving. This article explores the project from a knowledge management perspective, applying Nonaka and Takeuchi’s SECI model of organizational knowledge creation and the Manulab–Industry Competence Building Process as theoretical frameworks. The study focuses on how structured knowledge management practices, embedded from the earliest stages, enable successful navigation of uncertainty and foster innovation in machine learning-based product development. Workshops were employed as the primary research method, functioning not only as operational project checkpoints but also as structured environments for facilitating interdisciplinary collaboration, surfacing tacit knowledge, externalizing insights, and iteratively refining concepts and prototypes. Through this action research methodology, the study captures the dynamic interplay between knowledge creation activities and project advancement. Preliminary findings indicate that, unlike traditional automation projects where knowledge transfer follows linear trajectories, machine learning-driven innovation demands multiple, parallel SECI cycles running simultaneously across different knowledge domains. Sustained involvement of industry partners from the earliest stages, coupled with interdisciplinary teamwork among engineers, biologists, simulation specialists, and students, has proven critical for effective knowledge creation. Trial-and-error learning, real-time adaptation, and continuous feedback loops emerged as key mechanisms for accelerating organizational learning. The article concludes with recommendations for future research, emphasizing the need to explore structured models for parallel SECI cycle management, interdisciplinary knowledge transfer optimization, and the institutionalization of trial-and-error learning processes. The findings contribute to a deeper understanding of how knowledge management practices must evolve to support sustainable, human-centric innovation in emerging machine learning-driven industrial environments.
Hansen et al. (Fri,) studied this question.