Knowledge Operationalization from iNformal To EXecutable Models

Abstract

Forty years ago, Fred Brooks argued that the hard part of building software is “the specification, design, and testing of this conceptual construct, not the labor of representing it and testing the fidelity of the representation.” Since then, we have made spectacular progress in the labor of representation: high-level languages, model-driven engineering, cloud platforms, CI/CD pipelines, and, most recently, generative AI able to produce large parts of an implementation from a textual description. Yet the essential complexity Brooks identified has barely moved. Going from informal knowledge about what a system is for, whom it serves, and the context in which it operates to a conceptual model formal enough to be analyzed, executed, and evolved remains a dominant source of cost, defects, and failed projects.

Part of the reason, this talk argues, is that conceptual models systematically leave out the dimensions of the world that are hardest to formalize. Complex software systems operate within, and must account for, up to four dimensions: engineered systems, natural and biological processes, individual humans, and social structures such as teams, organizations, and institutions. Modelling practices, notations, and tools are mature for the first dimension, adequate for the second, and largely silent on the last two, even though this is often where the purpose of the system lies. A conceptual model with high fidelity in one dimension and none in another does not simply capture less; it can produce systems that are confident, actionable, and wrong.

Digital twins make this gap particularly visible, because their purpose is to represent a system. They will therefore serve as a running example throughout the talk: the smart building that exists for its occupants but is twinned primarily as plant equipment, or the software delivery process dominated by coordination but twinned as pipeline telemetry. Drawing on more than three decades of industrial and academic experience in software engineering, modelling and model-based engineering, DevOps, and digital twins, the talk will argue that dimensional scoping is a requirements and conceptual-modelling activity rather than a data or implementation activity. Fidelity, validation, drift, and decision authority must be defined for each dimension.

The talk will then examine how generative AI reshapes this picture. Generative AI reduces accidental complexity faster than any previous technology and can help elicit and structure informal human and social knowledge. However, it also makes it easier than ever to produce confident answers about dimensions that have never been modelled. The talk concludes with open challenges and deliberately provocative claims about what the conceptual-modelling community should stop—and start—doing.

Short bio

Francis Bordeleau is a professor at École de technologie supérieure in Montréal, where he has been based since 2018. Since 2021, he has led the Kaloom–TELUS ÉTS Industrial Research Chair in DevOps. His research focuses on improving industrial DevOps processes and engineering digital twins for different application domains, combining conventional and AI/ML technologies and approaches. He has more than 30 years of experience in software engineering, software process improvement, and model-based engineering, working with academic and industrial partners worldwide. Before joining ÉTS, he was Product Manager of Software Development at Ericsson, founder and CEO of Zeligsoft, and an assistant professor at Carleton University. He has also served on the organising and programme committees of numerous international conferences and workshops.