Why Science Needs Causal Problem Modelling

Science now operates in an era of unprecedented data, computational power, and AI (artificial intelligence)-driven ideation, yet scientific inquiry seemingly remains largely fragmented across disciplinary boundaries. Econometrics, machine learning, nonlinear dynamics, evolutionary economics, sociology, psychology, political science, and AI engineering all study causal processes, but each relies on distinct conceptual vocabularies, representational logics, and methodological assumptions. This fragmentation can limit the capacity of researchers and policymakers to communicate, to integrate evidence, and to design effective interventions in complex adaptive systems.&nbsp; <div> <br> </div> <div> We argue that contemporary science lacks a unifying upstream discipline that takes the causal problem itself, rather than data, statistical tools, or domain-specific models, as the primary unit of analysis. We introduce Causal Problem Modelling (CPM) as this missing discipline. CPM formalises what researchers implicitly construct but rarely articulate, the causal architecture, the feasible intervention set, the data-generating situation, the identification surface, and the counterfactual constraint set. These primitives provide a domain-general language for representing causal systems that can be applied uniformly across the natural and social sciences.&nbsp; </div> <div> <br> </div> <div> CPM extends the Sussex lineage of innovation studies, evolutionary economics, sectoral systems, and institutional theory by giving these traditions a formal causal structure they historically lacked. It also unifies insights from nonlinear dynamics, behavioural science, interpretive social science, and metascience, clarifying where different methods contribute and where identification is inherently limited. By distinguishing what CPM is, and is not, the framework avoids conflation with potential outcomes, structural causal models, econometric identification strategies, machine learning, and complexity models, positioning CPM as an upstream integrative discipline rather than a competing methodology.&nbsp; </div> <div> <br> </div> <div> CPM also provides a principled foundation for AI integration in science and policy. Treating AI systems as structural instruments for architecture discovery, dynamic simulation, and counterfactual exploration, CPM enables responsible, transparent, and ethically bounded use of AI in high-stakes domains. We illustrate how CPM can guide intervention design, policy evaluation, and scientific governance in systems characterised by feedback, emergence, heterogeneity, uncertainty, and institutional constraints.&nbsp; </div> <div> <br> </div> <div> In establishing CPM as a field in its own right, this article offers a coherent framework for designing interventions, integrating heterogeneous evidence, coordinating interdisciplinary teams, and governing complex socio-technical systems. CPM provides a foundation for the next phase of scientific practice, a unifying discipline for understanding and shaping the causal systems that define the modern world. </div>