Abstract for: AI for System Dynamics: Mapping Progress across Six Modelling Stages
Current research shows that AI techniques like NLP and ML offer promising support for traditional SD modelling. However, most studies focus on isolated tasks and lack a structured framework to assess AI’s role across all SD stages. This paper addresses that gap by systematically reviewing how AI supports all SD modelling stages (Martinez-Moyano and Richardson, 2013), mapping current practices, highlighting limitations, and offering recommendations for integrating AI into SD workflow. This study conducts a literature review to investigate how modern AI have been integrated into the stages of SD modelling. The literature was collected from a combination of peer-reviewed journals, conference proceedings, and preprints published 2015- 2025 with keywords such as "System Dynamics", "SD", "Artificial Intelligence", "AI", "Natural Language Processing", "Large Language Models", "Causal Extraction", "Causal Loop Diagram". The selection process itself adapting from PRISMA 2020 flow diagram with adjustment. Strong focus on conceptualization via AI-supported CLD building Problem identification rarely addressed except Akhavan & Jalali (2024) Promising progress in simulation and parameter tuning using ML and GPT Gaps in later stages: implementation, learning strategy design Challenges: traceability, usability, stakeholder trust Fragmented Integration= No end-to-end AI-supported SD modelling pipeline exists. Most focus on one stage only and require human validation Lack of Human-AI Collaboration=Few studies incorporate user feedback Inconsistent Evaluation Methods, Most rely on visual inspection, lacking formal evaluation Model Completeness Unclear=No study discusses. Raises concern on stopping criteria and AI-supported model validity. Lack of Traceability= Only one provide reasoning and relevant source text, others are AI black boxes. ChatGPT and grammarly for grammar checking, sentences and paragraph flow checking