Artificial Intelligence in Sustainable Transportation Planning: Issues, State of the Art, and the Potential of Neurosymbolic AI
TL;DR
It is argued that neurosymbolic AI architectures offer a promising research direction by combining neural learning from heterogeneous mobility data with symbolic representations of behavioral rules, network constraints, as well as accessibility, equity, and environmental objectives.
Abstract
Artificial intelligence (AI) is changing the way we study transport systems, yet its contribution to sustainable transportation planning remains uneven. While AI-based methods have shown strong performance in operational tasks such as traffic prediction and signal control, their integration into strategic planning is still fragmented. This paper provides a narrative review of AI applications in transportation planning and, more specifically, on-demand modeling, arguably the most challenging side of transport planning, and will focus on three arbitrary field macro-aggregations: machine learning (ML), artificial neural networks (ANN), and neurosymbolic AI (NeSy). The findings show that AI methods can improve predictive performance and capture complex nonlinear mobility patterns; however, predictive accuracy alone is insufficient for planning practice. Strategic planning requires models that are interpretable, transparent, and able to accommodate expert-defined constraints. The European Union AI Act further reinforces this requirement by classifying AI systems used in critical infrastructure, including road-traffic management, as high risk; this creates a regulatory need that many current AI applications do not yet satisfy. The paper argues that neurosymbolic AI architectures offer a promising research direction by combining neural learning from heterogeneous mobility data with symbolic representations of behavioral rules, network constraints, as well as accessibility, equity, and environmental objectives.