Benchmarking Generative AI Platforms and Deep Learning for Pavement Distress Diagnosis: A Case Study from Iraq and Türkiye

Timely pavement distress diagnosis is essential for resilient and sustainable road infrastructure, yet conventional visual inspection remains labor-intensive, subjective, and difficult to scale across urban networks. This study evaluates the potential of artificial intelligence to support faster and more consistent pavement maintenance decisions by benchmarking six generative AI platforms, namely Gemini, ChatGPT, Claude, Grok, Microsoft Copilot, and DeepSeek, against a YOLOv8-based object detection model for photo-based pavement distress diagnosis. A curated dataset of 21 field photographs from urban roads in Iraq and Türkiye was analyzed, covering eight distress categories, including rutting, plastic flow, fatigue cracking, longitudinal and transverse cracking, low-temperature cracking, potholes, settlement, and drainage-related defects. Each generative AI platform was assessed using a standardized prompt requiring distress classification, severity interpretation, root-cause analysis, and rehabilitation recommendations, while YOLOv8 was used for automated localized defect detection.

The results show that generative AI platforms provide valuable engineering interpretation but vary considerably in diagnostic reliability, with platform accuracies ranging from 25% to 100% across different image subsets. YOLOv8 produced 31 localized detections with confidence scores ranging from 26.1% to 83.7%, demonstrating its suitability for network-level screening. In contrast, language-based AI tools offered broader explanations related to material behavior, structural causes, and maintenance strategies. A self-recognition test further revealed that ChatGPT misclassified its own AI-generated fatigue-cracking image as block cracking, indicating that generative image production does not necessarily imply reliable diagnostic recognition. Overall, the findings indicate that object-detection models and generative AI should be viewed as complementary tools. A hybrid YOLOv8 and generative AI workflow is proposed to support scalable pavement monitoring, early intervention, and more sustainable infrastructure asset management.