[Watch on YouTube.]
Artificial Intelligence (AI) is revolutionizing the construction materials testing industry, and Idaho DOT's Quality Program Manager Mike Copeland is leading the charge. In a recent AASHTO Resource Q&A podcast episode, Copeland shared his journey of incorporating artificial intelligence into quality assurance processes, demonstrating how these powerful tools are reshaping traditional workflows while also introducing new challenges that require careful consideration.
Copeland's journey began with a straightforward problem: how to extract testing data locked away in PDF documents. Like many transportation departments, Idaho DOT had transitioned from paper to digital documentation, but without a Laboratory Information Management System (LIMS), their valuable data remained difficult to access for analysis. Traditional optical character recognition (OCR) offered limited success, but when generative AI models with vision capabilities emerged, Copeland discovered they could extract structured data from PDFs with remarkable accuracy, maintaining the proper relationships between values and retaining context that OCR typically lost.
This capability transformed data extraction from a tedious, error-prone process of manual data entry into a quick, reliable operation. What once might have taken 15-30 minutes per document now takes mere seconds. The implications are profound - by investing just a few hours into developing these tools, Copeland estimates the Idaho DOT could save thousands of manhours annually across their testing operations, freeing staff to focus on more valuable analysis rather than data entry.
Beyond data extraction, Copeland has employed AI in numerous innovative ways. He created a chatbot that contains all of Idaho DOT's specifications, manuals, and research documents, allowing staff to quickly find answers to technical questions without hunting through multiple documents. An unexpected benefit emerged as the chatbot revealed conflicts between different manuals, helping identify inconsistencies that might otherwise go unnoticed. He's also developed tools that can automatically process testing data, create visualizations, and even analyze gyratory compactor data to identify equipment-based testing variability.
Perhaps most provocatively, Copeland has explored whether AI could predict material properties without physical testing. Using multilinear regression models built with AI assistance and trained on years of laboratory data, he demonstrated that bulk specific gravity values could be predicted within acceptable precision limits without performing the actual test. While this raises exciting possibilities for efficiency, it also introduces critical concerns about the potential for fraudulent practices in the industry.
This leads to what may be the most important revelation from Copeland's work: our quality assurance systems need fundamental reconsideration in the AI era. Traditional QA programs were built around paper-based workflows, with chain of custody focused on physical samples. In today's digital environment, data security becomes equally critical. Copeland noted that with proper prompting, AI tools could be used to modify test results without changing metadata, creating plausible but fabricated data that might pass verification checks. This represents an evolving threat that quality programs must address.
Rather than viewing AI as inherently problematic, Copeland advocates for thoughtful integration paired with updated security measures. He suggests focusing on data provenance and chain of custody for digital information, increased independent verification testing, and implementing continuous improvement processes to stay ahead of rapid technological changes. He also notes that AI itself can be part of the solution, helping to identify suspicious patterns or anomalies in testing data.
The opportunities AI presents for standardization are equally compelling. By aggregating test methods from multiple organizations (DOTs, AASHTO, ASTM), AI could help identify unnecessary variations in procedures, potentially streamlining standards and eliminating arbitrary differences. This could lead to greater consistency across the industry while preserving essential regional adaptations.
As transportation departments and testing laboratories navigate this technology revolution, Copeland's experience offers valuable insights into both the tremendous potential and serious challenges AI brings to construction materials testing. The future clearly involves AI integration but requires thoughtful implementation that enhances rather than undermines quality assurance principles. Those who adopt these tools early, with appropriate safeguards, may gain significant advantages in efficiency and insight while helping shape responsible practices for the entire industry.
Important Note: Buzzsprout Cohost AI was used as a resource when creating this post.