Introduction to Generative AI
Generative AI as we know it is very prominent in daily life activities. Dr. Bo Xiao, a college of business faculty member specializing in Construction Informatics, AI Applications in Construction, and Construction Robotics research explores the ideas of Generative AI and its uses in the workspace.
According to Dr. Xiao, construction is an interesting field for AI research because it involves real-world problems such as safety, scheduling, cost control, quality management, and on-site decision-making. Due to the nature of construction zones, projects are unique and dynamic. Because of this, exploring these ideas for the implementation of Generative AI in construction zones becomes challenging.
When comparing Generative AI in its current form to its early stages of development, various ideas and new understandings of the topic come to light. Generative AI helps brainstorm topic ideas, knowledge retrieval, report generation, etc. Generative AI can be helpful to students interested in Building Information Modeling (BIM) course materials and/or provide help to practitioners to swiftly locate OSHA (Occupational Safety and Health Administration) safety regulations.
Dr. Xiao provides insight into the future of Generative AI, mentioning that it could become much more integrated into construction management and field operations ten years from now. “By combining BIM, sensors, robotics, scheduling data, and safety information, AI systems may evolve into intelligent project assistants that can proactively identify risks, recommend construction strategies, support resource allocation, and even collaborate with robotic systems on site.” (Dr. Xiao, 2026). To ensure AI practices are implemented correctly, it is important to understand all potential risks of this process and how AI systems can be grounded to avoid any unrestricted responses.
Implementational Risks of Generative AI in Construction
Due to the nature of construction zones being a high-stakes industry, inaccurate responses from AI systems can be detrimental. Generative AI may provide misleading and/or inaccurate responses leading to financial losses and even injuries. “In my research, I focus on grounding AI outputs in verified sources through retrieval-Augmented Generation (RAG). This means the system answers questions based on trusted materials such as course content, OSHA regulations, or project documents.” (Dr. Xiao, 2026). AI systems must provide correct citations in order for the users to trace where information is coming from. When environments become high-risk, human review and the implementation of validation mechanisms would still be necessary. This would allow AI to act as a support to professional decision-making rather than replacing it entirely.
Another concern includes bias and poor-quality training; the data poses as a concern when implementing AI systems in construction zones. AI models trained in one specific area of construction work would not operate well in situations that are unfamiliar to its training. Construction data frequently contains noise such as incomplete reports, unclear site images, or missing incident records. As a whole, these issues affect the reliability and overall fairness of Generative AI systems.
Dr. Bo Xiao’s Research
Dr. Xiao’s research includes Construction Education; AI-supported learning platforms created for BIM and other construction-related courses, and Construction Safety Knowledge Access; Retrieval-Augmented Generation (RAG) systems built around OSHA regulations to support safety-related information retrieval.
He studies the ways in which Generative AI can be integrated into multi-agent systems, learning science, and construction data management. He states that the AI systems he works with learn from verified domain-specific materials. E.g. course slides, PDFs, BIM educational content, OSHA safety regulations, construction safety documents, and structured question-answer datasets. “As these systems continue to enhance their learning in real-world construction projects, they can incorporate daily resorts, inspection records, BIM models, sensor data, and project management documents, ensuring these datasets are carefully cleaned, organized, and validated before being used.” (Dr. Xiao, 2026).
The Future of Generative AI in Construction
Fragmented knowledge remains one of the biggest risks. Information spread across specifications, drawings, reports, and personal experiences makes it difficult to access and reuse in an efficient manner. Dr. Xiao mentions that it is not enough to evaluate whether AI systems can produce fluent responses. But rather, whether it can improve decision-making and provide support to actual workflows. Currently, he is focusing on RAG frameworks, multi-agent architectures, citation verification, and instructional interaction design.
“In real construction environments, usefulness should also be measured by whether the system reduces errors, improves efficiency, supports safety awareness, and fits naturally into existing work processes.” (Dr. Xiao, 2026).
He aims to adapt existing large language models instead of training completely new AI models from scratch. Training foundation models would require large computational resources. “In construction, the bigger challenge is not usually the model itself, but how to connect the model with reliable domain knowledge.” (Dr. Xiao, 2026).
While Generative AI presents new opportunities for construction education, safety, and project management, its effectiveness ultimately depends on how it is developed and implemented. By balancing innovation with responsibility, Generative AI has the potential to become a valuable tool in shaping the future of construction management and workplace safety.