Overcoming the data crisis in AI-powered construction

By: Margaret Harding McGill 

Date: May 29, 2026

A shortage of standardized, high-quality data is a major bottleneck in using artificial intelligence to revolutionize American construction safety and efficiency.

To address this challenge, researchers have developed OpenConstruction, an open-source catalog designed to provide a central index for the construction industry to discover and reuse high-quality AI training datasets.

Why it matters: Construction is one of the most dangerous and least digitized industries in the U.S. AI can provide automated safety monitoring, real-time progress tracking, and predictive maintenance, but these systems are only as smart as the data used to train them.

Catch up quick: Construction data is trapped in proprietary silos, lacks consistent labeling, and rarely captures the complex reality of an active job site. The industry lacks the centralized, "FAIR" (Findable, Accessible, Interoperable, and Reusable) data repositories that have fueled AI breakthroughs in other fields.

Action: Researchers from Carnegie Mellon University and six other institutions systematically reviewed nearly 20 years of construction visual data. They synthesized their findings into a central data index for the construction industry called OpenConstruction. The OpenConstruction catalog is live and being updated by a growing community of researchers. 

Policy takeaways: The project revealed critical data gaps, such as a lack of 3D visuals and unified terminology, that could be addressed by policies such as:

  • Standardizing data language with unified industry ontologies (adapting existing systems like OmniClass) so AI can accurately identify site hazards across different regions and projects.
  • Incentivizing data trusts through the creation of centralized, long-term repositories managed by professional societies that provide secure, anonymized data for research while protecting proprietary company information.
  • Mandating ethical AI pipelines by requiring federally funded infrastructure projects to use privacy-preserving data collection (e.g., face blurring, skeletal mapping) to foster worker trust and enable safer data reuse.

The big picture: This research is part of CMU’s broader mission to create human-machine harmony in infrastructure. By fixing the underlying data crisis, we can move from reactive safety measures to proactive AI systems that save lives and reduce waste.

Faculty expert