“Worker-in-the-loop” AI beefs up job safety
Workers are a vital element in construction processes, and their safety is paramount. Yet more than 60,000 fatal injuries occur every year in construction projects globally. In the U.S., there are at least 1,000 fatalities and 200,000 non-fatal injuries annually, with days away from work costing nearly $6 billion in lost production and income.
In a world where technology is embedded throughout all communities, and where the boundaries between humans and technology are shrinking considerably, the question in construction becomes this: How can the construction community harness technology and artificial intelligence (AI) to ensure that its essential employees can work safely, learn rapidly, and communicate effectively?
Construction sites also are changing quickly, and workers soon will need to interact with co-bots, or collaborative robots, and communicate with drones. Are we ready for this change? The answer is no. Construction communities need to better understand how these technological advances are set to revolutionize our industry, and to mitigate negative perceptions, such as those involving privacy concerns and other ethical questions in addition to safety issues.
To narrow these knowledge gaps, my research team and I monitor the neuro/psychophysiological responses of workers in real time — and design and implement data-driven models that predict human behavior — to augment workers’ interactions with their physical environment on current and future job sites and to keep them safe. Our studies have expanded fundamental safety knowledge around human error detection to transform the identification, assessment and control of erroneous worker decisions.
This innovative research represents a substantive departure from the status quo, by proposing novel pathways for proactive injury prevention in the current and future construction industry. It challenges the traditional paradigm of construction safety, which has concentrated on reacting to hazards, minimizing unsafe physical conditions.
Instead, we focus on workers who may be influenced by cognitive biases, exhibited as systematic information-processing shortcuts that lead to judgment errors and risky decisions. Despite all safety protections in place, these human factors — such as cognitive failure, momentary lapse, overreliance on trust, or risk compensation — can significantly encourage worker risk-taking behavior and increase the likelihood of injuries or fatalities.
We use wearable technologies to capture a worker’s neurophysiological and psychophysiological responses, combined with biomechanical data, in a simulated mixed-reality environment. We harness this “worker-in-the-loop” AI to address human error; monitor workers’ decision dynamics; predict at-risk behavior; and provide real-time, adaptive feedback.
Our methods include monitoring workers’ visual attention and search strategy through eye-tracking technologies; perceived risks, using photoplethysmography (PPG), an optical technique that provides indicators of blood circulation and heart rate, and galvanic skin response (GSR), which measures sweat gland activity to capture emotional intensity; and brain activity and decision dynamics, using functional near-infrared spectroscopy (fNIRS), a noninvasive technology for measuring brain oxygenation changes.
With appropriate AI models, a worker’s decision dynamics can then be translated into automated control signals in a physical construction environment to provide real-time feedback or warnings for workers about unsafe behavior triggers detected. When opportunities for risk arise during a worker’s interactions with the physical environment, the control system foreseeably will be able to prompt the person to make optimal and informed decisions.
This well-developed, highly-adaptive AI and personalized training can be used to promote active learning and enhance the skill performance of the next-generation infrastructure workforce, at individual and team levels.
Our interdisciplinary research lies at the intersection of civil engineering, cognitive and behavioral psychology, data science, and computer science. It explores multiple aspects of technology and applied science, and it suggests engineering solutions and behavioral interventions to respond to current and future challenges in the construction community and with complex projects.
These proposed improvements using AI and technological innovations are integral to developing critical infrastructure systems and enhancing occupational safety. They will provide a crucial validation measure to confirm the effectiveness of training programs. In addition, they will build a foundation for advanced accident causation modeling and the design of effective injury prevention practices that can protect thousands of workers and save billions of dollars in lost productivity and wages every year.
Sogand Hasanzadeh, PhD
Assistant Professor, Lyles School of Civil Engineering, and Construction Engineering and Management
Member, Purdue Engineering Initiative in Data and Engineering Applications
College of Engineering, Purdue University
