Most day laborers have experienced some form of exploitation, with many reporting wage theft. Bureau of Business researchers have designed an AI-based methodology to help address the problem. Photo Credit: The Salt Lake Tribune
Published December 8, 2025
The working conditions of a day laborer are precarious.
A 2019 survey found that 86% of day laborers had experienced some form of exploitation, with 66% reporting wage theft. Yet laborers continue to accept jobs because partial pay is better than no pay at all.
According to Matt Kammer-Kerwick, director of the Bureau of Business Research at the IC² Institute, past interventions to reduce wage theft have proven largely unsuccessful. Law enforcement policing is rarely effective, because day laborer hiring and payment are informal, undocumented, and only partially observable.
Taking a Behavioral Approach
Kammer-Kerwick has been tackling the issue of wage theft and other forms of exploitation since 2016, when he and his team began interviewing day laborers in Houston and Austin. Continued on-the-ground research proved difficult, so the team turned to computer simulations to understand the laborer/employer dynamic.
Kammer-Kerwick’s research into wage theft focuses not on enforcement strategies, but, instead, on system change; that is, “How do we get people — both laborers and employers — to change their behavior?”
Kammer-Kerwick’s latest research, conducted with Evan Aldrich, and funded by the National Science Foundation, uses an AI-based reinforcement learning model to build a framework to test hypothetical wage theft interventions. The process is described in a recent article published n PLOS Complex Systems.
Testing Interventions in Simulated Environments
Laborers who experience wage theft rarely report it, as they imagine a time-consuming reporting process and little chance of ever receiving their missing pay. But what if laborers perceived a better chance of success — would they report their missing wages (to a city’s day laborer center, for example) more often? And would more regular reporting perhaps change the behavior of employers?
To explore these questions, the researchers used agent-based modeling to characterize and predict interactions between system participants (laborer and employer). The simulated environment allows the researchers (and future users such as policy-makers) to improve their understanding of interventional strategies before applying them in the real world.
Researchers deployed both a single-agent version of the reinforcement learning model that examines the decision-making of a laborer and then expanded to a multi-agent formulation that encompasses decisions by both the laborer and the employer.
Using a method called sensitivity analysis, the researchers found that with a modest increase in the likelihood of a successful reporting outcome, laborers learn to report theft and employers learn not to steal.
The researchers stop short of outlining specific ways to ensure the “successful reporting outcome”, though Kammer-Kerwick says, “A worker center can provide useful training and a forum that encourages ‘good neighbor’ behavior .” And he adds, “There are simple ways to ‘nudge’ laborers towards reporting, such as using labor center advocates and public service announcements to make sure laborers know how and where to file a report.”
Future Research Application: Community Health Clinics
Kammer-Kerwick admits that systems interventions (as opposed to “blunt force” policing) can take time to create real behavioral change. However, this research suggests that taking a behavioral approach, and leveraging AI simulations to test interventions and outcomes, holds promise for reducing wage theft and discouraging other illicit behaviors.
The biggest takeaway for Kammer-Kerwick is that he and Aldrich have arrived at an AI-based methodological framework which can be applied to other complex sociological systems. They cite community health clinics (where patients often present with a web of social, economic and health challenges) as one such system. The simulation/reinforcement learning methods could be used to develop educational interventions to increase patients’ adherence to treatment plans and appointment schedules.
Kammer-Kerwick explains: “The point of these methods — the simulation and the reinforcement learning — is that, when applied in the real world, they will allow policy-makers to develop dynamic policies that can adapt as a complex system changes. These methods will help us design strategies to disrupt bad behaviors like wage theft — and they can also be leveraged to encourage positive behaviors like cooperation.”