Sander Renes Receives NWO Funding for Automating Economic Behaviour Experiments with AI Agents

Sander Renes is one of the 60 researchers receiving funding from the SSH Open Competition XS of the Dutch Research Council. These grants, amounting to a maximum of €50,000, are intended to support promising, innovative, and risky initiatives within the Social Sciences and Humanities (SSH). With the research project ‘Large Language Model Multi-Actor Tool to Automate Economic Experiments’, Renes, along with co-applicant Rutger van Bergem, aims to simplify economic and behaviour experiments—often costly and time-consuming—by replacing human participants with AI agents.

About the project

In many economic design problems, you want to try multiple versions of a solution. This can be challenging to achieve. Renes explains, "Experiments are costly and sometimes not even feasible. The setting might be too difficult for random participants, the task could take too long for a simple experiment, or there might be ethical objections to certain interventions. We can mitigate these limitations by replacing human participants with AI agents trained to behave like human participants."

The first experiments Renes and Van Bergem plan to test this with are the ones they are currently concluding in the IBEX lab at TPM. "In these experiments, we are testing a decision-making mechanism designed to reduce Not In My Back Yard (NIMBY) problems in spatial planning. Testing different variants with groups of people requires numerous sessions, making the research expensive and time-consuming." By employing AI agents instead of people, the researchers can try all variations in a few minutes and use the collected data to identify the most interesting ones to conduct in the lab with real participants. This allows for exploring a much larger portion of the design space.