Experimental economics for machine learning—A methodological contribution on lie detection

Experimental economics for machine learning—A methodological contribution on lie detection

We explore how experimental economics and machine learning (ML) can benefit each other. While economics has long used technology for data collection and analysis, ML now offers a chance for economics to shape technology — especially in areas like lie detection, where algorithms aim to uncover private information.

To show this, we replicate the well-known “Lies in Disguise” experiment with a twist: participants are recorded while lying, but remain unobserved by the experimenter. This setup maintains privacy while enabling algorithm training.

Despite being monitored, participants show lying behavior similar to the original study. Our modified design allows for individual-level analysis and enables the development of a lie detection model with 67% accuracy — demonstrating how experimental economics can create better data for future AI research.

Citing

@article{bershadskyy2024experimental,
  title={Experimental economics for machine learning—a methodological contribution on lie detection},
  author={Bershadskyy, Dmitri and Dinges, Laslo and Fiedler, Marc-Andr{\'e} and Al-Hamadi, Ayoub and Ostermaier, Nina and Weimann, Joachim},
  journal={PloS one},
  volume={19},
  number={12},
  pages={e0314806},
  year={2024},
  publisher={Public Library of Science San Francisco, CA USA}
}