Human-Machine Symbiosis Framework to Understand Human Deception

Abstract

Advancing our understanding of deception is a national priority. Researchers have studied extensively the role of nonverbal behavior in recognizing deception during face-to-face interactions. Past studies have repeatedly concluded that humansÑeven expert interrogatorsÑrarely do better than chance in detecting deceit, with estimates not varying far from an accuracy rate of 54 percent. Even the state-of-the-art screening techniques developed by the TSA, known as Screening Passengers by Observation Techniques (SPOT), have come under serious scrutiny. A public statement from the American Civil Liberties Union (ACLU) called SPOT ÒunscientificÓ and ÒunreliableÓ. What makes this problem seemingly intractable? In this proposal, the PI envisions the phenomenon of deception as a 3-dimensional cube where the dimensions are stakes, questioning techniques, and realism. The PI proposes to conduct five unique experiments consisting of 800 interactions using a semi-automated video-conferencing framework. The scenarios of the interaction involve participants being dishonest in a fun game for more money (low-stake deception) to scenarios where participants may encounter substantial negative consequences if caught lying (high-stake deceptions). PI proposes to model the unique aspects of the interrogation (e.g., establishing rapport, explain, engage, closure, evaluation etc.) using a Partially Observable Markov Decision Process (POMDP). The observations include facial expressions and language spoken (verbal expression). The technique will involve use of reinforcement learning to learn the states of the interrogations independently towards providing real-time feedback to the interrogator. The validation of the proposed framework will involve running a controlled experiment with participants as well as practitioners to determine whether a computer-aided interrogation system works better than a human interrogation.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 14, 2019
Source ID
W911NF1910029

Entities

People

  • Mohammad Hoque

Organizations

  • Army Contracting Command
  • United States Army
  • University of Rochester

Tags

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Educational Psychology
  • Sensor Fusion and Tracking Systems.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms