In this work we present the development of an autonomous agent capable
of competing with humans in a deception-based game. The agent predicts
whether a given statement is true or false based on vocal cues. To
this end, we develop a game for collecting a large scale and high quality
labeled sound data-set in a controlled environment in English and Hebrew.
We develop a model that can detect deception based on vocal statements
from the participants of the experiment, and show that the model is
more accurate than humans.
We develop an agent
that uses the developed deception model and interacts with humans within
our deceptive environment. We show that our agent significantly outperforms
a simple agent that does not use the deception model; that is, it wins
significantly more games when played against human players. In addition,
we use our model to detect whether a statement will be perceived as
a lie or not by human subjects, based on its vocal cues.
Cite as: Mansbach, N., Neiterman, E.H., Azaria, A. (2021) An Agent for Competing with Humans in a Deceptive Game Based on Vocal Cues. Proc. Interspeech 2021, 4134-4138, doi: 10.21437/Interspeech.2021-83
@inproceedings{mansbach21_interspeech,
author={Noa Mansbach and Evgeny Hershkovitch Neiterman and Amos Azaria},
title={{An Agent for Competing with Humans in a Deceptive Game Based on Vocal Cues}},
year=2021,
booktitle={Proc. Interspeech 2021},
pages={4134--4138},
doi={10.21437/Interspeech.2021-83}
}