Voicebot and Deception

How to cite

Zhang, Shanshan. 2022. "Voicebot and Deception." AEA RCT Registry. AEARCTR-0010368. https://doi.org/10.1257/rct.10368-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Shanshan Zhang*

* Primary investigator

Registry ID
AEARCTR-0010368
DOI
https://doi.org/10.1257/rct.10368-1.0
First registered
Last updated
Status
Completed
Trial dates
2022-04-01 – 2022-08-01
Countries
United States of America (California )
Registry record
http://www.socialscienceregistry.org/trials/10368

Abstract

With technological advancements, interacting with a voicebot has become increasingly more common. Attitudes that promote or prevent dishonesty in human-human interactions may not carry over into human-machine interactions. There is little evidence of how interaction with a voicebot might affect dishonest behavior. We test whether individuals are more or less likely to lie to a voicebot than to a human using an online experiment in which subjects perform a coin-toss task and report the coin-toss outcomes. We vary the reporting channels, i.e., reporting via text, reporting to a voicebot, reporting to a sophisticated vociebot (AIbot), reporting to a human. For all the voice interaction treatments, we design an homogeneous online platform to mimic the real-world setting of voicebot interactions that varies only the robotic voice and pre-recorded human voice. We also vary the gender of voice for each treatment, i.e., feminine and masculine voice. We compare the difference between reporting to a machine (text, voicebot, or AIbot) and reporting to a human with pre-recorded voice. We also compare reporting via text and reporting via voice. We also analysis the effect of voice gender and the effect of sophistication of voicebot.

Intervention

We run an online experiment via Prolific. Subjects perform a coin-toss task and report their outcomes. Subjects are randomly assigned to one of the four treatments. In the text treatment, subjects are told to report their coin-toss outcomes to a chatbot by answering a questionnaire. In the voicebot treatment, subjects are told to report the outcomes to a voicebot. In the AIbot treatment, subjects are told to report the outcomes to a sophisticated voicebot. In the human treatment, subjects are told to report the outcomes to an experimenter. At the end of the experiment, we also ask questions that measure the subject's attitudes towards AI and chatbot, trust, and demographics.

Experimental design

We run an online experiment via Prolific, an online platform for recruiting online. We design an online voice chat platform for this expeirment. We use Qualtrics and the designed online platform (voice interaction platform) to collect answers. The study consists of three parts. In the first part, participants will be given an instruction on the study. In the second part, participants will toss a coin 7 times, record the results of coin tosses on a paper, and report the results to a human voice or a voicebot. Participants will be told that they will get 30 cents for each head they toss. The link to the voicebot or the human voice will be provided in Qualtrics. During the interaction with the voicebot or the human voice, the participants' voices will be recorded. After this, the participants will be redirected to the Qualtrics and answer questions in the third part. Participants will get an additional $1 for answering questions in this part. The questions include measurements of participants' perceptions of the Chatbot, measurements of their personality, and demographic information.

Design details

Randomization method
Randomization was done in qualtrics
Randomization unit
individual
Clusters
350
Observations
350
Arms
50 for each treatment
Primary outcomes
The number of reported heads
Secondary outcomes
The acoustic feature of the recorded voice

Cited by

No citation data available for this registration yet.

Keywords