Designing and Evaluating Multi-Chatbot Interface for Human-AI Communication: Preliminary Findings from a Persuasion Task

Overview

• This study examines how the availability of multiple AI chatbots impacts human-AI communication, particularly in a persuasion setting aimed at promoting charitable donations.
• The researchers developed an online environment that enables multi-chatbot communication and conducted a pilot experiment using two AI chatbots - one representing Save the Children and one representing UNICEF - to promote charitable donations.
• The paper presents the development process of the multi-chatbot interface and preliminary findings from the pilot experiment, including analysis of qualitative and quantitative feedback.

Plain English Explanation

The rise of language models like ChatGPT has reshaped how humans communicate with AI. However, most research has focused on one-on-one interactions, leaving much to be explored about how humans interact with multiple AI chatbots at once.

This study looked at what happens when people communicate with not just one, but two AI chatbots - one representing the charity Save the Children, and one representing UNICEF. The researchers built a special online environment to enable this multi-chatbot interaction, and then ran an experiment where people were encouraged to donate to charity through conversations with the two chatbots.

The researchers share how they developed this multi-chatbot interface, and present some early findings from the experiment. They analyzed both the qualitative feedback (what people said) and quantitative data (measurable outcomes) to understand the impact of this multi-chatbot communication. The paper also discusses the limitations of the study and areas for further research.

Technical Explanation

The researchers designed an online environment that enabled participants to communicate with two separate AI chatbots - one representing Save the Children and one representing UNICEF - in a persuasion setting aimed at promoting charitable donations.

The chatbots were built using GPT-based language models, which are a type of large language model capable of engaging in human-like conversations. The researchers developed a custom interface to facilitate the multi-chatbot interaction, allowing participants to switch between conversing with the two chatbots.

In the pilot experiment, participants were prompted to engage in conversations with the two chatbots and were encouraged to donate to charity. The researchers collected both qualitative feedback (e.g., participant comments) and quantitative data (e.g., donation amounts) to analyze the impact of this multi-chatbot communication.

The paper presents the researchers' process for developing the multi-chatbot interface, as well as preliminary findings from the pilot study. The analysis suggests that the presence of multiple chatbots influenced the persuasion dynamics, with participants engaging in more empathetic and multi-modal interactions compared to typical one-on-one chatbot scenarios.

Critical Analysis

The paper acknowledges the limited scope of the pilot study and calls for further research to better understand the dynamics of human-AI communication in group settings. Some potential limitations include the small sample size, the specific persuasion context, and the use of only two chatbots.

Additionally, the researchers do not delve deeply into the ethical implications of using AI chatbots for persuasion purposes, such as concerns around transparency, human-to-human-to-human communication, and the potential for manipulation.

Future research could explore the impact of multi-chatbot communication in a wider range of contexts, examine the long-term effects on user engagement and behavior, and address the ethical considerations more thoroughly.

Conclusion

This study represents an important step in understanding the evolving dynamics of human-AI communication, particularly in group settings. The researchers' development of a multi-chatbot interface and their preliminary findings suggest that the presence of multiple chatbots can influence the persuasion dynamics in ways that differ from typical one-on-one interactions.

While further research is needed to fully explore the implications of this technology, this paper highlights the potential for multi-chatbot communication to shape various domains, from charitable giving to other persuasion-oriented applications. As AI language models continue to advance, understanding these emerging communication patterns will be crucial for designing ethical and effective human-AI interactions.