One Pager - Towards Persuasive Chatbots with LLMs
Working Title: Towards Persuasive Chatbots with LLMs
General Description
The primary goal of the research paper is to create and evaluate prompts to make a LLM-powered chatbot persuasive. The prompts are derived from various research publications from different domains and evaluated by a working prototype. The research follows the design science research method and makes a contribution to practice by providing a new and effective artifact for a real-world problem.
Research Objective/Questions
The main research question is how can we make a chatbot persuasive. The term chatbot set the limit of the research scope, which is the text-based conversational agents powered by LLMs. To alter the models is not easily feasible and the way to interact with LLM has predominantly shifted into prompt engineering, which is the design (noun) in context of DSR, hence this paper treats prompts als design artifacts.
Expected Outcomes
- Innovative artifacts to make a chatbot persuasive
- Examplary prompts to be implemented in a chatbot design
- A set of useful design practices for a chatbot with persuasive abilities
- #TODO: add more?
Necessary Data/Measurements
The artifacts (prompts) are derived from various literature resources. Many research papers and publications in the domain of persuasive communication, psychology and psychiatry have developed a wide range of techniques and factors which could be implemented in chatbot design.
Intended Analysis
This paper will try to follow a typical DSR iterative process with 6 steps or elements as shown in the followings.
#TODO: Demonstration/Evaluation need to be discussed
1. Problem Identification and Motivation
Problem statement: the rapid development of conversational AI (LLMs) fostered the incorporation of the chatbot into consumer products, business services and profession workflows. However, how to make a chatbot persuasive is not fully researched in the context of prompt engineering. This research paper tries to bridge the gap by creating and evaluating artifacts to make a LLM-powered chatbot persuasive.
2. Define Objectives for a Solution
Objectives of a solution: through text-based interaction, the chatbot should has the persuasive ability to drive the counter-part (the user) to the desired outcome, more specifically, to enable expected change of behaviors.
3. Design and Development
The off-the-shelf one: a table with persuasive technologies and prompts based upon.
A technical prototype can also be developed, this can either be an implementation of PROMISE framework, or a quick one based on OpenAI (assistent) API with steamlit to show off.
4. Demonstration
The artifact is used to demonstrate how it solves one or more instances of the problem. The demonstration serves as a proof of concept.
The table of prompts (off-the-shelf one): the demonstration can be some examples of conversations between chatbot and user.
The technical one: this can also be put online, hosted on streamlit, netify, etc.
5. Evaluation
This can involve empirical tests, simulations, case studies, analysis, or experiments.
#TODO: how to conduct experiment? Due to time limit, a small online experiment with surveys can be conducted, assuming a working prototype is developed. Or is it necessary?.
6. Communication
The process, its results, and its contributions are communicated to both academic and practical audiences (DC project).
Target Community
DSR community is the target, because this research paper is planned to be submitted to DESRIST 2024, the conference on Design Science Research. The theme of conference encourages the DSR community to contemplate DSR-studies from an innovative and pragmatically useful perspective for contributing to a resilient future. The paper could also potentially be interested to the community of Prompt Engineering and Conversational AI.
(Optional): Exemplary Literature or Model Papers from That Community
Notes mentioning this note
There are no notes linking to this note.