The Influence of Prior Discourse on Conversational Agent-Driven Decision-Making
Stephen Pilli, Vivek Nallur
TL;DR
The paper examines how the cognitive load from prior dialogue tasks in a task-oriented chatbot affects susceptibility to status-quo bias in subsequent decisions. Using a between-subjects design with simple and complex prior tasks across three Samuelson-style scenarios, the study demonstrates that higher task complexity generally shifts effect sizes toward the hypothesized bias, with a reliable bias detected in the Budget Allocation scenario under complex discourse ($p=0.002$). Across investments and college jobs, results show directionally similar trends but less robust or non-significant effects, while cognitive load increases response times and NASA-TLX dimensions, and recall remains high. These findings illuminate how conversational nudging can be sensitive to prior dialogue complexity, informing the design of digital decision aids and the ethical considerations of bias amplification in chatbots.
Abstract
Persuasion through conversation has been the focus of much research. Nudging is a popular strategy to influence decision-making in physical and digital settings. However, conversational agents employing "nudging" have not received significant attention. We explore the manifestation of cognitive biases-the underlying psychological mechanisms of nudging-and investigate how the complexity of prior dialogue tasks impacts decision-making facilitated by conversational agents. Our research used a between-group experimental design, involving 756 participants randomly assigned to either a simple or complex task before encountering a decision-making scenario. Three scenarios were adapted from Samuelson's classic experiments on status-quo bias, the underlying mechanism of default nudges. Our results aligned with previous studies in two out of three simple-task scenarios. Increasing task complexity consistently shifted effect-sizes toward our hypothesis, though bias was significant in only one case. These findings inform conversational nudging strategies and highlight inherent biases relevant to behavioural economics.