Beyond Recommender: An Exploratory Study of the Effects of Different AI Roles in AI-Assisted Decision Making

TL;DR

Three AI roles are examined: Recommender, Analyzer, and Devil's Advocate, and their effects across two AI performance levels are evaluated, showing each role's distinct strengths and limitations in task performance, reliance appropriateness, and user experience.

Abstract

Artificial Intelligence (AI) is increasingly employed in various decision-making tasks, typically as a Recommender, providing recommendations that the AI deems correct. However, recent studies suggest this may diminish human analytical thinking and lead to humans' inappropriate reliance on AI, impairing the synergy in human-AI teams. In contrast, human advisors in group decision-making perform various roles, such as analyzing alternative options or criticizing decision-makers to encourage their critical thinking. This diversity of roles has not yet been empirically explored in AI assistance. In this paper, we examine three AI roles: Recommender, Analyzer, and Devil's Advocate, and evaluate their effects across two AI performance levels. Our results show each role's distinct strengths and limitations in task performance, reliance appropriateness, and user experience. Notably, the Recommender role is not always the most effective, especially if the AI performance level is low, the Analyzer role may be preferable. These insights offer valuable implications for designing AI assistants with adaptive functional roles according to different situations.

Figures (5)

• Figure 1: The interfaces of three AI roles. (a) Recommender: In this interface, the AI directly presents a recommendation along with its explanation. (b) Analyzer: The AI assists in evaluating the pros and cons of each option. By default, it displays an overview of all options and highlights two highly probable choices to help users refine their selection. Users can explore detailed AI analyses for any specific option by clicking on its corresponding tab. (c) Devil's Advocate: This interface features the AI providing arguments against the user's current selection, irrespective of whether it aligns with the AI's own prediction. Additionally, the AI suggests an alternative option, which represents the most viable alternative to the user's initial choice.
• Figure 2: The task accuracy in three AI role conditions. (a) When AI performance is high. (b) When AI performance is low. Error bars represent standard error. (+: $p$<0.1, *: $p$<0.05, **: $p$<0.01, ***: $p$<0.001)
• Figure 3: A detailed analysis of participants' final decision accuracy based on the correctness of human initial prediction and AI recommendation. (a) When AI performance is high. (b) When AI performance is low. Error bars represent standard error. (Since categorizing participants' prediction data into different situations reduces the sample size, we did not perform statistical analysis.)
• Figure 4: Participants' reliance and the appropriateness of their reliance on AI. (a) When AI performance is high. (b) When AI performance is low. Error bars represent standard error. (+: $p$<0.1, *: $p$<0.05, **: $p$<0.01, ***: $p$<0.001)
• Figure 5: Participants' user experience. (a) When AI performance is high. (b) When AI performance is low. Error bars represent standard error. (+: $p$<0.1, *: $p$<0.05, **: $p$<0.01, ***: $p$<0.001)