The Impact and Feasibility of Self-Confidence Shaping for AI-Assisted Decision-Making

TL;DR

This work addresses appropriate reliance in AI-assisted decision-making by shifting focus from AI-centric explanations to shaping human self-confidence. It formalizes a self-confidence shaping framework within a Judge-Advisor System, using a self-confidence function $F_\boldsymbol{\phi}$ to integrate human factors into decisions. Key findings show potential performance gains up to $\approx$ $47.6\%$ under under-reliance and $\approx$ $51.6\%$ under over-reliance in ideal conditions, a self-confidence predictor with ~${67}\%$ accuracy leveraging user and task traits, and a significant but modest link between sentiment and self-confidence ($r\approx0.221$, $p<0.001$) that motivates text-based interventions. The results demonstrate feasibility of sentiment-modulation strategies to shape self-confidence and outline a path toward deployment in high-stakes domains like finance, where calibrated human-AI collaboration can reduce error and improve outcomes.

Abstract

In AI-assisted decision-making, it is crucial but challenging for humans to appropriately rely on AI, especially in high-stakes domains such as finance and healthcare. This paper addresses this problem from a human-centered perspective by presenting an intervention for self-confidence shaping, designed to calibrate self-confidence at a targeted level. We first demonstrate the impact of self-confidence shaping by quantifying the upper-bound improvement in human-AI team performance. Our behavioral experiments with 121 participants show that self-confidence shaping can improve human-AI team performance by nearly 50% by mitigating both over- and under-reliance on AI. We then introduce a self-confidence prediction task to identify when our intervention is needed. Our results show that simple machine-learning models achieve 67% accuracy in predicting self-confidence. We further illustrate the feasibility of such interventions. The observed relationship between sentiment and self-confidence suggests that modifying sentiment could be a viable strategy for shaping self-confidence. Finally, we outline future research directions to support the deployment of self-confidence shaping in a real-world scenario for effective human-AI collaboration.

Figures (4)

• Figure 1: Conceptual illustration of the two-round procedure: 1st round, independent judgment; 2nd round, final team decision with AI recommendation.
• Figure 2: Box plots of sentiment scores for excerpts grouped by the assigned sentiment. Welch’s t-test indicate a highly significant difference between Positive and Neutral (***, $p\mkern-4mu<\mkern-4mu0.001$) and a marginally significant difference (*, $p\mkern-4mu<\mkern-4mu0.1$) between Negative and Neutral.
• Figure 3: Violin plots showing the distribution of accuracy for machine learning and rule-based approaches across 10 seeds. Each violin depicts the kernel density of accuracy values. The box within indicates the interquartile range and the horizontal line marks the median.
• Figure 4: Box plots of self-confidence $c$ for the three task‐sentiment variants. Welch’s t-test indicates a highly significant difference between Positive and Neutral (***, $p\mkern-4mu<\mkern-4mu0.001$) and a highly significant difference (***, $p\mkern-4mu<\mkern-4mu0.001$) between Negative and Neutral.