Beyond Algorithms: A Multifaceted Exploration of Trust, Confidence, and Blame in Generative AI-Assisted Advice
通过四个实验场景比较用户对生成式AI工具与人类税务专家的反应，发现用户更偏好人类专家，AI版本或成本不影响信心，但人机互补模型能增强信心，且用户倾向于为AI错误自责。
SYNOPSIS This study examines user interactions with a hypothetical generative AI tool, “TaxAssistAI,” compared with a human tax expert (CPA) through four experimental scenarios. Our findings reveal a persistent preference for human expertise, even as generative AI tools become increasingly prevalent. Notably, the cost or version of the AI tool did not significantly influence user confidence or willingness to act, suggesting that users evaluate AI differently from human advisors. Moreover, a complementary human-AI model boosted confidence in the advice provided, reinforcing the potential for collaborative decision-making approaches. However, participants demonstrated a tendency to internalize blame following incorrect AI advice, showing a continued reliance on AI despite errors. These insights contribute to a deeper understanding of trust, confidence, and blame attribution in AI-assisted tax advisory, offering important implications for the integration of generative AI in professional advisory contexts and expanding decision support systems (DSS) research.