Dorsal raphe nucleus enkephalin peptide modulates behavioral preference.

Artificial intelligence (AI) systems are increasingly integrated into human decision-making across domains such as smart city construction, scientific research, financial risk assessment, medical diagnostics, and autonomous driving. By leveraging large-scale data and advanced learning algorithms, these systems can enhance decision processes and, in some contexts, improve decision quality and efficiency. However, the real-world benefits of AI depend not only on technical performance but also on whether people are willing to adopt and appropriately rely on AI recommendations. Appropriate trust is essential to the acceptance and effective deployment of AI systems. Trust shapes whether people follow, ignore, or over-rely on AI recommendations, often beyond what can be explained by objective system performance alone. Miscalibrated trust poses clear risks: insufficient trust can lead to underuse of beneficial AI decision support, whereas excessive trust can produce uncritical acceptance of flawed recommendations, potentially resulting in harmful outcomes. As AI systems gain autonomy in high-stakes settings, understanding how trust forms—and how it can be calibrated—remains an important research concern. One difficulty in forming trust is that AI decision-making can be opaque. Machine-learning systems can be difficult to interpret because their internal decision logic may be inaccessible, technically complex, or hard to map onto human-understandable reasons. Explainable-AI research has therefore treated interpretability as relevant to whether users can decide when to trust a model’s predictions. Trust in automation is also shaped by information about system performance, process, and purpose, as well as by experience with system behavior over time. In value-laden decision contexts, an advisor’s observable choices can be especially informative because they reveal how competing priorities are weighted.