Explainable Intelligence for Ethical Human-Centric AI: Trust Modeling Using Gradient Boosting

The accelerated adoption of artificial intelligence in human decision-making processes has highlighted the importance of human-centric and ethical artificial intelligence systems that are transparent, trustworthy, and accountable. A key part of adoption of such systems is trust, which determines how humans value, trust and interact with AI assisted decisions. In this paper, we propose an explainable and ethical machine learning framework to address this challenge, i.e., trust modeling. The research uses the Human-AI Interaction dataset that describes behavioural, cognitive, and ethical characteristics of human trust in AI systems. An Explainable Gradient Boosting model for classifying trust in low, medium, and high is developed which allows intuitive and interpretable trust assessment. The classification accuracy of the proposed approach is 92 % overall, which shows the power in classification based on the complex data of mixed types of interactions. To ensure transparency and ethical accountability, explainability mechanisms of SHAP-based feature attribution are incorporated to provide both global- and instancelevel feature attribution information for trust prediction. The results underline the effectiveness of a combination of highperformance ensemble learning with explainable intelligence, which provides a reliable and human-centric solution to the deployment of ethical AI in trust-sensitive applications.