Optimizing Human Decision-Making through Explainable Artificial Intelligence A Behavioral and Technical Analysis

Abstract

Background: The problems of opacity and lack of transparency in an artificial intelligence (AI) system continue to arise as more of them are increasingly applied in making crucial decisions in various industries. This can be addressed with the help of Explainable Artificial Intelligence (XAI) which can improve the ability to interpret models, instill trust, and align AI outputs and human reasoning. Yet, the effectiveness of various formats of explanations in terms of their behavioral and technical efficiency is not sufficiently studied.

Aim: The purpose of this experiment is to determine the impact of different XAI methods on human beings in terms of accuracy of decision-making, trust, and cognitive load, as well as determining the technical fidelity, stability, and interpretability of these models.

Procedure: Mixed-methods research design was used that included a sample of 180 healthcare, finance, and education professionals. The participants used AI systems with various types of explanation: none and textual (SHAP) explanations and visual contrastive (counterfactual) explanations. Hand eye (accuracy, trust, response time and cognitive load) and technical measures of model fidelity, stability, and complexity were recorded. The preferences and mental models by the user were further probed using qualitative interviews.

Results: Visual and contrastive explanations have enormous effects on decision-making (M = 86.5%) and trust (M = 4.33/5) as compared to textual and non-explanatory types. SHAP provided the maximum model fidelity, whereas counterfactuals received the most points in terms of interpretability and cognitive effectiveness. They found contrastive and simplified explanations better and ones that matched their pattern of thinking.

Conclusion: Both behavioral and technical dimensions of optimization of XAI systems improve the way decisions are made, trust, and engagement of consumers. The XAI design of the future needs to be built based on cognitive regulations and needs to be suited to the context of the user in order to be both ethically and practically superior