INTERPRETATION OF ARTIFICIAL INTELLIGENCE DECISIONS: HOW TO ENSURE TRANSPARENCY AND COMPREHENSIBILITY FOR END USERS

Authors

DOI:

https://doi.org/10.31891/2219-9365-2024-79-27

Keywords:

artificial intelligence, interpretability, transparency, explainable AI, LIME, SHAP, visualization, accountability, trust, end-user comprehension, decision-making transparency

Abstract

This article delves into the pressing issues of interpretation and transparency in artificial intelligence (AI) decision-making systems, particularly in critical domains such as healthcare, finance, legal decision-making, and risk management. As AI models become increasingly complex—especially with the rise of neural networks and deep learning models—the decisions produced by these systems are often opaque to end users. This opacity can lead to decreased trust, limited adoption, and challenges in practical implementation, particularly in sectors where decisions directly impact human lives and well-being. Ensuring that users can understand, evaluate, and, if necessary, question AI decisions is thus becoming a vital part of responsible AI development.

The article provides an in-depth review of foundational methods for achieving AI interpretability. Key techniques covered include Local Interpretable Model-agnostic Explanations (LIME), which offers localized insights into specific model outputs, and SHapley Additive exPlanations (SHAP), a game-theoretic approach that quantifies the contribution of each feature to a given decision. In addition to these, the article examines global interpretive methods that aim to reveal overarching patterns and rules that guide the AI model’s behavior.

Moreover, the article highlights the pivotal role of visualization tools—such as feature importance charts, heatmaps, and temporal visualizations—that help make complex AI processes more understandable and intuitive for non-technical users. Visualization techniques are instrumental in conveying how certain features or data points influence model outcomes, thus bridging the gap between technical model outputs and user-friendly insights.

The article also critically assesses the strengths and limitations of current interpretative approaches, such as the computational demands of SHAP and the localized nature of LIME, which may not generalize across datasets. Finally, the article discusses future directions for advancing interpretability, including the development of real-time interpretive methods, hybrid approaches combining local and global explanations, and interactive user interfaces that allow end users to query AI models for explanations. By addressing these challenges and evolving interpretative frameworks, AI systems can be better aligned with the values of transparency, accountability, and trustworthiness, making them more suitable for sensitive, high-stakes applications.

Published

2024-08-29

How to Cite

SERDYUK Ю. (2024). INTERPRETATION OF ARTIFICIAL INTELLIGENCE DECISIONS: HOW TO ENSURE TRANSPARENCY AND COMPREHENSIBILITY FOR END USERS. MEASURING AND COMPUTING DEVICES IN TECHNOLOGICAL PROCESSES, (3), 212–216. https://doi.org/10.31891/2219-9365-2024-79-27