Enhancing Transparency and Trust in AI: the Role of Explainable AI and Visualization Techniques

The concepts of Explainable AI (XAI), and visualization methods are crucial in improving the interpretability of the ML models. With AI solutions being used in important industries including healthcare, finance, and autonomous, there is a need to explain the results produced by the models. The following survey focuses on the perspective of visualization in XAI to better illustrate the mechanisms by which models function and which are understandable to users. We present and compare several types of visualizations that might be applied for the given model: model-specific and model-agnostic ones as well as more complex approaches such as saliency maps and counterfactual visualization. The paper also considers the primary use cases for XAI and visualization across the contexts and responds to concerns, including scalability, ethicality, and the dilemma of the explainable accuracy vs. interpretability. Last, it presents future work findings in evaluation metrics, real-time system incorporation, and generative AI for visualization.