Multimodal Explainable Artificial Intelligence: A Comprehensive Review of Methodological Advances and Future Research Directions

AI systems need better transparency: The paper emphasizes that while AI has achieved remarkable results across many tasks, these systems often lack transparency and trustworthiness—a gap that explainable AI (XAI) research aims to address by providing meaningful explanations of how AI models make decisions.

Multimodal explanations are increasingly important: Multimodal XAI methods, which use multiple types of data (such as text, images, and audio together), require specialized explanation approaches that go beyond single-modality systems, reflecting how modern AI systems often process diverse information sources simultaneously.

Explanation timing matters: The paper identifies that explanations can be generated at different processing stages within AI systems, suggesting that when and how explanations are produced affects their usefulness for understanding model reasoning—an important consideration for practitioners implementing XAI solutions.

Evaluation of explanations is complex: The research highlights that assessing whether explanations are actually good and trustworthy requires multiple evaluation metrics, indicating that there's no single standard way to measure explanation quality, which poses challenges for organizations trying to verify their AI systems are truly explainable.

The field still faces significant open challenges: The paper concludes by discussing ongoing challenges and future research directions in multimodal XAI, suggesting that while progress has been made, achieving truly transparent and trustworthy AI systems in complex, real-world scenarios remains an active area of development.