Abstract
Objective: This review explores the trustworthiness of multimodal artificialintelligence (AI) systems, specifically focusing on vision-language tasks. Itaddresses critical challenges related to fairness, transparency, and ethicalimplications in these systems, providing a comparative analysis of key taskssuch as Visual Question Answering (VQA), image captioning, and visual dialogue.Background: Multimodal models, particularly vision-language models, enhanceartificial intelligence (AI) capabilities by integrating visual and textualdata, mimicking human learning processes. Despite significant advancements, thetrustworthiness of these models remains a crucial concern, particularly as AIsystems increasingly confront issues regarding fairness, transparency, andethics. Methods: This review examines research conducted from 2017 to 2024focusing on forenamed core vision-language tasks. It employs a comparativeapproach to analyze these tasks through the lens of trustworthiness,underlining fairness, explainability, and ethics. This study synthesizesfindings from recent literature to identify trends, challenges, andstate-of-the-art solutions. Results: Several key findings were highlighted.Transparency: Explainability of vision language tasks is important for usertrust. Techniques, such as attention maps and gradient-based methods, havesuccessfully addressed this issue. Fairness: Bias mitigation in VQA and visualdialogue systems is essential for ensuring unbiased outcomes across diversedemographic groups. Ethical Implications: Addressing biases in multilingualmodels and ensuring ethical data handling is critical for the responsibledeployment of vision-language systems. Conclusion: This study underscores theimportance of integrating fairness, transparency, and ethical considerations indeveloping vision-language models within a unified framework.