Unveiling the Black Box: Explainability and trust in AI-based banking systems

Sammanfattning

Introduction The integration of AI systems into the banking sector presents both significant opportunities and potential risks. As AI handles high-stakes decisions such as credit scoring or fraud detection, the lack of transparency, referred to as the black box phenomenon, has emerged as a critical concern. This thesis examines how explainability can be achieved in these complex systems and how it impacts and influences user trust. Research question In response to these challenges, the thesis seeks to answer the following research question: What are the challenges, opportunities, and effective strategies for achieving explainability and trust in AI-based banking systems? Method To answer the research question, a qualitative literature review was conducted. Following guidelines based on academic sources for a systematic review, relevant articles and case studies were identified through key words such as: explainability, trust, finance, AI models, black box AI, explainable AI in banking, autonomous systems, trust in AI and high risk. An analysis was conducted on the articles to connect them in relation to the thesis’s key concepts. Result Through systematic literature review and thematic analysis of 17 academic and institutional sources, the study identifies three key challenges in achieving explainability: technical and organizational barriers, psychological factors affecting stakeholder trust, and gaps between regulatory demands and real-world implementation. Practical strategies used by banks, like model-agnostic XAI tools, human-AI hybrid systems and conceptual alternatives such as envelopment, are also highlighted in the analysis. Discussion The results suggest that explainability alone isn’t sufficient to establish trust in AI-based banking systems. Instead, a combination of technical transparency, regulatory compliance and human oversight should be the way to achieve trust between human and machine. The study indicates that strategies for achieving explainability should correspond to the complexity and risk level of each application. Explainability should not be treated as a technical add-on but rather as a foundational component of responsible AI deployment. Further research is needed on integrated explainability methods, as well as clearer regulatory guidance to ensure a safe deployment of AI within the banking sector.