Transparency and Accountability
Princy Pappachan, Massoud Moslehpour, Ritika Bansal, Mosiur Rahaman · Advances in computational intelligence and robotics book series · 2024
The rapid growth and application of AI has ushered in ground-breaking technologies like LLMs. However, these innovations also bring significant challenges related to transparency and accountability, especially considering the complex neural network architectures and vast training datasets. This chapter thus explores the journey of AI from rule-based systems to the current ML and deep neural network, identifying the black box problem that plagues the decision-making process in LLMs. The chapter introduces strategies for enhancing transparency using explainable AI (XAI) frameworks to address these issues, offering practical solutions to quantify and improve transparency. Accountability is also emphasized through a detailed protocol for assigning responsibility across AI development phases, reinforced by ethical auditing and reporting methodologies. Mathematical equations and frameworks are also presented to compute transparency scores and accountability measures, providing organizations with structured, actionable guidelines for building transparent, fair, and ethical AI systems.