Explainable Artificial Intelligence and Big Language Models: Transparent and Reliable Decision Support Systems
Batyr Orazbayev Yerlan Izbassarov Lyailya K urmangaziyeva
AbstractIn recent years, the transparency and explainability of artificial intelligence systems' decision-making processes have gained a great deal of importance, both technically and ethically. Explainable artificial intelligence (XAI) plays an important role in the development of more trustworthy systems, especially in areas such as healthcare, law, and economics. This paper discusses model-independent explanation methods such as LIME and SHAP and evaluates their applicability to classical machine learning models. Moreover, the recent proliferation of large language models (LLMs) such as GPT, PaLM, LLaMA, etc., has led to a different approach in XAI, both in terms of their powerful text generation capabilities and the need for the explainability of their output. The methods developed to analyze the decision logic of LLMs have been evaluated through approaches such as chain of thought, attention visualization, and in-context explanation. However, due to the generative nature of LLMs, their accuracy, stability, and capacity to provide confidence to the user are still open to debate. This paper compares classical XAI methods and annotation approaches applied to LLMs and provides examples to illustrate how explainability can be achieved in LLM-based decision support systems. The findings show that technical and end users can better understand the reasons for LLM outputs