A Review of Methods for Reducing Hallucinations in Generative Artificial Intelligence to Enhance Knowledge Economy(مقاله علمی وزارت علوم)
حوزههای تخصصی:
Generative Artificial Intelligence (AI) models, such as large language models, are transforming various sectors of the knowledge economy, including education, research, development, and data analysis, due to their ability to generate new contents. However, hallucination, which refers to the generation of content that appears plausible but lacks scientific basis, presents a significant challenge to the safe adoption of this technology in critical applications such as financial analysis and market prediction. This study aims to explore the effects of hallucinations on productivity of the knowledge economy and proposes some approaches to mitigate them. Through conducting a systematic literature review and qualitative analysis, different types of hallucinations in generative AI models are identified, and their effects on trust and productivity in knowledge-based systems are examined. The review of hallucination reduction methods indicates that the approaches utilizing reinforcement learning with human feedback enhance the reliability of the generated content by correcting errors in the model’s output through repeated adjustments. Finally, this study presents a hybrid approach to hallucination reduction in knowledge economy. This approach is based on three techniques of prompt engineering, retrieval-based generation, and self-improvement through feedback and reasoning. The results provide a foundation for future research on managing AI risks and enhancing the productivity of the knowledge economy.