Assessing the Reliability of Artificial Intelligence Systems: Challenges, Metrics, and Future Directions

Purpose: As artificial intelligence (AI) systems become integral to diverse applications, ensuring their reliability is of paramount importance. This paper explores the multifaceted landscape of AI reliability, encompassing challenges, evaluation metrics, and prospective advancements. Methodology: This paper employs a comprehensive literature review approach to assess the existing body of knowledge on the reliability of AI systems. The review aims to synthesize insights into the challenges faced in evaluating AI reliability, the metrics used for assessment, and the potential future directions in this critical research domain. Findings: In this paper, challenges in AI reliability assessment, including explainability, data quality, and susceptibility to adversarial attacks, are scrutinized. Metrics for evaluating AI reliability, such as robustness, accuracy, precision, and explainability, are also elucidated. In addition, case studies illustrate instances where AI reliability has been successfully assessed or has fallen short, offering valuable insights. Originality/value: This paper sheds light on the complexities surrounding the assessment of artificial intelligence (AI) reliability and contributes to the ongoing discourse on AI reliability by providing a comprehensive examination of its challenges, metrics, and future trajectories.Purpose: As artificial intelligence (AI) systems become integral to diverse applications, ensuring their reliability is of paramount importance. This paper explores the multifaceted landscape of AI reliability, encompassing challenges, evaluation metrics, and prospective advancements. Methodology: This paper employs a comprehensive literature review approach to assess the existing body of knowledge on the reliability of AI systems. The review aims to synthesize insights into the challenges faced in evaluating AI reliability, the metrics used for assessment, and the potential future directions in this critical research domain. Findings: In this paper, challenges in AI reliability assessment, including explainability, data quality, and susceptibility to adversarial attacks, are scrutinized. Metrics for evaluating AI reliability, such as robustness, accuracy, precision, and explainability, are also elucidated. In addition, case studies illustrate instances where AI reliability has been successfully assessed or has fallen short, offering valuable insights. Originality/value: This paper sheds light on the complexities surrounding the assessment of artificial intelligence (AI) reliability and contributes to the ongoing discourse on AI reliability by providing a comprehensive examination of its challenges, metrics, and future trajectories.

کلیدواژه ها:

Artificial Intelligence ، Reliability ، Challenges in AI Reliability Assessment ، Metrics for AI Reliability ، Explainability and Transparency ، Data Quality ، Bias