Ethical Challenges and Bias in Artificial Intelligence Systems
Author : Nikhith Vasa and Vasavi Eneshetty
Abstract :
Artificial intelligence systems are increasingly used in high-stakes decisions across the healthcare sector, hiring, criminal justice, and law enforcement. However, growing evidence discloses systematic ethical issues and measurable bias. This paper reviews the types and sources of bias in AI, examines key ethical issues, including fairness, transparency, accountability, and privacy, and analyses real-world studies such as Amazon’s hiring tool, UK police facial recognition, COMPAS recidivism algorithm, and biased clinical AI. Quantitative findings show false positive rates for Black populations in facial recognition up to 5.5% compared to 0.04% for white people, over 15% of AI hiring tools fail basic fairness tests, and nearly 38% of generative AI outputs contain discrimination. Current mitigation approaches include fairness-aware machine learning, algorithmic audits, explainable AI, and diversified data collection. Regulatory systems such as the EU AI Act impose fines up to €35 million or 7% of global turnover. Despite promising technical advances, fundamental open problems remain, including the fairness-accuracy trade-off, dynamic bias amplification, and the mathematical impossibility of achieving perfect fairness in general-purpose AI. The paper concludes that no single solution exists; instead, a combination of transparent data, continuous auditing, strong regulation, and cross-sector collaboration is required to build AI systems that remain both powerful and just.
Keywords :
Artificial intelligence bias, ethical issues, algorithm fairness, facial recognition discrimination, AI regulation, fairness-aware machine learning, COMPAS recidivism bias.
