Fairness in Artificial Intelligence: A Comprehensive Review of Bias Detection: A Systematic Literature Review
- Version
- Download 5
- File Size 440.21 KB
- File Count 1
- Create Date 20 May 2025
- Last Updated 20 May 2025
Fairness in Artificial Intelligence: A Comprehensive Review of Bias Detection: A Systematic Literature Review
Authors:
Lokesh Kumar, Navneet Kumar Bind
ABSTRACT
In this paper, bias refers to the systematic preference of AIs for some groups and against others, which can cause harm. Despite the progress in AI technologies like recommendation systems, generative models, and predictive analytics, AI systems suck up biases from datasets, algorithms, and operational processes. It is important to tackle bias as AI is already, or plans to be, used in areas like healthcare, education, and governance. The main causes of bias are data bias, where datasets are unbalanced, and algorithm bias, where the design of the algorithm is unfair.
The focus of this review is to identify the types of biases and the importance of fairness in AI systems. Current research is trying to develop datasets, fairness metrics, and debiasing heuristics, but each has its own drawbacks. The majority of metrics do not capture intersectional biases properly, and the mitigation techniques generally lead to residual or domain-agnostic biases being left unmitigated. Also, most of the frameworks do not consider the contextual biases that are specific to non-Western societies, particularly Indian society.
To address these gaps, this review assesses the current datasets, bias quantification metrics, and debiasing approaches. It also discusses the weaknesses of current solutions and suggests future research directions. Some of the suggested directions include developing comprehensive, high- quality, and region-specific datasets, developing new fairness metrics that are suitable for various application domains of AI, and developing efficient and scalable debiasing approaches for both generative and multimodal AI systems. This comprehensive review is expected to advance the quest for fair and reliable AI systems with a view on fairness in various settings around the world.
Keywords: Bias Detection in AI, Algorithmic Bias, Dataset Bias, Fairness Metrics, LLMs