Data-driven justice: Survey on approaches to identify and mitigate biases to build fair artificial intelligent models for Indian justice system

Artificial Intelligence (AI) is widely used in decision-making systems, including the criminal justice system. Automated decision-making systems can speed up the handling of cases and improve consistency and efficiency. These systems were expected to enhance transparency and equip judges with data-driven insights. AI in criminal justice also pointed out concerns about bias and fairness. The idea is to build more inclusive legal systems. This paper explored different sources of potential biases in the judiciary. We compared different approaches to identify and measure the biases. We also reviewed techniques for bias mitigation, such as in-processing, pre-processing, and post-processing approaches. This work aims to comprehensively understand how to build fair AI models. We examined the widely used datasets and the fairness metrics used for evaluation. Most of the work on addressing biases is done in the Western context, leaving a notable gap in the Indian context. India, a country with rich diversity and a complex legal structure, needs AI models that are accurate and also equitable across different demographics to ensure justice and equity for all citizens. To achieve that, India needs bias detection and mitigation approaches that suit the Indian context, as well as evaluation metrics to measure fairness in decisions influenced by gender, caste, religion, etc. The approaches discussed in this paper were supported by case studies that explain the historical and cultural dimensions of the Indian judiciary.