India Pushes for Localized Legal AI Amid Challenges
Experts urge India's legal sector to develop AI tailored for its complex system.
Why it matters: Localized AI can address India's unique legal intricacies and improve workflow, despite regulatory gaps.
- India's diverse system needs jurisdiction-specific AI for precise legal assistance.
- Global AI models often fail, with examples of inaccurate legal references.
- Building local AI is costly, ₹4–10 crore for processing in two years.
- Absence of AI-specific regulations adds to legal professionals' challenges.
India's legal system is intricate, involving multiple courts and specialized regulations that challenge global AI models. According to Manupatra's CEO Deepak Kapoor, such models often fail in providing accurate legal citations and lack jurisdiction awareness, leading to potential case mishandling.
Localized AI solutions are crucial for addressing these issues. A study highlights a "domain-partitioned hybrid RAG system" that achieved a notable 70% pass rate on specialized benchmarks, demonstrating the potential of native AI to outperform global counterparts in understanding India's legal nuances.
Significant resources are necessary to convert unstructured legal data into machine-readable formats, consuming 70–80% of engineering efforts. Sakshi Sadashiv notes the expense of this process, estimating costs between ₹4–10 crore over two years, emphasizing the investment needed to build reliable local AI systems that reduce bias and protect confidentiality.
While tools like Kira Systems exist, they don't account for India's unique legal context, underscoring the need for indigenous AI development. However, India currently lacks specific AI regulations, which complicates integration efforts. Although the Digital Personal Data Protection Act provides a framework for managing AI tools and sensitive data, more comprehensive legal tech policies are necessary.
By the numbers:
- 70% — Pass rate of localized AI in specialized legal benchmarks.
- ₹4–10 crore — Cost estimated for developing native AI over two years.