Whalesbook
India's AI Policy: Balancing Deployment with Creation
India has positioned itself as a responsible leader in Artificial Intelligence governance with its latest guidelines. The approach favors digital public infrastructure and voluntary measures over stringent regulations, a philosophy resonating with many emerging economies. However, beneath the applause, a critical question looms: is India doing enough to foster global-scale AI creation, not just deployment?
The Digital Public Infrastructure Advantage
India's strengths lie in its world-class digital public infrastructure (DPI) such as Aadhaar, UPI, and DigiLocker. These platforms provide a robust foundation for developing AI applications efficiently, offering unique advantages in identity, payments, and data sharing. Applying AI solutions atop these existing systems is a clear Indian success story.
The Creation Gap
Despite excelling in AI application, India faces challenges in foundational AI development. Key issues include:
- Data Usage Clarity: There is no clear legal framework for using publicly available data to train AI models. The Copyright Act lacks text-and-data mining exemptions, creating uncertainty and pushing some developers offshore or towards fine-tuning existing models instead of building new ones. This impacts technological sovereignty, as India may end up using applications whose core technology is not domestically owned.
- Unresolved Liability: The question of who is responsible when an AI system causes harm remains unanswered. Current guidelines defer this to future legislation, introducing risk and potentially deterring smaller companies, especially in sensitive sectors like finance and healthcare, from engaging in foundational AI development.
- Underleveraged Research: India possesses strong academic talent in AI but lacks accessible, robust compute infrastructure for research. Initiatives like AIRAWAT are promising but face operational and transparency challenges, creating delays and slowing down experimentation compared to nations aggressively funding research.
Global Landscape and Lessons
Other nations are actively enabling AI creation. The UAE launched its open-source Falcon model, Singapore is shaping rules for explainability, the EU provides developer certainty, and the US offers a broad fair use doctrine. China's DeepSeek models showcased impressive capabilities, though privacy concerns limited global traction. The lesson is clear: trust, openness, and alignment are as crucial as model performance for global scale.
Recommendations for Policy Evolution
To bridge the gap from deployment to creation, India needs targeted interventions:
- Clarify that using publicly available data for AI research and training is legal.
- Establish safe harbors for AI developers, similar to those for internet intermediaries, making liability proportional and predictable.
- Make AIRAWAT fully usable by publishing access norms and creating simplified onboarding and shared compute clusters.
- Set up structured sandboxes for AI development in regulated sectors, with legal guidance.
- Implement a lightweight certification regime for AI models, rewarding fairness, transparency, and robustness.
Impact
Implementing these changes could unlock India's potential to not only use AI safely but to build, own, and shape its future, securing a competitive edge and offering a roadmap for other nations. This is crucial for technological sovereignty, economic growth, and global influence. The potential impact on the Indian technology sector and its global standing is significant.
Impact Rating: 8/10
Difficult Terms Explained
- Digital Public Infrastructure (DPI): Foundational digital systems like Aadhaar and UPI that enable widespread digital services.
- Copyright Act: Laws that protect original works of authorship, including data compilations.
- Text-and-data mining exemption: A legal provision allowing the copying of copyrighted works for analysis, often used in AI training.
- Compute: The processing power and infrastructure required for digital tasks, especially intensive AI model training.
- Foundational models: Large AI models trained on vast datasets, capable of performing a wide range of tasks.
- Fair Use Doctrine (US): A legal principle allowing limited use of copyrighted material without permission for purposes like criticism, commentary, news reporting, teaching, scholarship, or research.
- AI Governance: The framework of rules, standards, and practices for managing and directing AI development and deployment.
- AIRAWAT: A supercomputing mission in India aimed at boosting AI research and development capabilities.
- Sandboxes (Regulatory): Controlled environments where businesses can test new products or services under regulatory supervision.
- Certification regime: A system for verifying that AI models meet certain standards for safety, fairness, or robustness.