India floats AI plan under which ‘The model travels to the data, rather than the data travelling to the model’

If you develop AI models using Indian data, you may soon need to comply with India's DEPA framework and use its secure data environments.

India's Ministry of Electronics and Information Technology (MeitY) has proposed a Data Empowerment and Protection Architecture (DEPA) for AI training. This initiative aims to foster the development of AI models trained on local Indian data, addressing national issues and ensuring data sovereignty. Under DEPA, the concept is that AI models will "travel to the data" rather than the data being moved to the models. This is achieved through "hardware-based Secure Data Environments" where data remains encrypted and isolated during processing. "Training Data Providers" (TDPs), such as hospitals and banks, will share anonymized and encrypted data with "Data Exchange Platforms" (DEPs). "Training Data Consumers" (TDCs), the model developers, will access this data within the DEPs to train their models. The framework emphasizes the use of differential privacy techniques and immutable logging for security and transparency, with TDPs setting rules for data usage.

This approach contrasts with current global AI development, where large datasets are often consolidated and processed by a few major tech players. DEPA's focus on "model-to-data" processing and secure environments could mitigate privacy concerns and data localization requirements, potentially encouraging more localized AI development across various sectors like healthcare, finance, and education within India. It also aligns with India's broader digital public infrastructure (DPI) strategy to empower local entities and reduce reliance on foreign Big Tech.

The DEPA initiative signals a potential shift in how AI training data is managed and accessed, especially in regions prioritizing data sovereignty and domestic innovation. By creating secure, controlled environments for AI training, India aims to unlock valuable datasets for specific national needs, such as personalized learning or improved diagnostics, without compromising user privacy or granting unfettered access to large corporations.