MeitY introduced a consent-based framework to unlock large-scale India-specific datasets for AI training. The framework aims to give Indian startups robust data access for AI models without violating citizen privacy under the DPDP Act. This could directly address India's long-standing challenge of data scarcity for domestic AI innovation.
How We Got Here
This DEPA for AI training builds on India's existing DEPA framework, which already enables consent-based data sharing through initiatives like Account Aggregator. Formulated with think tank iSPIRT, the new guidelines address the critical need for curated datasets to train India-specific AI models since the DPDP Act came into force.
The Numbers
- The framework outlines three key roles: Training Data Providers (TDPs) such as banks and government agencies, Training Data Consumers (TDCs) including startups, and Data Collectives for aggregation.
- It mandates four technical pillars, notably Confidential Clean Rooms for secure data collaboration and Differential Privacy to safeguard individual information in aggregate datasets.
- TDPs must anonymise and encrypt their datasets, then register them on a data exchange platform before any access is granted to TDCs.
- Trained AI models developed using this framework will undergo independent checks for bias, fairness, security, safety, and explainability.
What Happens Next
🇮🇳 Why This Matters for India
For Bangalore-based AI founders building sector-specific models in healthtech or fintech, this framework could finally open up crucial, anonymised Indian datasets, accelerating product development.
The Take
This framework's success hinges entirely on top-tier TDPs like major public banks and hospitals actively joining Data Collectives. Without their anonymised, large-scale data, Indian AI startups will remain dependent on global datasets, missing out on crucial local nuance.
Source:
Inc42 ↗