A false start to the AI revolution

The IndiaStack team should revisit their approach with the DEPA training framework for creating AI models. It ignores customer consent and violates the provisions of Personal Data Protection Act

DEPA stands for Data Empowerment and Protection Architecture. The Account Aggregator, the DEPA team's first offering, embodies this spirit with customer consent as the bedrock of the system. There are a few implementation challenges with it at the moment.

The team recently launched a DEPA training framework geared towards the uptake of AI models in the country. The premise is that fine-grained consent is impractical for analytics and training purposes. See the 'Limits of Consent' section in this link. They are advocating a data sharing architecture with Training Data Provider (TDP), Training Data Consumer (TDC), Data Clean Room and Contract Service with customers nowhere in the picture.

When the Personal Data Protection Act is enacted, all organizations must collect customer consent for specific purposes. In its current form, the DEPA training framework can get the operators into trouble. Also, when we introduce too many digital intermediaries in the ecosystem, the data will be scattered across multiple organizations, resulting in high cumulative compliance costs.

The correct approach for data fiduciaries is to hand over the data to the customers on their digital wallets. The AI model developers can get them from the customers for a small fee (or a token gift). It is similar to how traditional primary research companies directly collect data from end customers by sponsoring their lunch or giving them a gift card.