Training AI models using data may need a rethink

With consent placed at the core of the Digital Personal Data Protection (DPDP) Rules, companies will need to rethink how they use customer data to train internal artificial intelligence (AI) models, including removing any training data where consent is not granted.

- AJINKYA KAWALE & AASHISH ARYAN

Retraining AI models may require significant cost and effort.

However, the data used for already trained models can be deleted from repositories on the lack of consent, without affecting the model’s efficiency, said industry experts.

This comes as companies processing user data — known as data fiduciaries — must clearly explain to users, or data principals, how their personal data will be used. Data fiduciaries must also provide an easy way for them to withdraw consent, according to the rules.

“The real problem is less about the trained models and more about the data behind them. Even if the model’s weights stay, user data stored in training repositories must be deleted on time, as it often remains in version-controlled systems used during training. It increases the governance of data,” said Ashok Hariharan, chief executive officer (CEO), IDfy.

The new rules require informed consent along with easy provisions to revoke it for any personal data processing.