The IT ministry (MeitY) has proposed a consent-based framework to facilitate data sharing for training AI models in India. 

Called the Data Empowerment and Protection Architecture (DEPA) for AI training, the framework aims to help startups, businesses and academic institutions access large-scale, India-specific datasets to develop AI models while protecting citizens’ privacy and ensuring compliance with the DPDP Act. 

The DEPA for AI training also proposes a techno-legal architecture to enable organisations to collaborate on AI model training without directly transferring underlying datasets. 

With this, the ministry aims to foster access to datasets for researchers and businesses looking to train AI models. The guidelines have been formulated in collaboration with think tank iSPIRT.

Explaining The New Proposed Framework

The proposed architecture builds on India’s existing DEPA framework, which already supports consent-based data sharing through initiatives such as Account Aggregator. The new framework encompasses the following technical pillars: 

• Confidential Clean Rooms: Secure computing environments where organisations can collaborate on data processing and AI model training without exposing the underlying datasets to one another
• Differential Privacy: A privacy-preserving technique used to ensure that individual-level information cannot be identified or extracted from an aggregate dataset
• Electronic Contracts: Machine-readable agreements that define and enforce data-sharing conditions, including permitted uses, data protection and security requirements
• Zero-Trust Security: An approach where every access request is authenticated and authorised, with encryption, continuous monitoring and auditing to protect data 

The proposed framework also defines key stakeholders for enabling safe sharing of data for training AI models:

• Training Data Providers (TDPs): Organisations that make their data accessible through the framework. For instance: banks, hospitals, government agencies 
• Training Data Consumers (TDCs): Startups, enterprises, researchers and individuals seeking access to datasets to train AI models 
• Data Collectives or Data Collabs: Intermediaries that aggregate, standardise, and organise data from various sources and define data-sharing policies and governance rules for participants 

Under the DEPA framework, TDPs would first register their datasets on a data exchange platform after anonymising and encrypting them. The datasets would then be tagged with metadata and cryptographic identifiers for traceability. 

TDCs can then request a dataset for their specific use cases, and an electronic contract system checks their requests against data-sharing policies and governance rules before granting access. 

Once access is given, the model is then trained within a secure data environment using differential privacy methods, ensuring that the raw data remains inaccessible to consumers. 

The framework also proposes that trained models must undergo independent checks for bias, fairness, security, safety and explainability. Only models that pass receive approval for pilot deployment and scaling. 

Even after deployment, the oversight will continue in the interest of long-term trust and accountability. In line with this, the framework will ensure that TDCs remain subject to continuous monitoring, periodic reviews, and traceability of the model’s outputs back to the data and contracts used in its training. 

Govt’s Focus On AI Governance

The launch of the new DEPA framework for AI training comes soon after TRAI also released draft guidelines on the responsible use of AI in telecom networks and service delivery. The telecom regulator has sought inputs from stakeholders by November 27. 

As per TRAI, its draft guidelines are rooted in the IndiaAI Governance Guidelines. The centre had earlier released these guidelines to focus on fostering innovation and safe development and deployment of AI offerings, while mitigating risks posed by the technology. 

A few weeks ago, union IT minister Ashwini Vaishnaw had also said that the centre will release a consultation paper for regulation on AI safety and security within a month. Earlier reports had also hinted that the Centre was looking at formulating a new framework for AI regulation. 

The post MeitY Floats New Privacy-Led Framework To Unlock Data For Training AI Models appeared first on Inc42 Media.