IT for Change responded to the Ministry of Electronics and Information Technology (“MeitY”) sub-committee’s Report on AI Governance Guidelines Development. Our comments largely addressed the need to address issues around generative AI, mitigating bias in AI training, and improving transparency.
Highlights from our feedback are as follows:
- It is necessary to adopt an integrated and indivisible human rights approach toward the development, deployment, and use of AI systems.
- On the issue of deepfakes, inspiration can be taken from Article 50(4), of the EU AI Act which requires deployers of AI systems that generate deepfakes to disclose that the content has been artificially generated/manipulated.
- To ensure cybersecurity, safeguards must be robust and not left to interpretation. For instance, Article 25, General Data Protection Regulation, provides for data protection by design and by default.
- To adequately respond to the issues arising from the advent of generative AI models, there may be a need to review copyright law so as to ascertain whether the concept of “fair learning” should be introduced.
- IP laws need to be revised to ensure that IP claims, particularly, trade secrets, are not misused to limit algorithmic transparency and evade accountability for AI models.
- Social and economic value generated from work produced by AI trained on public datasets should be fairly distributed to relevant communities.
- AI developers/deployers should undertake regular impact evaluation (including the impact of AI on human and fundamental rights).
- AI developers/deployers should ensure AI is trained using quality training data sets to mitigate the risk of bias/discrimination.
- Sectoral guidelines (for health and other sensitive sectors) are vital to provide key recommendations across the AI lifecycle.
- A public algorithms repository should be created to enhance transparency.
- Serious AI incidents should be mandatorily reported in the AI incident database.
- Baseline commitments for all AI systems are crucial to ensure the effective implementation of AI regulation.
Our detailed feedback is here.