India is withdrawing a recently released AI advisory after receiving criticism from many local and global entrepreneurs and investors.
The Ministry of Electronics and IT on Friday shared an updated AI advisory to industry players, now not requiring them to hunt government approval before launching or deploying an AI model to users within the South Asian market.
According to the revised guidelines, firms are as a substitute beneficial to label under-tested and unreliable AI models to tell users of their potential fallibility or unreliability.
The revision comes after India's IT ministry received severe criticism from many senior figures earlier this month. Martin Casado, partner at enterprise firm Andreessen Horowitz, had called India's move “a travesty.”
The March 1 suggestion also marked an about-face from India's previous, hands-off approach to AI regulation. Less than a yr ago, the ministry declined to manage AI growth, calling the sector critical to India's strategic interests.
The latest advisory, like the unique earlier this month, has not been published online, but TechCrunch has reviewed a replica of it.
The ministry said earlier this month that while the suggestion was not legally binding, it signaled that it was the “way forward for regulation” and that the federal government demanded compliance.
The advisory emphasizes that under Indian law, AI models mustn’t be used to share illegal content and mustn’t allow bias, discrimination or threats to the integrity of the electoral process. Intermediaries are also beneficial to make use of “consent pop-ups” or similar mechanisms to explicitly inform users of the unreliability of AI-generated output.
The Department continues to emphasise ensuring that deepfakes and misinformation are easily identifiable and recommends that intermediaries label or embed content with unique metadata or identifiers. It isn’t any longer essential for firms to develop a way to discover the “creator” of a specific message.
