After backlash from numerous national and international companies and entrepreneurs, India is changing a previous AI guidance. The Indian Ministry of Information Technology and Electronics modified its artificial intelligence guidance on Friday. They discontinued the requirement that stakeholders from the industry obtain government permission before releasing or distributing an AI model to consumers for the South Asian call.
Instead, under the updated regulations, companies are encouraged to identify artificial intelligence models that haven’t been tested sufficiently or are inefficient so that customers are aware of any potential vulnerability.
The change comes after numerous prominent figures harshly criticized India’s Technology minister earlier this month. Associate at Andreessen Horowitz, Martin Casado, had referred to India’s action as “a mockery.”
The 1st of March recommendation also represented a change from India’s prior free-market capitalism attitude toward artificial intelligence legislation. The ministry rejected attempts to control the rise of artificial intelligence almost a year ago, highlighting the industry’s significance to India’s geopolitical objectives.
Similar to the first recommendations from earlier this month, the updated one has yet to be posted online, although we have reviewed a copy.
Although the advice was not enforceable by law, the ministry stated early in the month that it indicated the “future of legislation” and government participation was necessary.
The advice highlights that artificial intelligence models must not be used to distribute illegal content in violation of Indian law, nor should they support prejudice, discrimination, or actions that threaten the credibility of the election process. Additionally, intermediaries are recommended to notify consumers directly regarding the unpredictable nature of AI-generated content by using “permission notifications” or other techniques.
The ministry continues to highlight the need for deepfakes and disinformation to be immediately distinguishable, and it advises intermediaries to mark or incorporate information with specific identification numbers or metadata. Organizations do not need to devise a method to determine the “originator” of a given data.
