The story so far: On March 1, the Ministry of Electronics and Information Technology (MeitY) issued an advisory to the Artificial Intelligence industry. It said that all generative AI products, like large language models on the lines of ChatGPT and Google’s Gemini, would have to be made available “with [the] explicit permission of the Government of India” if they are “under-testing/ unreliable”.

What is the government’s stand?

The advisory represents a starkly different approach to AI research and policy that the government had previously signalled. It came soon after Rajeev Chandrasekhar, the Minister of State for Electronics and Information Technology, reacted sharply to Google’s Gemini chatbot, whose response to a query, “Is [Prime Minister Narendra] Modi a fascist?” went viral. Mr. Chandrasekhar said the ambivalent response by the chatbot violated India’s IT law.

How has it been received?

The advisory has divided industry and observers on a key question: was this an ‘advisory’ in the classic sense that was reminding companies of existing legal obligations, or was this a mandate? “It sounds like a mandate,” Prasanth Sugathan, Legal Director at the Delhi-based Software Freedom Law Centre said at an event on Thursday. The document, sent to large tech platforms, including Google, instructed recipients to submit an “[a]ction taken-cum-status Report to the Ministry within 15 days.” Mr. Chandrasekhar insisted that there were “legal consequences under existing laws (both criminal and tech laws) for platforms that enable or directly output unlawful content,” and that the advisory was put out for firms “to be aware that, platforms have clear existing obligations under IT and criminal law.” Mr. Chandrasekhar referred to rule 3(1)(b) of the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, which prohibits unlawful content like defamation, pornography, disinformation and anything that “threatens the unity … and sovereignty of India.” He added that the rules were intended for large tech firms and wouldn’t apply to startups.

The government hasn’t elaborated in detail on how IT laws can apply to automated AI systems in this way. Pranesh Prakash, a technology lawyer who is an affiliated fellow at the Yale Law School’s Information Society Project, said the advisory was “legally unsound,” and compared it to the Draft National Encryption Rules of 2015, a quickly withdrawn proposal to outlaw strong encryption of data in India.

The advisory also included a requirement for AI-generated imagery to be labelled as such, something that the industry has vacillated between taking serious efforts on doing. Amazon Web Services has tried implementing an ‘invisible’ watermark, but has expressed concern that such a move would be of little use as watermarks can be edited out fairly easily.

Rahul Matthan, a technology lawyer and partner at the firm Trilegal, urged a more permissive approach to AI systems. “In most instances, the only way an invention will get better is if it is released into the wild — beyond the confines of the laboratory in which it was created,” Mr. Matthan wrote after the advisory was released. “If we are to have any hope of developing into a nation of innovators, we should grant our entrepreneurs the liberty to make some mistakes without any fear of consequences,” he added, pointing to the aviation industry as an example, where he said air safety improved as a result of planemakers’ willingness to share information on failure with each other to collectively improve air safety.

What has been the government’s approach to the AI industry?

Until recently, the government itself shared optimism on AI, where Big Tech firms have often struck a balance between seeking regulation and seeking to control the direction these regulations take. The IT Ministry last April categorically said that “the government is not considering bringing a law or regulating the growth of artificial intelligence in the country”.

But in the last few months, even before the now viral Gemini response, Mr. Chandrasekhar has expressed dissatisfaction with AI models spitting out uncomfortable responses. “You can’t ‘trial’ a car on the road and when there is an accident say, ‘whoops, it is just on trial. You need to sandbox that,” Mr. Chandrasekhar said on AI firms’ responses to criticism on bias. The tension underlines the conflict inherent to widely testing an experimental technology — which is that wide testing is what allows these often unruly models to detect mistakes and improve. That dynamic was on display when Gemini generated racially incorrect photos of historical events, leading to a storm of criticism that led to the firm pausing the photo generation feature until it worked on a fix.

Will it benefit local developers?

“This is just a poor job in communication, resulting from the need to do something in an election year,” Aakrit Vaish, co-founder of Haptik, a conversational AI firm founded in 2013 said on X. Mr. Vaish amplified subsequent clarifications on the advisory’s applicability as good news for startups, and sought inputs to collect from local firms to send to the ministry.

Also read | Cabinet approves ₹10,372 crore corpus for AI infrastructure

Atul Mehra, founder of Vaayushop, an AI finance firm, expressed hope that the advisory could actually translate to a benefit for local developers. While it was a “short term hassle,” he conceded on X, “it’s a huge opportunity in disguise. It points to [a] need for local AI stacks, datasets, [and] GPUs [graphics processing units] … Let’s keep building and wait for our right moment to even beat Microsoft and Google.”

Source link