Discussions on labelling AI content over, rules to follow soon: IT secy

The government has completed consultations with the industry on the proposed mandatory labelling of AI-generated content, and the related rules are set to be issued shortly, IT Secretary S Krishnan has said.

Krishnan told PTI in an interview that the industry has been "fairly responsible", understands the logic behind labelling of AI content and that, as such, there has not been serious pushback against it.

The primary feedback from the industry is around wanting clarity on the degree of modification to make clear distinctions between substantive, material changes through AI, and routine technical enhancements.

"Based on the inputs we have received, we are just consulting the other ministries within government, saying that these are the changes which have been suggested... so which changes we accept, which changes we make and what tweaks we make... that portion is on right now, and I think we should come out with the new rules very shortly," Krishnan said.

About the feedback and inputs coming in from the industry on this, Krishnan said, "I don't think they are against it."  "And again, this is not something which we are either asking them to register or go to a third party entity or placing some restriction of any kind. All that is being asked is label the content," he said, asserting that citizens have the right to know whether a particular piece of content has been generated synthetically or is authentic.

Krishnan said, at times, even minor AI edits may significantly alter meanings, while routine technical enhancements, say smartphone's camera related improvements, may enhance quality without changing facts.

"Most of the reaction is the degree and what kind of change because now advanced technology is such that there is some modification or the other in some sense. In some cases, modification can be very small, but that, in itself, can make a difference... One word in a sentence can make a huge difference to what the outcome is," he explained.

As such, use of technology and modern devices involve some level of enhancements.

"...the way that you photograph or take a video or record something, the phone itself enhances some of this, tries to make it better. So they (industry) want some clarity that those kind of technical changes, which don't alter anything in substance, but they are enhancements, are not simultaneously called into question when you do something like this... I think those kind of reasonable asks, we can certainly accommodate," Krishnan said.

That said, excluding all kinds of modifications can be an issue.

"Because, as I pointed out, even one or two words changing in a particular sequence of conversation could have a completely different effect and impact... say, the rest of it is a real conversation, but you used AI to change two or three words in a particular sequence of things that somebody says, it can make all the difference. And creativity also has its place and we are not against creativity, but people have a right to know this is actual real stuff, and this is not real," he said.

In October, the government had proposed changes to IT rules, mandating the clear labelling of AI-generated content and increasing the accountability of large platforms like Facebook and YouTube for verifying and flagging synthetic information, to curb user harm from deepfakes and misinformation.

The IT ministry noted that deepfake audio, videos, and synthetic media going viral on social platforms have demonstrated the potential of generative AI to create "convincing falsehoods", where such content can be "weaponised" to spread misinformation, damage reputations, manipulate or influence elections, or commit financial fraud.

The proposed amendments to IT rules provide a clear legal basis for labelling, traceability, and accountability related to synthetically-generated information, the ministry had said.

The ministry had invited comments from stakeholders on the draft amendment mandating labelling, visibility, and metadata embedding for synthetically generated or modified information to distinguish such content from authentic media.

The draft rules involved mandating platforms to label AI-generated content with prominent markers and identifiers, covering a minimum of 10 per cent of the visual display or the initial 10 per cent of the duration of an audio clip.