Need a human touch: Govt tells Meta not to leave takedowns solely to AI

The government has asked Meta to ensure human review in AI-led content moderation, warning that keyword-based takedowns can remove legitimate posts without context

Don't leave decisions to artificial intelligence (AI); ensure there’s a human in the loop. This is the government’s message to Meta, according to sources in the know. 

Flagging the absence of human reviewers in the tech giant’s AI-enabled content moderation systems, the government is said to have asked the company to put in place systems to ensure that user content is not taken down solely on decisions taken by AI engines. 

In recently concluded meetings with senior Meta executives, including the Chief of Global Affairs Joel Kaplan, officials from the Ministry of Electronics and Information Technology (Meity) said AI systems of the intermediary had been trained to detect only “static” keywords to flag and take down content across all three of the company’s platforms. The tech major invited scrutiny after it took down Prime Minister Narendra Modi’s post during the student protests in Delhi. 

This approach was problematic and could lead to any content being taken down even if it did not violate any laws or community guidelines, an official present at the meetings said. 

“Take for example a recent case where a government organisation shared details of how the law enforcement agencies had successfully extradited pro-Khalistan fugitives from certain countries. When the details of this operation were shared on Facebook by that government department, the post was automatically taken down by AI-enabled content moderation systems due to the presence of the word ‘pro-Khalistan’, without the AI system understanding the context,” a senior government official present at these meetings told Business Standard. 

A human in the loop would have prevented such an error from being made, the official said. 

Apart from Meta, the lack of a human agent in the AI decision making system has also been observed across other intermediaries such as X, which also restricts access to any posts that contain keywords that have been banned or barred as per their community standards, another official said.

 

“Right now, we are studying such instances where the decision to take down a post has been taken down solely by the respective AI-enabled content moderation systems of social media intermediaries due to one particular word. We might take some action based on the report,” the official said.

 

Meta and the IT ministry did not respond to emails seeking their response on the issue till press time.

 

Between August 5 and August 10 this year, senior officials from the IT ministry, including Union Electronics and Information Technology Minister Ashwini Vaishnaw, met senior global and Indian executives, including Kaplan, from Meta and questioned them on the company’s handling of Child Sexual Abuse Material (CSAM), as well as other content.

 

During these meetings, Meta executives were also questioned about the inadvertent 5-hour takedown of Prime Minister Narendra Modi’s video. in which he was addressing students protesting the leak of the National Eligibility-cum-Entrance Test (Neet) exam’s question papers.

 

Kaplan was accompanied by Neil Potts, vice-president of public policy at Meta, Rafael Frankel, the head of Asia Pacific public policy at the company, and Aman Jain, head of public policy in India, among other senior executives.

 

On August 5, during his meeting with Vaishnaw, Kaplan apologised to the minister on behalf of Meta for the erroneous restrictionoin PM Modi's post.

 

Earlier that day, Kaplan and his team also met Meity Secretary S Krishnan and other senior officials from the ministry. During the meeting, Meta executives were questioned on multiple aspects, including the presence of Child Sexual Exploitative and Abuse Material (CSEAM), the takedown of Modi’s post, the presence and proliferation of deepfake content across various platforms of Meta, as well as the intermediary’s inaction on the issues raised by the Grievance Appellate Committee (GAC).

 

Plugging the gaps

• AI systems trained to detect only ‘static’ keywords to flag content
• Human reviewers necessary to prevent such content takedown
• Meta asked to put humans in the AI decision-making loop
• Govt studying such takedown instances across other platforms
• Will reach out to other intermediaries if needed