What is the context?
In contrast to deepfakes, which use AI-driven tools to spread disinformation, cheapfake videos can be made by just about anyone.
- AI-powered deepfakes still rare on Indian social media
- Misleading videos and images are more common
- Fact checkers say platforms' new policies fall short
NEW DELHI – With India's elections in full swing, hundreds of social media users shared a video that appears to show Home Minister Amit Shah saying the ruling party wants to scrap a quota system aimed at eliminating centuries-old caste discrimination.
The controversial comments caused a brief uproar, but fact-checkers stepped in and declared the video a cheap fake, a fake made by altering old footage with basic editing tools.
Ahead of the ongoing election, the results of which are due to be announced on June 4, politicians and digital rights groups have expressed concern that voters could be swayed by disinformation contained in AI-powered “deepfake” videos.
But fact-checkers say most of the fake photos and videos posted online during the six-week election were not created using artificial intelligence, but rather used relatively cheap and simple techniques such as video editing and false labels to present content in a misleading context.
“Maybe 1 percent of the content we see is AI-generated,” said Kiran Garimella, an assistant professor at Rutgers University in India who studies WhatsApp. “From what we know, the proportion of misinformation is still very small.”
Fact checkers say the results can be equally persuasive whether it's a cheapfake or a deepfake, and that social media companies have a responsibility to do more to root out all forms of misinformation spread on their platforms.
“You can use AI to bring a dead leader back to life and people will realise it's propaganda… but if you mislabel a video or take it out of context, people are more likely to believe it,” said Pratik Sinha of Alt News, an Indian non-profit fact-checking website.
“Instead of falling into the dichotomy of deepfakes versus cheapfakes, we need to find ways to combat misinformation more effectively,” Sinha told Context.
Meta Platforms, which owns Facebook and Instagram, and X, formerly Twitter, have introduced new policies aimed at cracking down on various forms of misinformation in a crucial global election year, but fact-checking groups say the results have been disappointing.
Updated Guidelines
In response to criticism from the oversight board, Meta updated its guidelines in April to add prominent labels to all forms of misinformation. Meta's previous policy only applied to content that was altered or created using AI.
“We agree with the Oversight Board that the current approach is too narrow, as it targets only videos that are created or altered by AI to make it appear as if a human is saying something they did not,” Monica Bickert, the company's vice president of content policy, said in a blog post last month.
Under the new approach, which took effect ahead of the start of India's elections on April 19, fact-checkers working with Meta will review and rate posts on the social media network – including ads, articles, photos, videos, reels and audio – with six labels to provide more information to users.
You can use the labels “Error,” “Partially Incorrect,” “Altered,” “Uncontextualized,” “Satire,” and “True.”
Sinha questioned the effectiveness of the policy in dealing with false or misleading posts that are digitally manipulated during elections.
“We don't know how effective meth labeling has been,” he said, calling on the company to release data about its fact-checking program.
Contextual analysis found that many fact-checked videos on Facebook were not labeled correctly or did not carry warning labels at all.
One edited video appears to show Prime Minister Narendra Modi urging his supporters to vote for a rival party. Rather than being labeled “altered,” the video is labeled “partly false,” meaning it contains “factual inaccuracies.”
Karen Rebello, deputy editor at fact-checking site Boom Live, said X introduced new features to combat misinformation for its Indian users in April, but that these also fell short.
X says the Community Notes feature is designed to invite users from different backgrounds to be note-takers and combat misinformation by setting facts straight.
But Rebello said the authors of the memos often contradicted each other, creating further confusion as there was no clear agreement on the authenticity of the posts in question.
“A lot of misinformation has notes attached to it but doesn't surface because other posters don't agree with it. X needs to find a way to resolve this, otherwise it defeats the purpose of community notes,” she said.
Context found a cheap fake video of Mallikarjun Kharge, leader of the opposition Indian National Congress party, on X, which has generated no comment despite the video being debunked by a fact-checking website. In the mislabeled footage, which has been viewed 43,000 times, Kharge appears to say that his party will distribute Hindu wealth to minority Muslims.
A wider threat
Garimella said even when doctored videos are labelled as fake by social media platforms, they often still continue to spread on messaging apps such as WhatApp.
“40 percent of forwarded viral content has already been fact-checked multiple times, but messaging apps don't have that moderation, so it continues to spread,” Garimella said.
“This suggests that people may not be aware of it,” he said, warning that without stricter controls from the platforms, the situation is likely to continue.
Ahead of the elections, Meta, which owns WhatsApp, has launched a fact-checking helpline on its app in partnership with the Misinformation Alliance (MCA) to combat AI-generated misinformation in India.
Panposh Raina, head of the MCA's deepfake analysis unit, said most of the content reported to the helpline was manipulated using simple techniques rather than AI-driven tools.
But caution about deepfakes may be distracting platforms from the broader threat of misinformation, Sinha said.
“I have seen very few deepfake videos spreading disinformation, but (social media) are dedicating money and resources to exposing deepfakes. They should have done more research into the market,” he said.
(Reporting by Adnan Butt and Helen Popper Editing)
