Indian election: Misinformation, AI deepfakes surge on social media

Claims circulating online in India recently have misstated details about casting a ballot, claimed without evidence that the election will be rigged, and called for violence against India’s Muslims.

Researchers who track misinformation and hate speech in India say tech companies' poor enforcement of their policies has created perfect conditions for harmful content that could distort public opinion, spur violence and leave millions of voters wondering what to believe.

"A non-discerning user or regular user has no idea whether it's someone, an individual sharing his or her thoughts on the other end, or is it a bot?" Rekha Singh, a 49-year-old voter, told The Associated Press. Singh said she worries that social media algorithms distort voters' view of reality. “So you are biased without even realising it,” she said.

In a year crowded with big elections, the sprawling vote in India stands out. The world’s most populous country boasts dozens of languages, the greatest number of WhatsApp users as well as the largest number of YouTube subscribers. Nearly 1 billion voters are eligible to cast a ballot in the election, which runs into June.

Tech companies like Google and Meta, the owners of Facebook, WhatsApp and Instagram, say they are working to combat deceptive or hateful content while helping voters find reliable sources. But researchers who have long tracked disinformation in India say their promises ring hollow after years of failed enforcement and “cookie-cutter” approaches that fail to account for India's linguistic, religious, geographic and cultural diversity.

AI deepfakes

Research by the group and another organisation, India Civil Watch International, found that Meta all owed political advertisements and posts that contained anti-Muslim hate speech, Hindu nationalist narratives, misogynistic posts about female candidates as well as ads encouraging violence against political opponents.

The ads were seen more than 65 million times over 90 days earlier this year. Together they cost more than $1M.

Meta defends its work on global elections and disputed the findings of the research on India, noting that it has expanded its work with independent fact-checking organizations ahead of the election, and has employees around the world ready to act in case its platforms are misused to spread misinformation. Nick Clegg, Meta’s president of global affairs, said of India's election: “It’s a huge, huge test for us.”

“We have months and months and months of preparation in India,” he told The Associated Press during a recent interview. “We have teams working around the clock. We have fact-checkers in multiple languages operating in India. We have a 24-hour escalation system.”

Google, YouTube's owner, criticised the research and noted that it has multiple procedures in place to catch ads that violate its rules. Global Witness removed the ads before they could be spotted and blocked, the company said.

“Our policies explicitly prohibit ads making demonstrably false claims that could undermine participation or trust in an election, which we enforce in several Indian languages,” Google said in a statement. The company also noted its partnerships with fact-checking groups.

AI is this year's newest threat, as advances in programs make it easier than ever to create lifelike images, video or audio. AI deepfakes are popping up in elections across the world, from Moldova to Bangladesh.

Senthil Nayagam, founder of an AI startup called Muonium AI, believes there is growing demand for deepfakes, especially for politicians. In the run-up to the election, he had several inquiries about making political videos using AI. “There’s a market for this, no doubt,” he said.