Can algorithms solve the fake news problem in India?

Fake news has drawn many comparisons with creatures from mythology: from the Lernaean Hydra, the many-headed snake that sprouted two new heads when one was chopped off, to Raktabīja from the Indian mythos, a demon whose blood was so potent, every drop of it spilled on the ground would instantly transform into a new asura. No wonder then that for some time now, media and technology companies around the world have looked at Artificial Intelligence (AI)-led algorithms as the saviour—the demon-slayers, if you will, to extend the fantastical metaphors.

There is serious work happening in the field. Delhi-based MetaFact, for instance, is a tech-media start-up passionate about fighting fake news using fact-checking and content validation tools, which it is close to launching. Founded in 2017, MetaFact offers a fact-checking tool (in the form of an AI-based chatbot) that can detect, monitor and counter fake news or misinformation in real-time. Powered by the IBM Watson platform, it uses a combination of natural language processing programmes and machine learning to create fake news detection tools.

The MetaFact tool is designed in such a way that users can validate everything—text and images—from a single window. The tool will have four dashboards: detection, monitoring, investigation and results. “It will work in the form of an editorial workflow," says Sagar Kaul, founder, MetaFact. “The detection will work not only for misinformation and breaking news, but it will also give you trending news scenarios based on your location," he adds.

MetaFact acknowledges that checking the spread of misinformation requires a mix of technology and journalistic skills. Its six-member core team is divided neatly in half between journalists and developers. “Our team was founded with the core belief that AI with a blend of domain expertise has the power of solving some of the most pressing global issues, including the spread of misinformation," says the company website.

About a month ago, researchers at the University of Michigan announced that they had developed an algorithm-based system that identifies linguistic cues in fake news stories. They were then able to demonstrate in a study that the system was “comparable to and sometimes better than humans at correctly identifying fake news stories"; while the system found fakes up to 76% of the time, the human success rate was around 70%, according to an official news release on the university’s website. The system uses linguistic cues and analysis, the researchers say, analysing aspects of text-based news like grammatical structure, word choice, punctuation and complexity.

However, newsrooms in India dedicated to fighting the onslaught of fake news descending upon our inboxes, chat messages, Twitter and Facebook feeds have a different take on the capability of algorithms that can detect and debunk fake news.

“Fake news in India is very complex. I’d say it’s much more complex than fake news or biased news in Western countries, which is mostly disseminated through websites. In India, most fake news is distributed via other channels and the formats are not text-based, so techniques like linguistic pattern detection and natural language processing are not easy to use," says Pratik Sinha, a former software engineer, who co-founded the Indian fact-check website Alt News in February 2017.

Language adds another layer of complexity to the problem, says Sinha. While websites like Alt News and BoomLive.in do run parallel websites in Hindi, covering all Indian languages—either manually or using AI—is a huge challenge as most algorithms available today are trained only to detect patterns in English text.

Then there’s the problem of distribution. “Platforms like WhatsApp and Facebook, which are used extensively to share fake news, are closed systems that do not allow independent developers to build third-party algorithms and apps on top of them," says Sinha. In fact, many fake news nuggets shared on social media are not even indexed on search engines and don’t show up on a general Google search.

Moreover, in India fake news items are usually photo-heavy with less text or with embedded text, which defy algorithmic scrutiny, says Sinha.

Ultimately, in newsrooms such as Alt News and BoomLive, the job of sifting through fake news submissions that come in via Twitter DMs, WhatsApp messages and emails, running fact-checks on them (which often includes doing primary reporting and talking to sources to authenticate or debunk claims), and writing short pieces deconstructing them falls on reporters and editors, which is why campaigns like the Google News Initiative have roped in fake news busting experts to train Indian journalists, especially those from regional language media.

In an email, Jency Jacob, managing editor at BoomLive, who is conducting a workshop on news verification in Chennai for the Google News Initiative, says that “machines have not learnt it all yet". “Our work in India over the last 18 months shows that a lot of nuances and the need for external reporting make it impossible for algorithms to do our job with precision," he adds.

However, both Jacob and Sinha say that fact-check sites do use technology quite heavily in their investigations, such as looking at the IP addresses from where fake news bulletins have emerged, checking the unique content ID of videos on YouTube and Facebook, and doing reverse image searches to determine the provenance of stories and images.

Evidently, fighting fake news is a mammoth task that needs both artificial and human intelligence.