Kept off the editorial: News organisations set limits for AI use

News publishers and broadcasters are using AI for mundane tasks but not editorial

She is reading the news on Navbharat Times online. Her Hindi is flawless. Her clothes and hair are stylish like most news anchors. Except for the somewhat mechanical enunciation of words and the tag on-screen that says ‘AI Anchor’ you probably wouldn’t have known that this is not a human being. That it is an artificial intelligence or AI-created anchor talking about some minister in Bihar from a script fed into it. Why bother? “A real human will work for eight hours. The AI version can work for 24 hours,” says Puneet Gupt, chief operating officer, Times Internet. There is Sana on Aaj Tak and AI Kaur on News18 among many others. “We brought her simply to understand what the technology is capable of doing. Maybe have a fully functional AI anchor in the future,” says Puneet Singhvi, chief executive officer (CEO), digital, Network18 Media and Investments, which operates CNBC-TV18 among 16 other news channels.
 

AI anchors are fashionable these days. However, a bulk of the usage of AI at Times Internet, Network18, Inshorts and many other news firms is about bread and butter stuff, like translating from one language to another, from text to audio, making summaries, giving headlines, video editing, writing tweets and sometimes creating ad campaigns.
 

Times Internet, which had a revenue of Rs 1,700 crore in FY23, reaches 110 million users everyday with brands like Maharashtra Times, Economic Times, The Times of India and Vijay Karnataka. “My goal is that 100 per cent of the content for any one language should be transformable to podcast, videos, other languages. What we are doing currently is Hindi to English, but eventually this should happen across all languages. If The Times of India (ToI) is in 40 cities now, the ability to take it to 100 goes up,” says Gupt. This pushes down the cost of growth and eventually improves revenues, much like in entertainment. For instance, Navbharat Times has reporters in cities that may not be covered by ToI. Earlier, ToI would take the copy of say a reporter in Moradabad and translate into English. It would take 20-30 minutes; AI can do it in three minutes.
 

Multiply this efficiency across markets, editions and anywhere from 6-12 languages for almost every major news publisher/broadcaster in India. You begin to understand why translation is the first and most crucial leg in the journey of integrating AI into news media.
 

“In English digital we have been playing around with AI but not much in Malayalam because AI is not trained in regional languages,” says Jayant Mathew, director, Malayala Manorama, the publishers of the largest-selling news brands in Kerala.

News aggregator Inshorts’ main product is a feed with 60-word summaries of the key stories of the day. It also owns Public, a news-based short video app. Azhar Iqubal, co-founder and CEO of Inshorts, says a small team is testing about 40,000 user-generated videos a day using AI. The idea is to see if it manages to ensure that the videos comply with Public’s guidelines. 

“We are using it for marketing to write captions, etc. It eliminates the mundane and cuts costs,” says Vignesh Vellore, co-founder, The News Minute. At Network18’s Moneycontrol AI is used to analyse reams of stock market data.
 

Almost nobody uses AI in editorial. Why?
 

“What we are providing is first-hand information. How do we use ChatGPT there because the information is never public before that. From an editorial perspective our stand is that we will not use it. For ground reporting, interviews, analyses, the possibility is zero. You can’t bring the nuance that we want,” says Vellore, referring to the AI chatbot that responds to a range of written queries. 
 

“For now, there are no plans for using AI for news generation due to ethical concerns,” says Gupt. That is a sentiment repeated across publishers. 
 

That is also where the hype around AI gets tested.
 

The hallucinating AI
 

All generative AI, the stuff that is hot right now, grows up on data. You feed tonnes of data into a software, it trains on it and then starts morphing humans by developing a cognitive ability. It creates ‘content’ on command. While this works wonderfully for entertainment, in news it holds dangers. “News is a sensitive topic. Human intervention to make a decision on factuality, tonality, neutrality and what is ethically acceptable will be important and cannot be replaced by any AI tool,” says Chandrashekar Mantha, partner and media and entertainment sector leader at Deloitte India.
 

There is the possibility of fake news, biases that can be fed into the data, and of course hallucination or making things up. If you ask ChatGPT to write an article about, say, the conflict in Mongolia but there isn’t enough data on it, the AI hallucinates. “One (entertainment) client said 'make it hallucinate',” says Garret Goodman, vice-president, sales at Papercup, which offers AI-based dubbing of videos into any language. He was speaking at IBC in Amsterdam this September. 
 

Getting AI to deliberately hallucinate to see what twists and turns a film or a show can take is fine. But do that on the news and you could end up with a riot on your hands. Many publishers reckon, off the record, that the coming general elections will test what AI gone rogue could do. “It is very dangerous. Today anyone can give information. What publishers offer is trust and credibility. For now, we are sceptical about using it in our newsrooms,” said Pawan Agarwal, deputy managing director, DB Group, at an International News Media Association (INMA) event this August. The Rs 2,168 crore (FY 2023 revenue) DB Group publishes Dainik Bhaskar, among other brands, and operates 30 radio stations.
 

Network18’s Singhvi thinks AI could be used to gather, tag and edit content for its 16 news channels, including CNBC-TV18, and for its online properties from its hyperlocal brand Local18.  He clarifies it will not be used for political stories but only for crime, human interest and civic issue ones. The lines then are clearly drawn.
 

“Technology always gets misused before we use it well,” said Shailesh Kumar, chief data scientist, Jio, at the INMA event. He thinks there is an urgent need for regulating AI, especially since it is open source. 
 

Most publishers like Agarwal say that they will be using AI only “on the business side.” 
 

The safe AI
 

Times Prime is a membership programme that gives users access to over 20 brands including Disney+Hotstar, EazyDiner and others. Members then enjoy discounts and offers from some 40 brands such Myntra, Swiggy and Uber. In May this year Times Prime released a campaign across online, out-of-home and print media that was created by AI. This includes the copy, model, everything. Harshita Singh, business head for Times Prime, reckons that if it had been a campaign designed by a real agency it would have taken over a month of planning and editing. The AI campaign was created and released in less than half that time and at one-tenth the cost. The tough part was “familiarising ourselves with the AI, giving it a sense of what we wanted,” she says. 
 

On the editorial side though, even within translating, “There isn’t 100 per cent accuracy, we need a human editor,” says Mathew.
 

To trust AI tools with generating news for dissemination and decision-making in a democracy then is a big question mark.