A chatbot that won't take bribes for giving advice is hit in India

By Saritha Rai

ChatGPT quickly found a home in the sophisticated echelons of investment banks and drug design firms. Now, advanced artificial intelligence is coming to a huge workforce that’s largely tech-illiterate and non-English speaking: India’s domestic workers, waste recyclers and struggling farmers.

In the crowded neighbourhoods of Bangalore, ragpickers, cooks and cleaners are taking part in an AI trial aimed at helping some of the nation’s poorest people access money from government anti-poverty programs without getting snarled in red tape and corruption. It’s an opportunity for people such as Vijayalakshmi, who earns just $100 a month cooking for households in Jayanagar. She only uses her smartphone for basic purposes and speaks no English. Yet, that sweltering afternoon in April, she joined a gaggle of domestic workers to experiment with AI technology.

Also Read | Generative AI: Take the lead Bengaluru, for India

Vijayalakshmi, who goes by a single name as is common in southern India, voiced a question to a bot in her native Kannada language on education scholarships. Moments later, a human-like voice responded to explain the government aid available to her 15-year-old son. While the sea change that was ushered in by OpenAI’s release of ChatGPT in November has brought to the fore concerns about AI’s role in the spread of disinformation and the potential loss of jobs, the tests in Bangalore and in Mewat in northern India show it’s also a tool that can aid social equality. The tech can assist in professional communications, empower people who don’t have language skills and help those who have a disability, such as users of BeMyEyes, a personal assistant for the visually impaired.

Helping the Underprivileged

Removing language and tech barriers is particularly important in India, where about 16 per cent of people live in poverty, according to the United Nations. The world's most populous nation is positioning itself to be fully open to AI developments, in contrast to China, which bars the use of ChatGPT, and the US and the UK, which are studying how to regulate AI. India’s ministers say the country is in no rush to bring in regulations, and instead may find ways to innovate and use technology to level-up language, education and cultural inequalities.

Multiple AI chatbots are being built in India to help the underprivileged seek legal justice, dispense advice to farmers and help migrant workers get support in cities.

“Billions are left behind by technology but AI can help them hurdle over the barriers of literacy and tech-savvy,” said Rahul Matthan, a partner who heads the technology practice at the law firm Trilegal and is an adviser to India’s Ministry of Finance on digital public infrastructure. “Blanket bans or sweeping regulation is not the path for India.”

Satya Nadella, the India-born chief executive officer of Microsoft Corp. — a major investor in OpenAI — discussed the difference the technology could make to a remote villager’s life at the World Economic Forum earlier this year. “A large foundational model developed in the West Coast of the US a few months before made its way to a developer in India,” Nadella said. “I’ve never seen that kind of diffusion before. We are waiting for the industrial revolution to reach large parts of the world after 250 years.”

Still, the technology’s lightning-fast spread is a cause for alarm for many. The trials in Bangalore took place just as OpenAI Chief Executive Officer Sam Altman urged US lawmakers to regulate AI, with concerns about political manipulation, health misinformation and hyper-targeted advertising being raised. Altman and other leaders of AI companies have since warned of the existential harms of the technology, including the risk of extinction.

There is also apprehension in India, with Jibu Elias, Mozilla senior fellow for responsible computing based in New Delhi, raising concerns around consent, data privacy and security. This becomes particularly problematic when dealing with people who may lack technical skills and formal education.

No Bribes Needed

None of the women in the Bangalore trial had heard of ChatGPT and some of them had given up on receiving aid after struggling with language barriers, and government officials and middlemen demanding bribes.

The Bangalore trials were led by Saurabh Karn and his team at the nonprofit OpenNyAI. By feeding a collection of millions of parallel sentences spoken in different Indian languages into machine translation software, and adding thousands of hours of dialogue for speech recognition, the bot, named Jugalbandi, offers text-to-speech multi-language translation on the fly. For instance, a rural farmer can pose a question in Haryanvi, the language spoken just outside Delhi, and the tool translates it into English, searches the database for an appropriate answer, and then translates the answer back to Haryanvi and voices it out in a human voice via Meta Platforms Inc.’s WhatsApp to the farmer.

Jugalbandi has been trained to filter out personally identifiable information, such as India’s unique digital identity number or details like phone numbers and location, even before the user's question is translated. Still, Karn acknowledges that India's vast social challenges are too big for AI alone to solve.

But for women accustomed to challenges from bureaucracy and corruption, it's a start. “The robot can’t throw our application in the waste bin like the government official does when he’s dissatisfied with the bribe amount,” said Yashoda, a home cleaner who tried the tool at the domestic workers’ trial.

A week after Vijayalakshmi’s trial, a group of waste pickers in a neighbourhood called Hebbal at the other end of Bangalore presented a new reality. The women spend their days scouring the streets and gathering waste plastic, metal scraps and paper, then eke out a living by selling the daily collection to the local recycler. Most of them don’t own smartphones, showing that while AI can bridge some chasms like language and literacy, it can also exacerbate the divide by totally excluding those without access to technology.

The AI chatbot, meanwhile, reared up survival instincts in Vijayalakshmi.

“Bribe-rejecting robots are OK but don’t build any that can do house chores,” she said. “I don’t want to lose my job to a robot.”