AI (Artificial Intelligence): India

From Indpaedia
Jump to: navigation, search

Hindi English French German Italian Portuguese Russian Spanish

This is a collection of articles archived for the excellence of their content.
Additional information may please be sent as messages to the Facebook
community, Indpaedia.com. All information used will be gratefully
acknowledged in your name.

Job creation

Data labelling/ 2019

Sonam Joshi, Sep 8, 2019: The Times of India

The economics of Data labelling
From: Sonam Joshi, Sep 8, 2019: The Times of India


Five years ago, Hyderabad resident Tulasi Mathi was forced to quit her job as a maths teacher due to health issues and the birth of her two children. But today, the 29-year-old does data labelling and makes up to Rs 15,000 a month. The money isn’t much but it’s more than she made as a teacher, and enough to pay her kids’ school fees and her own expenses.

She chanced on data labelling work through a You-Tube video in 2017. Today, she scans videos and marks and labels objects encountered by self-driving cars. Her output is used to train artificial intelligence algorithms powering such cars. All Mathi knows is that it makes her life easier. “I can work from home and don’t have to choose between work and family,” she says.

Mathi is one of the faceless workers helping companies in the US and Europe perfect their machine learning models. For instance, if you’re trying to get a driverless car to correctly identify a stop sign, you need to feed that algorithm thousands of images correctly labelled as stop signs. Sharmila Gupta of Gurgaon-based AI Touch likens data labelling to training a newborn. “Any AI model requires labelled data to get trained. This is like teaching a small child multiple times till they understand.” It’s a job that only humans can do and since it is quite labour intensive, it is being outsourced to countries with cheap labour like India, Malaysia, the Philippines and Kenya. There is a new AI workforce, says Ajay Shah of HR company TeamLease Services. From an opportunity point of view, there are about a lakh jobs posted on various portals currently.”

Mary Gray, a researcher and author of the book ‘Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass’, says there are two main reasons why companies are turning to India for data-labelling and annotation. “It has a workforce trained in English as a first language and an internet infrastructure created during the first outsourcing boom that relied heavily on India as an offshore labour market,” says Gray. There is also a growing demand for data-labelling services that are “localised”— both linguistically and culturally relevant to India — and this work can’t be done by workers in the United States.

Playment, a crowdsourced marketplace that trains annotators from scratch, where Hyderabad’s Mathi works, has 25,000 annotators between the age of 18 and 30 years working remotely across India, and its co-founder Siddharth Mall claims that anyone with a laptop and basic English skills can start working. “Everybody talks about how AI will make people lose their jobs, but there are also new kinds of jobs being created,” he says. These youngsters earn anywhere between Rs 20,000 and Rs 30,000 a month. It’s not a fortune but it’s helping many annotators — most of them stay-at-home moms, fresh graduates and even students, such as 21-year-old Shiekha Mahara from Nalgonda, Telangana — get by. Mahara, who recently completed a BTech degree, began looking for online work opportunities to help out with her family’s finances. She has earned Rs 1.3 lakh so far while doing occasional projects for Playment over the last year and a half.

Unlike Playment which pays people on a contractual basis, iMerit, a data annotation company with offices in India and the US and data labelling centres in Ranchi, Shillong, Vizag and Kolkata, has 2,500 people on their rolls. What they have in common is an overwhelmingly young workforce. At iMerit, the average age of employees is just 24. Jai Natarajan, VP, marketing and strategic business development, says that nearly 80% of their employees come from underprivileged backgrounds, while 50% are women. “Our employees are positioned for the future because they understand that they have to learn new things, that nothing stays still,” says Natarajan. iMerit’s employees do data labelling for drones in the agriculture sector, medical imagery such as MRI scans, e-commerce, and sports analytics. Mujeeb Kolasseri, a high school dropout from Mannarkkad in Kerala, founded his own data labelling company Infolks in 2015 after learning the work online. Today, the company employs nearly 250 people, nearly half of them from poor families in Kolasseri’s village. New employees get trained on image annotation tools for two months. “Nearly 80% are freshers. With proper training, anyone can work on image annotation without any technical knowledge — you just need to be a quick learner,” says Kolasseri, who was forced to quit his studies because of his family’s financial problems.

Jitendra Kumar, 27, would agree. Six months ago, Kumar, who used to drive a four-wheeler for weddings and parties in his hometown Etawah, Uttar Pradesh, found a data labelling job with Gurgaon-based firm AI Touch. “Now, I get a salary on time, work in an office and can spend some time with my family as well,” says Kumar. Kumar’s colleague Satyam Barthwal, a Chinese interpreter, was hired after the company got work from a Chinese AI company. “I hadn’t even heard of data labelling before I got the call,” says Barthwal, who sees the job as a way to earn money till he fulfils his dream of becoming a singer. “The work is easy — we just need to read labels.”

But as machine learning evolves, will it make the work of data labelling redundant in the future? “Since we started in 2013, the precision, nuance and sophistication required has gone up. Sometimes we need domain experts. But even then, you need humans to review, audit and keep track of results. There is going to be a role in AI for humans for a long time,” adds Natarajan of iMerit. 

Personal tools
Namespaces

Variants
Actions
Navigation
Toolbox
Translate