Google steps up AI efforts in India with eye on data monetisation
Google on Tuesday said it is stepping up efforts on artificial intelligence (AI) and machine learning (ML) to find solutions to local problems in India, with a focus on voice, video and vernacular. The last three verticals, the company hopes, will shape the future of India's internet as well as boost advertising business.
“Voice has been emerging as the preferred mode of use for new internet users. We’re seeing major growth of voice queries in India,” said Rajan Anandan, vice-president, India and South East Asia. “Furthermore, online video now accounts for 75% of all mobile traffic. And as for vernacular, the majority of the Internet users today are Indian language users, a number expected to reach 500 million plus in the next two years. 95% of video consumption is in vernacular languages,” he said.
However, the company finds too little content in Indic languages available on the internet. "While the internet is a vast repository of information, just like a big library, the Indian language that you speak in might just represent one shelf of books in that language," said Shashidhar Thakur, vice-president, engineering, Google Search.
The company will add more Indic-language content to the internet, he said. For this, the firm is targeting 135,000 or 90% of 150,000 government-registered newspapers and magazines that don't have websites because they are published in vernacular.
"In fact, the road from a print publication to a website is difficult because there is no mechanism to copy non-Unicode Indic language text from PDFs to webpages," Ken Tokusei, product management director at Google, said, adding that the company was launching a new programme called Navlekha using which vernacular publications can create content for their websites in Indic languages.
According to the company, Navlekha, powered by artificial intelligence, provides a console which has two windows showing old content and a blank window for new webpage content. Users can copy text and drag images from PDF files in one window and paste and drop data in the blank window to create a webpage. With Navlekha, Google is also offering free web hosting with a branded domain for three years and AdSense support to monetise content.
"Navlekha has started on-boarding publishers from Delhi and we will keep on adding more publications," Thakur said, adding that any media house could sign up for the programme.
Analysts believe that the push to create more Indic-language content is a monetisation strategy of the internet giant. "The more content you have, the more people will access it, creating a scope for advertisements to be placed and revenue to be earned. The company already supports adverts in Bengali, Tamil and Hindi and is planning to add more Indic languages," Navkender Singh, a senior analyst with market research firm IDC, told TechCircle.
Pavel Naiya, a senior analyst with Counterpoint Research, also said that the company's efforts seemed concentrated on generating more Indic language content for training its data engines for Google Assistant, which would lead to more voice computing and help the company in understanding consumer behaviour.
Thakur also said that the company was using artificial intelligence to bring Google Feed under Google News platform in Indic languages such as Bengali, Tamil, Marathi and Telugu and would add other languages soon.
But the company isn't stopping there. It is adding a lot of new abilities to its Assistant. "We are adding Marathi to the Assistant's ability of conversing in Hindi and English. We will be adding seven more languages soon," Pravir Gupta, director of software engineering, said, adding that the use of the Google Assistant has tripled in India since last year.
Gupta also said that Google was opening up Actions for developers to create third-party use cases such as finding a train or checking expiry of plan or data usage on Airtel or taking English language lessons from the Assistant. Actions on Google can be compared with Amazon's Alexa Skills -- basically, abilities of the digital assistant.
According to Singh, the changes in the Assistant mean that Google’s strategy is coming a full circle by adding more language abilities to its Assistant and extending them through its Android Go programme under which the company sells cheaper or entry-level smartphones.
"What you have to understand is that, most of these companies believe voice computing is the future because a lot of users are not comfortable with touch. If the Assistant can bring conversational computing interface, Google will succeed in its aim to bring these consumers under its fold in line with its Next Billion Users initiative, giving it the opportunity to collect more data and do a little more targeted advertising," he explained.
However, Singh was concerned about the success of the company's Android Go programme and shared that these phones were not doing well in retail chains and most people who use feature phones are not comfortable with touch, which stands in contrast to Google's claim that more than 50% of its Go phone buyers are feature phone users.
The company has also made some developments under the Go apps interface so that the Assistant can identify parts to read aloud in Hindi, English and Marathi and expects to add more languages here too. It also said that the Assistant could teach pronunciation in Hindi and English now, with support for more languages coming soon.
The company is also working on other local artificial intelligence and machine learning solutions. "We are looking to solve local problems in India using artificial intelligence and machine learning. One of our many projects is to work on flood-warning systems in countries such as India, China and Bangladesh, which are prone to floods. We are working with the water resources ministry here on a project basis to develop a flood-warning system based on satellite imaging and machine learning. The system can easily warn people or the government," Anita Vijaykumar, technical project manager at Google's TensorFlow divison, said, adding that the present warning systems don't get the job done accurately.
TensorFlow is Google's open-source software library for machine learning and can be used for a range of tasks for dataflow programming.
Vijaykumar added that the company was thinking of expanding the pilot to other regions in the country and was looking to provide data from the warning system to citizens and other people who live in flood-prone areas. She also said that the company was working on artificial-intelligence pilot projects in Tanzania and Amazon in order to save crops and prevent deforestation.
The TensorFlow technical manager also said that the company, via its acquisition, Verily Systems, was working with Aravind Eye Hospital and Sankara Nethralaya to prevent vision loss due to diabetic retinopathy. The company said it was also working with an Indian artificial intelligence startup, Nebula Innovations, via its Launchpad programme, to make sure that the farmer gets the right price for his/her product.