Like ruling regimes elsewhere, the Narendra Modi government, too, has seen artificial intelligence as a gamechanger for a while but it started work earnestly on tailoring a strategy around the technology only about a year ago. That was when the government’s policy think tank Niti Aayog called for a meeting, chaired by its CEO Amitabh Kant.
The meeting was the start of a journey to understand what AI means for India – could it be meaningful for a country of 1.3 billion people or was it just another flashy technology shining on the hype cycle?
For months after, at several meetings with the academia, government departments and agencies, technology companies, and amongst Niti Aayog officials themselves, the question posed itself in different ways. “The world is certainly looking at AI but does India need it?”, “If yes, what are its unique applications?”, “Are there high impact areas that India could bet on?”…
The meetings and the debates threw up one conclusion: in a country with one in five of the world’s poorest and lagging on healthcare, education, and farming incomes, AI would need to address needs of its teeming masses, the so-called bottom of the pyramid. The #AIforAll strategy is detailed in last week’s Niti Aayog discussion paper on India’s AI strategy.
“The maximum AI deployment has happened in sectors with commercial interest,” said Anna Roy, a 1992-batch officer of the Indian Economic Service who was instrumental in bringing out the paper. “There are sectors where public goods are in prominence, where there are high externalities, where we cannot depend on private motive to deploy AI. The government needs to play a bigger role,” she told FactorDaily in an interview.
Internationally, there is a race between China, the US, and Russia, in AI. It is clear, for instance, that China doesn’t just want to develop AI solutions for itself but also set global standards. International commentators have also noted how the gap between the US and other countries on the defence front is reducing as its rivals adopt AI solutions rapidly.
That the Indian government was serious of AI and other cyber physical technologies was signalled in its economic survey of 2017-18. It mooted a national mission on cyber physical systems as part of half a dozen such missions that would double down on technology and R&D in areas such as dark matter, genomics, energy storage systems, mathematics, and agriculture.
There is a “hugely multidisciplinary area including deep mathematics used in artificial intelligence, machine learning, big data analytics, block chains, expert systems, contextual learning going to integration of all of these with intelligent materials and machines, control systems, sensors and actuators, robotics and smart manufacturing,” the survey noted.
This was quickly followed by finance minister Arun Jaitley’s announcement in the Union Budget for 2018-19 that Niti Aayog would develop a national programme of artificial intelligence and machine learning.
The Roy-led discussion paper is an outcome of that decision. Niti Aayog has decided to focus on building AI solutions in five areas that will potentially impact millions of lives: education, agriculture, healthcare, smart cities, and smart mobility. The government sees partnering with tech giants such as Microsoft, IBM, and Google as also a raft of startups as critical to its push.
“This effort will consolidate the government’s role in providing AI,” said Kamakoti Veezhinathan, computer science and engineering professor at the IIT (Madras), who also chaired the AI task force report by the ministry of commerce and industry. “Now we see so many scattered efforts amongst institutions… interesting research projects will come up.”
Farmers of the future
Pilot projects are on and early results are encouraging: Niti Aayog’s work IBM aimed at farmers, work that the Andhra Pradesh government is doing with Microsoft (again, in agriculture), and diabetic retinopathy projects across the country.
P. Anandan, CEO at Mumbai-based Wadhwani Institute for Artificial Intelligence, a research institute with an ‘AI for social good’ mission, puts Niti Aayog’s paper in perspective for us. Despite India’s widespread adoption of mobile phone services, it has some 800 to 900 million people who are really not beneficiaries of mainstream technology. “Particularly, if you go to rural areas, the human capacity to serve needs there is quite limited. For example, there are not enough doctors. Small farmers don’t have the expertise to plan what they have to do appropriately and, therefore, are at risk,” he said.
India has the curious situation of people who are literate but still are not capable of fluent reading and writing, Anandan, a former director of Microsoft Research in India, added. That’s a hard problem to fix but not an unsolvable one. “Making machines usable by people speaking a wide variety of languages will itself be a breakthrough,” he pointed out.
Take agriculture, for instance. More than half of India’s population is engaged in farming but the sector contributes just 16% to the country’s gross domestic product. Niti Aayog has partnered with IBM to develop a crop yield prediction model and provide real time advice to farmers using AI. It aims to help the farmer with warnings of pest control and crop disease outbreaks with help of data from remote sensing (provided by ISRO), soil health cards, Indian Meteorological Department’s weather prediction, and other data such as soil moisture, temperature, crop phenology, among others.
“We have the weather company with us. With Watson and weather we have the ideal combination to go and help the government and farmers at large to build the model,” Viswanath Ramaswamy, director – systems at IBM India and South Asia. “The end-consumer, in this case the farmer, would not even realise that he is using AI.” The project is being implemented in 10 districts of Assam, Bihar, Jharkhand, Madhya Pradesh, Maharashtra, Rajasthan, and Uttar Pradesh.
But IBM and Niti Aayog are the not only ones using AI in agriculture. Microsoft started piloting deployment of its AI technologies with 175 farmers in Andhra Pradesh two years ago in June 2016. A Microsoft case study details work in Bairavanikunta village. Farmers there received messages on sowing date, tips on how well the land should be prepared, advice on the quanity and timing of fertiliser to be used, among other things. Crop yields improved 30%.
It wasn’t an easy task. The region’s climate data of 30 years, from 1986 to 2015, was analysed using AI. Then the moisture adequacy was calculated. The success led Microsoft and ICRISAT (Microsoft’s partner, which provided on ground data, in the programme) to take the pilot to 3,000 farmers in 2017. The results were the same: yield increased by 10% to 30%.
According to Niti Aayog, farmers get as low as 20% of the price paid by an end-consumer for fruits and vegetables – largely because of ineffective price discovery, supply chain inefficiency, and local regulations. AI can help solve those problems by improving crop yield with real time advisories, reduce crop failure with advance information of weather events and detection of pest attacks, spread best practices around sowing and farming, and help with a better prediction of crop prices. “Why should a farmer know it is AI as long as it solves his problem,” said Anant Maheshwari, president of Microsoft India. For the farmer, the solution in this case came via a series of text messages.
To make predictions of impact, Roy said, AI algorithms need data. Data from e-NAM, Agricultural Census (with data on over 138 million operational holdings), AGMARKET, and over 110 million soil health samples will help in predictive modelling.
“More data is better. As more data (is analysed) the output gets better,” Maheshwari said. Roy said that the government is already looking at solving the data problem by creating what is called a “marketplace”. We will come to that in a bit, but before that the government’s efforts on driving change in healthcare using AI.
For a healthy Bharat
The scale of growth is massive. By 2020, the healthcare industry in India is expected to grow to $280 billion from $100 billion in 2017, according to the Niti Aayog discussion document. But, still there are two problems: shortage of qualified healthcare professionals and non-uniform access to healthcare across the country.
To put things in context, India has 76 doctors and 209 nurses for every 100,000 Indians – about 20% to 25% short of recommendations from the World Health Organisation (WHO). Technology can help bridge some of the gaps.
Niti Aayog is in talks with Microsoft and medical technology startup Forus Health for early detection of diabetic retinopathy. WHO estimates that about 19% of the world’s 422 million diabetics are in India. One in three diabetic patients have some sort of diabetic retinopathy – a condition where blood capillaries burst in the retina causing black spots, which progressively debilitate vision.
The Microsoft-Forus partnership has already been deployed at eye check camps in villages. The problem with mass eyecare checkups is that the images taken during the camps are not often usable for diabetic retinopathy. AI prompts the person, who is clicking the picture to take another one if the first one isn’t usable. This allows a less skilled person to do the job.
“We started developing technology on the eye three years ago. We had started working on contract surgery to drive better diagnostics and manage better outcomes. The idea is to create a large body of work to use that data in other eye-related use cases,” said Maheshwari. Diabetic retinopathy is being addressed using AI by other players including Google, Eyenuk, and IDx, too.
As far as Niti Aayog is concerned, Roy said, “India doesn’t have enough supply of pathologists. AI can help in doing predictive pathology.” A semi-trained person can collect the samples in an remote area and upload images to the cloud. “You would not require a pathologist,” Roy added.
Government health centres can play a role here. “With very little effort you can train these foot soldiers. You can train the nurses with very simple bridge courses. Being an arts student, you can be trained to be a nurse in the village. This will open up new job opportunities,” Roy said.
A new opportunity
While there is a lot of fear that AI will eat up jobs, Roy believes that it will be the other way round. That is also the reason why Niti Aayog’s document focuses on skilling.
She points out at some numbers from IT services industry body Nasscom: by 2022, 46% of the Indian workforce will be engaged in entirely new jobs that do not exist today. She also said that the demand of AI specialists in India will grow by 60% by the end of this year.
The Niti Aayog discussion paper recommends incentivising creation of jobs that could constitute the new service industry – jobs that would ideally be a part of the AI solution development value chain but require a relatively low level of expertise so as to create employment at scale. Work like data annotation can employ a large quantum of human resources.
The other one is through recognition and standardisation of informal training institutions. Bridge courses help people getting AI-trained. According to the discussion paper, tech-hubs like Bengaluru have many traditional IT training institutions establishing courses in new age technologies yet they are not standardised.
Creation of open platforms for learning, like Nasscom’s Future Skills Platform, and online and self-learning platforms such as Coursera and edXare, will help. “India is substantially behind developed countries like the US and China (in frontier technologies),” said Mohan Lakhamraju, founder and CEO of Great Learning, a company that provides online courses.
“Everything has to be driven by implementation. In our country, people having expertise in deep learning… the number won’t be more than 10,000,” he added.
Roy felt that AI can come in handy in education, too.
The gross enrolment ratio in India is 96.9% at elementary school and 80% at secondary level, as per recent figures. But then the dropouts start happening. Retention rate is at 70.7% at elementary level and 57.4% at the secondary level. The National Achievement Survey (NAS) of November 2017 shows that there is increasing concern about the poor learning levels of children in school: less than 60% in most states.
According to Lakhamraju, AI can help track a student’s progress and predict what might likely happen with performance. For that, schools and colleges will have to co-opt the technology. “AI is as good as the data it gets,” he said. “There is a long way before it goes into villages… schools need to get internet first.”
Microsoft’s Maheshwari was a little more optimistic. “If you are able to engage students and their families better, you can reduce dropout rates.” He said that there are enough signals to predict the risk of dropouts: grade, attendance, health of the student, among other signs.
In a pilot done in Andhra Pradesh, Microsoft used 100 different parameters to predict if there will be a dropout. This was done for 100,000 students and is now being expanded to 600,000. “If we can reduce the dropout rate by 20%, it will help in overall development of the country,” Maheshwari said.
But, none of this is viable without data.
Marketplace for data
Niti Aayog has proposed what is called a “marketplace”. A three-pronged, formal marketplace could be created focusing on data collection and aggregation, data annotation, and deployable models. There could be a common platform called the National AI Marketplace (NAIM).
For that, the data – available with various government bodies – needs to be organised. Once that is done, the idea is to open up the data to build AI solutions in those five areas chosen by Niti Aayog. “We have to balance that out with ethics and privacy. There is technology available that without infringing on privacy. Also, we don’t want privacy to be over hyped in a manner where data doesn’t come in use for greater good,” said Roy.
A good governance structure for marketplace is also needed. “We want to promote startups , so we have proposed ICTAI, where startups can develop AI tools. A lot of startups will come up in data annotation, data generation. That is why we have proposed the marketplace,” said Roy. ICTAI is short for International Centres for Transformational Artificial Intelligence.
Niti has collaborated with universities such as NTU (Nanyang Technical University), Stanford and Carnegie Mellon, with the vision of international collaboration to get research done in India. But the real work will start when the government can throw a challenge that there is a problem that the country is facing and an AI solution needs to be build to solve it. ICTAI, which is lead by private research, Roy felt will collaborate with industries to solve some problems. “Industry is much better in doing it and they should collaborate with academia and other agencies in doing it… Government should come out with expression of interest. They have to come up with specific AI tools, to solve specific problems, in specific areas,” she added.
We have proposed government funding for this projects, she said. “We are not giving financial estimates because that will need wider consultation.”
Estimates are that investments in the beginning can go up to $250 million — each project costing anything between $10 million and $50 million.
In the social sector, there is no single party who is equipped to do what tech giants (such as Google or Facebook) have done, said P Anandan of the Wadhwani Institute
In the social sector, there is no single party who is equipped to do what tech giants (such as Google or Facebook) have done, said Anandan of the Wadhwani Institute. “Inherently, we’re going to rely on some kind of common ecosystem for data gathering, sharing, and use,” he added. “There has to be an ecosystem of social sector organisation partners, government organisations, and institutes like us, who have to come together.”
Roy said that implementation is the key for success of AI in India. Those are still in plans. “One consultation has been done on the strategy. Now will be looking at implementation,” she said.
Each recommendation would have to be taken up on its own – it would need consultation with academia, government agencies, and other stakeholders. For rolling out AI projects, one needs to consult with local bodies also. “This is a macro matter – whether it is a civic body or a state body will depend on what problem we are solving,” said Roy.
There are three pillars which will be foundational pillars for AI in India: technology advancements, collaboration, and adoption.
“We are not inward looking. Even other countries, which are evenly placed, have the same challenges – tools developed in India can come in use there also. We are looking at Africa, South East Asia, Latin America, where ever there are these problems,” Roy said.
The big push will come when the government will announce the big challenges. That is something we are working on. China, Roy said, just announces it and goes all out. “Ours is a democracy. We have to take everyone along. Without the states on board, we cannot implement anything because health, education, are all state subjects,” she said.
Subscribe to FactorDaily
Our daily brief keeps thousands of readers ahead of the curve. More signals, less noise.
Disclosure: FactorDaily is owned by SourceCode Media, which counts Accel Partners, Blume Ventures and Vijay Shekhar Sharma among its investors. Accel Partners is an early investor in Flipkart. Vijay Shekhar Sharma is the founder of Paytm. None of FactorDaily’s investors have any influence on its reporting about India’s technology and startup ecosystem.