A plan for India to carve its own space in the AI dominance race4 min read . Updated: 13 Dec 2020, 08:54 PM IST
We needn’t obsess over algorithm development for artificial intelligence so long as we focus on making the best use of its tools
Today, India lags behind significantly in the global artificial intelligence (AI) race, compared to the US or China. Indian universities or institutes are not yet top-notch AI-research entities, either measured by citations or other parameters such as winning prestigious AI contests. Most recently, Chinese scientists have demonstrated and claimed “quantum supremacy" in a race that seems to upend Google and the US.
However, it is not too late, nor is it to India’s disadvantage not to have top-notch algorithm developments taking place outside the country. The reality is that most useful algorithms make it to the public domain quickly, leading to their wide availability.
Algorithms are not hard to get, data is.
Indeed, one could argue that there could be advantages in not getting into the algorithm race, at least as a short-term winning strategy. We can let others do the complex and expensive job of creating these algorithms, and focus on India-specific use cases and data instead.
India’s national goal should be to turn data into useful “knowledge", specific to solving our own problems.
In data science, “knowledge" comes from general-purpose algorithms and good-quality data. Creating good quality data, which can be anywhere between 5–30% of the total data (and rarely more), takes up the majority of preparation time for any AI work.
Availability of a large amount of country-specific, contextual and clean training data could be a major source for creating useful and actionable knowledge about data.
Hence, rather than focusing on winning the algorithm race, India should focus on creating data infrastructure in key areas. Think of data being the new oil (you would have heard this cliche before), and an algorithm being the refinery. Oil here is the asset, and the latest refinery can be always be bought or imported.
The real win for India would be to create data infrastructure that will improve the quality of life at an everyday level, and significantly enhance governance. This would elevate interoperability, which in turn would reduce operational costs and serve as a “template" for developing a post-covid global AI economy.
This will a good fit for India, given its strength in process-driven white-collar and technology jobs, a heritage that goes back to the days of early call centres, business process outsourcing entities, and software development industries. Collectively, they trained millions of people and built formidable, world-class industries that powered the developed world.
Post-covid, this proposition is even more important. A jump in data usage in India during the covid pandemic has resulted in a significant increase in the amount of data in existence. Digital data in India was around 40,000 petabytes in 2010, and this number is projected to shoot up to 2.3 million petabytes by 2020, twice as fast as the worldwide rate. Also, the government’s Atmanirbhar Bharat (self-reliant India) initiative is focused on the localization of manufacturing, products and technology, including citizens’ data.
The key to success would be to solve the most pressing problems in the private sector, public and government services, and taking a five-year view to plan and fund these.
Projects for the private sector could be: Address, Identity and Gender Disambiguation. Doing so could make the delivery of everything from food and parcels to medicines more efficient, and lower the interest rates on loans. Gains through such efficiencies can add up to 1% of India’s gross domestic product (GDP). Our estimate is that building an AI-based product to identify a location correctly for an address in India can add almost 0.5% of GDP, or about $12 billion.
Projects for public services could be: Optimization of public transport in the post-pandemic world, using sensor-based data collected as part of the Indian government’s smart city programmes. Imagine the impact of even a 5% annual additional improvement in traffic capacity for the next 10 years in Indian cities that are plagued by traffic delays and productivity problems.
Projects for government services could be: Integration of public services and spending to ensure predictive usage of scarce resources for growth.
India has also built a world-class payments platform, the Unified Payments Interface (UPI). The results are dramatic. More UPI transactions were done in the past 18 months than through the country’s credit cards in 18 years.
Besides uplifting hundreds of millions of people, these planetary-scale platforms are enabling over 9,000 startups to build products and services on top, thus expanding the digital economy rapidly.
If India can build public infrastructure for the unique digital identity of a billion-plus people (Aadhaar), an open payment system (UPI) that has enabled direct payment of government benefits to hundreds of millions of its poor, then it can do the same for public AI infrastructure as well.
The opportunity to chart a new AI future is now. Each country will chart its own unique course in the AI race. India’s strength lies in its large workforce that can take part in collecting and codifying data for consumption and in creating large-scale technology infrastructure.
We should focus on that, rather than on algorithm games like the US or China. While doing so, we should also take advantage of technologies that protect user privacy and enable the generation of unbiased AI decisions to ensure that the outcomes and recommendations of it are acceptable in a country as diverse as India.
The journey is only 1% done so far, and the clock resets every single day, enabling us to start our AI race afresh.
Santanu Bhattacharya is chief data scientist at Airtel, visiting professor at Indian Institute of Science, and a former Nasa scientist