The Problem With India’s ‘AI for All’ Strategy

The government’s slogan seeks to position India as a leader in artificial intelligence but ignores crucial societal questions that will determine its effectiveness for the majority of Indians.

Expert comment Published 18 February 2019 Updated 26 August 2021 2 minute READ

Dr Urvashi Aneja

Former Associate Fellow, Asia-Pacific Programme (based in India)

The offices of SigTuples, an AI startup focused on improving healthcare, in Bangalore. Photo: Getty Images.

The offices of SigTuples, an AI startup focused on improving healthcare, in Bangalore. Photo: Getty Images.

With its brand of ‘AI for All’, India is seeking to position itself as a leader in the global AI race, innovating, testing and deploying solutions to address development challenges in the global south. The government’s official think tank, Niti Aayong, released a paper last year identifying five sectors for AI intervention in India – healthcare, agriculture, education, smart cities and smart mobility – and the government recently announced the creation of a National Centre on Artificial Intelligence.

But AI is not a silver bullet that can magically or autonomously address complex social problems. In fact, it can exacerbate existing socioeconomic inequities, lead to a concentration and collusion of power and even reconfigure the fundamental tenets of democracy. And while these issues are of concern across the world, two further challenges will complicate India’s vision of AI for All.

First, AI relies on the collection and analysis of data. However, existing data sets in India, whether for labour markets or health systems, are fragmented, unrepresentative or outdated. Further, there are large digital divides, for example between urban and rural areas and between men and women. Less than 30% of India’s internet users are women and only 14% of women in rural India own a mobile phone. Algorithms based on existing data sets will thus undoubtedly have a distorted picture of social reality, blind to the behaviours, needs and experiences of numerous social groups.

This could have particularly damaging consequences in precisely the sectors outlined by the government. For example, the Manipal Hospitals Group recently teamed up with IBM’s Watson AI to aid doctors with the diagnosis and treatment of cancer. But physicians noted that there was a strong risk of misdiagnosis, as Watson was trained on data from American patients which did not translate to a new context.

The rapid growth in the number of Indians online – one report projects 850 million internet users in India by 2025 – alongside the digitalization of government, financial and other services, could partially address this issue. But for most people, access to, and participation in, existing systems, whether for healthcare, social protection or employment, is through informal, unregistered and unaccounted channels and systems.

Unless these systems themselves transform or new means and metrics are found for capturing data, the challenge will remain. Getting more users online or digitizing access to public service will not create usable data sets by itself – this will require existing socioeconomic systems to transform and deeply embedded behavioural patterns and social practices to shift.

There is then the further issue of data privacy. India is yet to put in place a data protection framework; existing drafts rest on the idea of informed consent, but this seems far from adequate given the low-levels of literacy and education of much of the population.

Second, the deployment of AI solutions in industry will disrupt labour markets in India, to the detriment of a bulk of the labour force. The reduction in the cost of intelligent automation is already resulting in the re-shoring of numerous industries to industrialized economies in the global north. This will make it increasingly difficult for India to generate employment through an export-oriented manufacturing strategy.

This poses a particular challenge for India, given that a large part of its population is low-skilled, and thus traditionally best absorbed within large-scale manufacturing industries. Further, it is unlikely new job creation can offset such losses. The people who lose their jobs are unlikely to be the same ones to take up newly created jobs – a middle-aged low-skill worker will find it very difficult to re-skill or up-skill fast enough. The newly created high-skilled jobs are likely to be significantly fewer in number and unable to absorb India’s large labour surplus.

Instead, a new class of low-skill and low-paid jobs are being created to fuel an AI world – from data annotators to content moderators. Indian workers are one of the largest contributors to online micro-work platforms such as Amazon’s Mechanical Turk, to which many American workers displaced by technology have already reluctantly turned for repetitive, sub-minimum-wage work.

In the absence of thinking about both technical feasibility and social viability, there is a strong risk that AI-based technology gains are likely to benefit only a select few Indians. In this context, the ‘AI for All’ narrative obscures rather than answers many of the fundamental challenges that India faces.