Technology has always exerted tremendous influence on society and led to improved human productivity. According to a report by Barclaysanalysts, if human productivity was 100 units in 1765 it has increased to 3,000 units today (Barclays, 2018). In fact, it has doubled in just the last five decades and this steep increase coincides with the adoption of information technology—personal computers, software, internet, e-mail, mobile communications and more (Keller, Wieladek and Shelepko, 2018).
We are now in the next phase, the fourth industrial revolution, which is powered by advances in artificial intelligence and machine learning (AI/ML) (World Economic Forum, 2015) and brain science and is blurring the lines between the physical, digital, and biological domains.
To understand where India is placed in shaping the next revolution, itihaasa Research and Digital conducted landscape studies of AI/ML and brain research in India, interacting with over 55 Indian researchers to identify important focus areas. AI in India is believed to have begun with a course being taught at Indian Institute of Technology (IIT ), Kanpur as early as the 1960s. Subsequently, India initiated the knowledge-based computing systems (KBCS) project in 1986 as part of its Indian Fifth Generation Computer Systems (FGCS) research programme. Today, the national institutes of importance like Indian Institute of Science, Bangalore (IISc), International Institute of Information Technology (IIIT)-Hyderabad and the IITs are the top universities for AI/ML research in India. itihaasa estimates that there are about 50 to 75 principal investigators/researchers in the domain in India. itihaasa also analysed published bibliometric indicators database pertaining to India over the period 2013 to 2017 (Scimago Journal and Country Rank Data, 2019).It was observed that over five years, India ranked third in the world in terms of the number of citable documents in AI, and ranked fifth in terms of the number of citations (Koshy, 2019). While this is encouraging, we should also acknowledge the big gap between the top two countries and India as far as AI research is concerned. For instance, while the USA produced 2.7 times the number of citable documents and were cited 5.0 times more often, China produced 3.1 times the number of citable documents and were cited 6.3 times more often.
Based on the interviews among researchers, different domains where AI/ML research could be meaningfully applied were identified—healthcare, financial services, hi-tech communications, retail, education, agriculture, smart city and transportation. Also,India-specific challenges such as the monsoon prediction and Indian language processing figure prominently.
It is well understood that neuroscience research will have a bearing on AI/ML development. However, less than 10 per cent of the Indian AI/ML researchers currently have an active research project involving neuroscience/computational neuroscience. This contrasts with neuroscience researchers in India, where at least 40 per cent were found to be undertaking research that combined neuroscience and AI/ML (itihaasa Research and Digital study).
The focus of Indian AI/ML research can be categorised as those related to sensing—document analysis and computer vision; comprehending—natural language processing and probabilistic decision making; and, responding—creating a complete situated learning agent and disease incidence prediction. The emerging themes of research are:
Unsupervised learning: AI/ML research requires curated and labelled data as opposed to humans, who do not need labelled dataset to learn. New research is likely to focus on learning without supervision and creating a never-ending stream of learning agents.
Reinforcement learning: A child learns by trial-and-error. New research around reinforcement learning will use rewards and punishment as signals for positive and negative behaviour of the agent. Another area of research is imitation learning, with humans in the loop, which makes it possible to teach agents complex tasks with no explicit programming.
Explainable AI: We do not know exactly how AI systems arrive at their decisions and in a sense it is considered a black-box. Their algorithms may be inadvertently influenced by human-biases and may thus adversely impact people’s lives, especially when it comes to applications like medical diagnosis or risk assessment for loan disbursement. Researchers are increasingly focussing on ‘explainability’ or ‘interpretability’ and study the ethics and fairness of AI/ML.
Causal modelling of AI: The current statistical-mode of machine learning systems will combine with causal reasoning tools in defining the next paradigm of AI. Developing a common-sense in AI will become the next target for researchers.
Resource-efficient ML: An autonomous car needs to draw its intelligence from a local apparatus as opposed to a cloud, in cases where it needs to react instantly, such as application of brakes. New research will develop algorithms that will make edge devices and Internet of Things (IoT) sensors smarter and address issues of bandwidth, latency, privacy, and battery power.
AI and blockchain: Various business objects collect onto a block chain and the data typically belongs to multiple owners. New research will focus on how data can be made secure to enable a confidentiality preserving AI.
Interviewed researchers identified key challenges that they have to overcome in order to achieve a global impact in AI/ML research. Amongst the topmost concerns ranks the need for quantity and quality in students who enter AI/ML research in India. This is closely followed by a need for improved computing infrastructure. The researchers also identified resource and administrative bottlenecks, lack of good quality labelled data sets and a siloed research approach within a university that calls for urgent intervention. Although, Indian AI/ML researchers and universities seem to receive adequate funding support from both government agencies and the industry, the lacunae mandates a change in approach. Also, about three-fourths of the interviewed researchers had collaborated with universities and institutions other than their own as well as researchers outside India. Nearly 90 per cent of the researchers had active or future collaborations with the industry.
In the light of the above, itihaasa made six recommendations for furthering AI/ML research in India. First, to increase the number of PhD students in India. Creating a special fund by instituting research fellowships for PhD and post-doc students and developing programmes to inculcate an interest in AI/ML in the undergraduate students with an active intervention of world class faculty will help bolster the academic strength in AI. China in fact,has launched a five-year university programme to train at least 500 teachers and 5,000 students working on AI/ML technologies, which can serve as a reference point for policy makers. It plans to develop 50 world-class teaching and research institutes, 50 national-level high quality online open courses and 50 AI/ML facilities by 2020 as part of the ‘AI+X’ programme (National Strategy For Artificial Intelligence, 2018 a).
Second, there is a need to augment computing infrastructure for AI/ML research by setting up a national high-performance computing infrastructure that is rich in graphic processing units (GPUs) and specialised-hardware for AI research, while encouraging the capability to design and make such computing systems in India. The NITI Aayog report mentions a proposal to set up a national computing infrastructure Artificial Intelligence Research, Analytics and Knowledge Assimilation Platform (AIRAWAT), a 100-petaflop super computing system for AI/ML applications (National Strategy For Artificial Intelligence, 2018 b). Also, SHAKTI an open-source hardware design and processor development is an initiative by the Reconfigurable and Intelligent Systems Engineering group (RISE) at IIT Madras with funding from Ministry of Electronics and Information Technology (MeitY), aimed at developing industrial-grade processors (Shakti Processor Programme, 2018). It has released India’s first indigenously-built microprocessor.
Third, there is a need to create India-specific AI challenges, tools and data-sets by focusing on India specific problems that affect a large number of citizens by creating resource repositories for research. India is beginning to understand the power and importance of data. For example, a computer vision algorithm in an autonomous vehicle working at about 80 per cent efficiency in developed countries, may work only at about 40 per cent efficiency on Indian roads. Researchers in IIIT Hyderabad are capturing data on Indian road conditions and are planning to use it for training autonomous vehicles. Another initiative is a project on Indian languages by NITI Aayog, IIT Patna and IIT Bombay. The initiative aims to create a repository of basic tools and resources for Indian languages. Any industry or startup wanting to develop an Indian language application can access this repository (Sharma and Verma, 2018).
Fourth, India urgently needs to set up a number of centres of excellence for AI/ML research which are multi-disciplinary, even from engineering and social science disciplines, facilitating a close interaction with the industry. One such multi-disciplinary AI research centre involving 24 faculty from eight different departments is the Robert Bosch Centre for Data Science and Artificial Intelligence (RBC-DSAI), at IIT Madras. Similar centres of AI are in the process of being set up in IIT Kharagpur and IIT Delhi.
Fifth, the adoption of an AI grand challenge approach to channelise the efforts of different researchers from multiple disciplines for a common purpose is urgently needed. The four grand challenges for AI in India identified by the AI Task Force, constituted under the aegis of the Ministry of Commerce and Industry are improving manufacturing, especially in the small and medium-sized enterprises; improving healthcare quality, reach and cost; improving agriculture yields and profitability;and, improving delivery of public services (The Artificial Intelligence Task Force, 2018).
Last, institutional mechanisms needs to be linked to startup ecosystem to strengthen the academic incubators and help translate AI/ML research to market applications. Many of India’s premier institutions such as the IITs, IISc, and the IIITs have strong academic incubator programmes. At the IIT Madras Incubation Cell, about 7 per cent of all companies incubated have AI/ML as a core technology. Also, the Atal Innovation Mission (AIM) under NITI Aayog has an ongoing scheme to support academic incubators (Kant, 2019).
At a time when the AI revolution is poised to take the world by storm, India’s preparedness mandates evaluation. The country’s pre-eminent institutes have led the charge when it comes to research in AI/ML and the output of published research is commendable. At the same time, it must be kept in mind that China and the USA have a huge lead over India. There are several challenges to AI research which include computing infrastructure, resource and administrative bottlenecks among others. Nonetheless, there have been active inter-university collaborations and collaborations with the industry. An AI grand challenge approach is required to channelise research from all disciplines towards a common purpose. Rounding up, institutional mechanisms must be linked to the startup ecosystem.