The Case for AI4Bharat
AI4Bharat is a voluntary community founded by us - Mitesh Khapra and Pratyush Kumar. We both are faculty members at the Computer Science and Engineering department at Indian Institute of Technology, Madras. We have also co-founded One Fourth Labs (incubated at the IIT Madras Research Park) with a mission to design and deliver affordable hands-on courses on AI and related topics.
AI - Promises and un-promises
Artificial Intelligence (AI) is one of the defining technologies of our age. Its progress, especially in the last decade, has been remarkable. Aided by abundant data, cheaper compute, and advances in algorithms and hardware, the sub-field of Deep Learning (DL) has enabled machines to reach human-level performance on specific tasks such as image classification, speech recognition, gameplay in games such as Go and DoTA, and several others. Tech corporations have earned enormous commercial success in deploying AI in ubiquitously used products and services.
This progress of AI is expected to continue robustly. Fuelled by increasing investment in research and development, more talent is focussing on technical advances in AI. Also, our data sources continue to grow richer with more people joining the internet and more sensors sensing our physical world. Tasks as varied as designing drugs, driving vehicles, composing music, and simulating ecosystems are expected to benefit from increasing automation with advances in AI. This is the promise of AI.
With this backdrop, AI technology to most of us would appear magical, in the sense of Clarke’s famous third law. And magic naturally invokes an other-worldly awe. A magical AI is expected to be potentially all-powerful: We expect AI to understand language as we do, or have an open-ended conversation with us, or provide emotional support, or even take away half our jobs. No, there is no conclusive evidence that any of this is happening, immediately or in the near future. Even technically, there are valid debates ongoing on whether DL is reaching a plateau and needs to be supplemented with other methods such as symbol manipulation. So, AI is not a panacea. It is a tool, that is becoming ever more sophisticated. Reading it as anything more than that can detract focus from the tactical efforts of deploying the tool in the service of humanity.
AI4Bharat is founded with this tactical focus. We aim to build AI solutions to the problems of Bharat, today. We aim to bridge the wide gap between demonstrated AI technologies and our real-world challenges across domains of agriculture, healthcare, smart cities, digital India, and sustainability. Apart from bringing the latest technologies to these domains, the promise of AI can uncover data troves in these domains. These troves of data can drive efficiency, transparency, and equity. This is the promise of AI4Bharat.
Setting up the AI Platform
Platforms at Indian railway stations are happening places. The government sets up these platforms - the concrete structures, but then others come and setup book stalls, food stalls, water depots, porter services, and even WiFi services (recently setup by Google). This is the power of platforms: They have a simple beginning, but evolve remarkably over time as they are designed for participation and open innovation.
These design principles of the railways platform are our design principles in AI4Bharat. We are a platform, our concrete structures comprise of (a) emergent networks for domain experts to connect with AI engineers, (b) efficient processes for AI engineers to collaborate on projects, and (c) accessible resources for learning about AI. This platform then becomes an invitation for government bodies, engineers, academicians, and domain experts to pool in their expertise to build concrete solutions. The diversity and engagement across this diverse group will dictate the vitality and the impact of AI4Bharat.
And then there is the matter of nation building being a team sport. A sport where the role of government is best as a facilitator. AI4Bharat’s members representing an empowered group within the nation, are expected to lead this effort of nation building. Each member of this group will have their respective passions of contributing to the nation shaped by their experience and expertise. The rise of AI provides an occasion and reason to pool together these passions and direct it at solving concrete problems. AI4Bharat is thus a platform, on the journey to nation building.
Futurist Ray Kurzweil famously said, “fundamental measures of information technology follow predictable and exponential trajectories”. We do not see this is as an absolute law, but there is more than a grain of truth in that statement demonstrated by Ray’s 86% success in his predictions.
AI4Bharat is challenged to succeed with this predictable and exponential trajectory. We expect significant initial effort in setting up the said structures of our platform, in getting the first projects out. But we aim to benefit from the law of accelerating returns in the number of volunteering AI engineers and engaging domain experts, in the number of solutions we are able to demonstrate, in the number of countrymen we are able to impact.
While a trend provides a fashion, an outfit needs specific targets. Here are our targets for the first year of AI4Bharat:
Grow to a community of 100 selected AI experts and 50 engaged domain experts.
Establish best practices in collaborating-in-the-wild on problem discovery and building AI solutions.
Deploy at least 3 solutions in the wild targeting 3 different domains of focus.
Author 25 definitive pieces chronicling opportunities and experiences in practical deployment of AI in India.
We believe that the pace of innovation in AI for good can be significantly increased by systematising the process of problem discovery and solution engineering. In addition, there is need for education and robust commentary. AI4Bharat is built around these four functions:
Discover - Problem Finding
Problems exist in the real world and datasets exist in silos, both generally inaccessible to AI engineers. AI4Bharat is a platform for problem discovery wherein organisations and individuals post templated problem statements, provide accompanying datasets, and identify the corresponding value generated. Each problem is accompanied by community-led discussion to refine the problem statement and find connections with other problems or datasets.
Build - Solution Engineering
A large success of AI has been due to democratised technology. However, the effect of this democratisation is not readily accessible to problem solvers on the ground. Just as our IT industry succeeded in software engineering at scale, for AI solutioning we need to build the right processes: processes which enable distributed and cross-cutting teams to innovate effectively on AI solutions. This function will provide a streamlined pipeline combining access to cloud resources, design patterns, and practices for AI project management including reproducibility and maintainability.
Educate - Capacity Building
With the number of research papers in AI and related areas growing exponentially, it is hard for even researchers to keep pace with the latest findings. There is thus an acute need for innovations in pedagogy and delivery to ensure that engineers constantly update their practical and theoretical knowledge on AI. Beyond this, access should be provided to domain experts and policy makers on broadly understanding the technology and its potential impact. Misunderstanding the technology is a reality, and this can be very counterproductive.
Comment - Opinion Pieces
Framing opinion about AI at all levels - for the general public, domain experts, technologists, programmers, and government - is of high significance. We aim to robustly review the state of current adoption and future opportunities in domains such as healthcare, agriculture, vernacular languages, and finance. For instance, what are the opportunities of deploying AI to enhance the world’s largest employer the Indian Railways? Questions like these need to be definitively answered. Further, we also aim to provide balanced perspectives on topics such as AI ethics, potential job losses, and role of governance in AI.
1. keep it real | solve, today
AI4Bharat is oriented towards action. We will focus on solving real challenges, today. This requires careful and constant balancing between state-of-the-art techniques with practical domain knowledge and deployment. The latter is at times found not to be as glamorous, but is perhaps the most crucial piece in creating impact.
2. do it together | open community
AI4Bharat aims to create significant network effects in innovation in AI. This requires that people from different organisations and backgrounds join the community, have an effective communication, and build consensus on action plans. We will always prioritise open innovation with the imperative of the greater good.
3. impact at Speed | exponentiate
AI4Bharat is challenged to act at speed, to match up to the rapid pace in AI development and the gravity/scale of challenges Bharat faces. This requires thinking big and setting the right goals to work towards as a community. We will also build processes and platforms to benefit from the exponentiation power that comes with more data and partners.