It’s an Important Inflection Point in the Journey of AI says Sriram Raghavan Vice President IBM

IBM and IIT Bombay team up to accelerate Al research in India

L-R (Sriram Raghavan, Vice President, IBM Research – India & Singapore & CTO IBM India/ South Asia, Arvind Krishna, Senior Vice President, Hybrid Cloud and Director, IBM Research, Prof. Devang Khakhar, Director, IIT Bombay, Prof. Uday Khedkar, Head of Department of Computer Science and Engineering, IIT Bombay and Prof. Soumen Chakrabarti, Department of Computer Science and Engineering, IIT Bombay)

Sriram Raghavan VP IBM Research & CTO, IBM India, talks about Research and Development in AI in collaboration with IIT Bombay, a way ahead towards India’s AI Revolution in a Tet-e-Tet with Priyaankaa Mathur from Businessworld 

Q Please elaborate on the research collaboration between IBM and IIT Bombay. 

In IBM we have a global Academic collaboration Program for AI which we call as AI Horizons Network, the focus of the program is around advancing the Science technology and applications of AI and we are happy to announce that IIT Bombay is the first institution outside North America to join us along with eight leading global universities,that are working to accelerate the development of AI technologies such as deep learning, natural language processing, computer vision and others. 

Collaboration with IIT Bombay came naturally firstly, because we have a long standing relationship with them, since we have had strong collaborations working with their researchers and faculty in many IBM academic research initiatives, secondly, data is the foundation of AI and IIT Mumbai is one of  the pioneers of Data Management Research in India and we feel that it would be a great foundation for joint AI research and thirdly, they are the premier institute in the country and get fantastic students. So the intent is that jointly with IIT Bombay faculty and students IBM will collaborate for the research in AI. 

Q Could you highlight on the application of AI on Businesses and IBM specifically? 

For us AI is one of those transformative technologies that will cut across almost any industry you can name. There will be rate and pace of adoption in different industries and different geographies, but in terms of its potential, AI will absolutely touch every industry.

While there is tremendous potential already there are scientific challenges that need to be solved. In IBM we believe we to discuss this evolution from what we call Narrow AI to Broad AI and that journey has lot of fundamental scientific challenges to solve, which is why partnerships like this come in place, which allows us to understand the fundamentals of Science. 

Q What do you mean by Narrow Al and Broad AI? 

Narrow AI is like, if you have a problem that is sufficiently well defined and you have lot of data with you. For example look at the AI that you sort of experience around you like facial recognition in social media, that has lots of data say billion images to recognise human images, or basic understanding of our questions from our smartphones and mobile devices, where there is certain level of machine translation of a particular language, although it’s not perfect, so Narrow works with lot of Data and AI can do pretty good job with a huge amount of Data better than human accuracy in some cases.  

Broad AI has the ability to apply in wider domains, work with less data, is trusted, bias free and explainable and that’s the kind of AI we are working in IBM that is essential for the journey from narrow AI to broad AI. Trust in AI is another example, you come across models that are biased and that bring about higher accuracy of data by characterising human faces on the bases of race or gender. We took literature of last 50 years to conduct a research and used a lot of AI to find that, there has been a lot of gender bias. You can imagine AI to be used to help you in loan credit decisions. By law most countries cannot discriminate on the bases of gender or races and so there are certain protective models, so how can we use AI for decision making, we need controls and understand where it could be biased and should be able to measure bias and test a model, which does bias and correct it.  So that’s the problem to be solved from Narrow to Broad AI. 

Q Who are your target segments for this collaboration, what is IBM’s AI strategy? 

Our AI strategy has two parts first, is to bring in the prominent AI platforms for our clients and partners for building solutions for transport, health care or financial services and hence to become an enabler to provide an AI platform for businesses.

In addition IBM has invested in key verticals like health programs looking at health or creating applications for financial service industry, or for research in India for the retail. So there are a set of focus sectors for us as a company, but we are also looking at a generic platform that allows other companies and partners with tools to access performance scale, accuracy, bias free data analysis and build AI models for whatever verticals, so it could be like a software solution provider, which is very broad. 

Q Where do you see the future of AI in India, is it very new in the current scenario? 

I think AI will keep us busy for the next couple of decades, and there are a lot of hard problems to be solved so, I don’t thing adoption is going to wait for it. I think we all are on this journey already and would like to emphasise that this is an important inflection point in the journey of AI, which determines what will allow AI’s impact which will be even broad and pervasive than it is right now, since we have all our technologies working on solving problems right now and for future. 

Q What are the areas of research you are working on in collaboration with IIT Bombay. 

In this context we have picked up some important themes within the umbrella first is How can AI explain a decision? Or how can AI tell there is a fault and explain it, it’s a kind of an AI software. There are lot of cases which requires explain ability.Since AI should be able to explain not only what the answer is, but why the answer is? So we have to work on fundamental scientific challenges with IIT Bombay to develop AI that can explain its decision. 

Second, the question is how you develop AI that can work across different modalities like speech, text, and image, carry context and that we as humans do it very naturally, say talking with someone sitting across the tables, but AI cannot do it. So how do you make AI systems that are multimode and why are we doing it at IBM is, that we represent knowledge in a very uniform way. 

Third,How is fundamental AI in the core applied in advance technologies, areas called deep learning, looking at how we accelerate deep learning? As part of IIT Bombay’s, super computer cluster, IIT Bombay also has machines from IBM. We develop these models on multimodal AI and how would it help to optimize them on that type of hardware. We are looking to publish scientific papers together with IIT along with research data sets for the world wide community. 

Q Could you elaborate on field of training and building domain specific AI agents Chatbots, which aid in human decision making? 

This is part of the journey from narrow to broad AI which is today building an AI model for a problem. For example building a customer service through chatbots to answer for an insurance product for an insurance providers. There are a lot of clients who want to install our chatbots, which could be on a website or on WhatsApp, it is the way to interact with the machines on the basic questions. So, you don’t need to wait for a customer care executive or a human interaction, to answer for questions like what are the penalties for a late payment of insurance premiums and that’s how most of our insurance clients want to start interaction with their clients. 

There is this technology called transfer learning,that works on how do we adapt AI to work from one domain to other domain, so if an insurance company has developed a chatpot to answer 4 questions for 4 products in a particular domain,then how can it quickly transfer it to next 4 products? Or if a company has done business at one part of India in 4 months and wants to start operations in say Africa, can it quickly deploy the next one in Africa in 2-3 months. So, with IIT we are working on what is the technique that allows to transfer a learning, when I teach an AI something, how can it adopt to others? 

Q Can you name  the top IIT Faculty and Researchers who are part of this collaboration? 

We will be working closely with Prof. Devang Khakhar, Director, IIT Bombay, Prof. Uday Khedkar, Head of Department of Computer Science and Engineering, Prof. Soumen Chakrabarti from the Computer Science Department, who is an expert in Data and Machine learning, Prof. Sunita Sarawagi, Prof. Preethi Jyothi, Prof. Ganesh Ramakrishna’s. These are all luminary faculty members with expertise in AI and Data Management, who we have identified to work on joint projects with IBM. IIT has faculty and students and we have our scientists at IBM Research, India, so we will work together on explainable AI and create a research agenda to work jointly. 


Tags assigned to this article:
ibm india artificial intelligence ChatBot


Around The World