Tell us about your interaction with AI and other intelligent technologies that you work with, in your daily life at Mindtree.
At Mindtree, I lead an exciting new business that we call the ‘Reimagination Business’. The idea is that we can use the Digital Transformation foundation that has been laid over the past 10 years or so in companies around the world and add AI on top of that to fundamentally reimagine the businesses of these companies. This involves changing their business models, products, business processes as well as the experience of various stakeholders such as customers, employees, and the ecosystem. In our client engagements, we see firsthand the power of AI working its magic to produce profound business transformation and its ability to liberate human talent that is currently chained to drudgery so that they can do more engaging work. As part of my role, I also have the pleasure of working with leading AI researchers at Stanford University that leaves me inspired every time I talk to them.
In daily life, I use the services and products of companies that have truly become AI companies. I search online, talk to Siri and Alexa, use Google Maps, click on ads, take online classes driven by AI, etc. Being blessed with the opportunity to live in Silicon Valley, I encounter a few self-driving cars from Waymo and others every day on my way to work. I am simply amazed by the strategic thinking, long-term time horizons, and the audacity of the valley companies to push the boundaries of what AI can do for all of us.
How did you start in this space? What galvanized you to join Mindtree?
I first learned about Neural Networks way back in 1988 when I was getting my Masters in Computer Science at the Indian Institute of Science in Bangalore. Unfortunately, at that time it was all theory. The promise was there but the computational power that was needed to do iterative learning with backpropagation within reasonable time frames was simply not there. Until about 2011, nothing much happened – it was the ‘AI Winter’. In 2011, large enterprise customers that I was working with started doing ‘Big Data’ projects. These projects started to lay the foundation for the data-driven AI revolution that was about to unfold. In 2013, I had the honor of meeting NVIDIA, founder Jen-Hsun Huang, in person and was struck by his foresight on what massively parallel processing is driven by his GPUs could do. Prof. Geoff Hinton’s team had won the ImageNet competition using Convolutional Neural Networks and NVIDIA GPUs the previous year.
By about the end of 2013, I had concluded that the ‘AI Winter’ was about to be over and had started incorporating in my client discussions that AI was the fifth Digital force going to transform industries (the other four forces being mobile, big data, social media, and cloud). Sure enough, by 2015-2016, with the advent of even more powerful GPUs and FPGAs, we finally had all the ingredients needed to break into ‘AI Spring’ – the massive amounts of training data, the low cost (relatively speaking!) and low power massive parallel compute capability with GPUs was there and it was time to become ambitious with what one can do with AI within the enterprise. In 2015, I directed global research on the adoption of AI across 13 different industry sectors that showed that the ‘AI Spring’ was definitely starting. I have seen tremendous growth in the adoption of AI in the industry since then. The article that I wrote for Harvard Business Review Digital based on our 2015 research has already become outdated – the pace of progress has been that fast in the last 4-5 years!
In mid-2017, I was looking to move to a new role after having built a Digital services organization from the grounds up for seven years. I could see the impending disruptive power of AI in many industries, including the one I am currently in – the IT services industry. My career experience told me that companies without too much legacy in older ways of doing things are likely to be the most supportive of my ideas of the transformation of the IT industry with the help of AI. Mindtree was such a company. I talked to the leaders of the company who agreed with my vision and I decided to join them.
How do you differentiate Mindtree from other similar service providers?
Mindtree is a ‘born digital’ company and has built a lot of credibility with customers from various industries for laying down the digital foundation very well. Some of the early decisions of the company made almost 20 years ago, have become very advantageous for the company now. For example, the company decided early on that it will not participate in the human-powered Business Process Outsourcing market. There are hundreds of thousands of people around the world using their cognitive capabilities to act as a mere glue between the physical world and the digital world or between disparate digital systems that cannot be integrated programmatically. As such, we are able to be daring in our solutions by bringing the maximum power of AI to help liberate human capital from low value-added work. We are true believers in the power of AI to transform our customer industries and our own industry as well.
How do you see the raging trend of including ‘AI’ in everything’ impacting businesses?
I absolutely believe that AI must be woven into the very fabric of enterprise applications, enterprise infrastructure and the business itself. Every business has the potential to become an AI-driven business. We are inspired by Prof. Andrew Ng and his vision of AI Transformation. We are also pragmatic and pay particular attention to what is possible with AI today and what is not possible. As Prof. Ng often says, the sweet spot is at the intersection of valuable business opportunities on the one hand and what is possible to be reimagined with AI on the other.
What are the biggest challenges and opportunities for AI companies in dealing with rising technology prices?
AI will increasingly get democratized. There is no doubt about that. Once the business opportunities are identified, the key ingredients for the success of AI are data, compute and electric power. I truly believe that training data is the new gold and am very excited by the global movement towards the wide availability of interesting data sets. Governments have a special responsibility to facilitate the availability of data, just like they need to ensure steady availability of reasonably priced oil for the energy needs of the economy. Universities and even private companies are also contributing. Compute will only get cheaper as time goes by. They will also become more power-efficient. The real challenges, though, maybe in the availability of trained AI professionals and the trend towards concentration of data-power in the hands of a few super companies.
How should young technology professionals train themselves to work better with AI, especially Robotics?
Learn, learn, learn. If there is one piece of advice I will forcefully give to any young professional, it is that they never stop learning. They should recognize that developing an AI system is fundamentally different from developing a traditional software system. In earlier systems, it’s all about the code. In AI systems it’s all about the data. The real intelligence in AI systems is in the data.
The code is a secondary part of the whole effort. This is a big paradigm shift. There are excellent courses available online such as in Coursera. Legendary master practitioners such as Prof Andrew Ng are teaching courses in these systems. These courses have excellent tools such as Jupyter Notebooks and even automatic code evaluations with unit tests, all of which can make sure the young technology professionals can reach peak performance very efficiently.
How do you consume information on AI/ML and related topics?
It’s all on the web mostly. I am grateful for the global AI community on being open about their research and publishing their work. IEEE, ACM, you name it, they are all great resources. I am also lucky to gain enormously from the Stanford AI community. A visit to Stanford is always inspirational. Interactions with Prof. Silvio Savarese, the Mindtree Faculty Scholar in the Stanford Vision and Learning Lab, upgrade my AI-quotient by a few notches every time I speak with him.
I learned a lot about Computer Vision techniques by listening to Prof. Juan Carlos Niebles whose research in activity detection from videos is very relevant to the challenges we hope to solve for our customers in reimagining business processes using AI. Prof Clark Barrett does very interesting work on using formal methods to ensure the safety of AI systems which is of interest for our industrial customers. What I have gained from Stanford researchers is also the audacity of their ambition in collecting massive data sets in a speedy and cost-efficient manner to drive advancements in modern AI.
What makes deploying AI so hard?
Deploying AI is hard because it completely overturns the traditional thinking of deploying software systems. Selling AI system is different, developing it is different, contracts are different, maintaining AI systems is different and in general, setting expectations in an organization about AI systems is very different. Let me give an example. Suppose we are trying to deploy traditional point-of-sale software in a number of retail stores, it is fairly straightforward. Once it has been developed and tested, the performance of the system is not going to vary from store to store. Now, suppose you have to deploy a convolutional neural network vision solution in the same set of stores. After you have developed it in the lab with available data, the performance of the system could vary from store to store. This is because, in each store, the background for the images could be different, camera angles could be different, occlusions could be different, and all of these affect the performance of the AI system. These have to be explained to the stakeholders.
Maintenance of traditional software systems is mostly about doing bug fixes on the code and adding features in the code etc. Maintaining an AI system is mostly about updating the training data with more current data, fine-tuning the hyperparameters, retraining the model and so on. It requires a very different orientation. It all comes down to a code-orientation in the traditional systems versus a data-orientation in AI systems. For the larger ecosystem, this is a pretty big shift and we see it every day in our work with our customers.
Which is harder – choosing AI or working with it?
Choosing AI is a no-brainer. You have to do it. When your competitor’s systems are infused with AI, and you work in the good old way, guess who will have the advantage in the market? Working with AI is harder because as explained earlier, the way of thinking about traditional systems gets overturned at every step of the process in the case of AI systems.
How potent is the Human-Machine intelligence for businesses and society? Who owns Machine Learning results?
There is an incredible power to the collaborative relationship between humans and machines. That has always been the case in history. It’s not going to be any different from modern AI. I believe that true Artificial General Intelligence that can rival human intellect is several decades, if not centuries, away. After all, we are talking about matching biological systems that have evolved over hundreds of millions of years whose inner workings we don’t fully understand yet.
Neural networks, though inspired by the structure of the biological brain, is just a baby step. Therefore, there is a lot to be gained by the partnership between humans and machines today. The big breakthroughs in AI were aided by a heavy dose of human help. This is particularly so in supervised learning. The ImageNet database with its 14 million+ images in 20,000+ categories was classified by humans so that machines can learn from it. This collaboration between humans and machines will continue for the foreseeable future.
Where do you see AI/Machine learning and other smart technologies heading beyond 2020?
There is a massive amount of investment dollars and human capital going into AI technologies and initiatives and these will bear fruit. AI will become more pervasive by getting adopted by a wide variety of industries, well beyond the so-called ‘Internet companies’. This will happen in everyday functions and business processes of these industries. AI will become invisible so that we won’t know if it’s there. This will happen by AI transitioning from human-level performance to human-level experience.
The Good, Bad and Ugly about AI that you have heard or predict –
The Good of AI is the profound impact AI is already having in bettering our everyday life in a non-obtrusive manner. Most of the interactions we are having in the digital world are influenced and impacted by AI systems. They help us buy the best stuff, provide us with the most relevant knowledge and info, remind us to do things, help diagnose diseases, discover new things about our environment, catch the bad guys, protect us from attacks, etc.
The Bad of AI is obviously the misuse or even incorrect use of AI and also the generalization and sensationalization of that. Sometimes I feel that practically every social ill that one can think of is being projected on to AI. This ranges from bias issues to diversity issues and so on. No doubt these are very serious problems that society has to tackle and solve, but I sense a tendency out there to make a convenient scapegoat out of AI. This would be a big mistake.
The Ugly would be the extreme views on AI that pits us, humans, against AI. Despite all the advances in AI, we are far from the time when we have to spend serious cycles worrying about this. It’s healthy that people are thinking that far ahead but it will be harmful if this creates a general perception that makes an enemy out of AI. It is particularly problematic when industry leaders, who are otherwise helping humanity, make great progress indulge in this fear-mongering!
What is your opinion on “Weaponization of AI”? How do you deal with the challenge here?
I am not fully convinced the ‘weaponization of AI’ is that special to warrant a separate discussion. Any technology can be weaponized. Why are we especially concerned about the weaponization of AI – is it because of the ‘intelligence’ aspect of it? Is it because of the fear that it could become ‘autonomous’, free from human control and supervision? The flight control software in airplanes can cause serious problems if it’s not designed well. AI is no different. I do think industry consensus and government oversight will be required for certain applications of AI.
For example, its use in autonomous transport systems. There must be safety standards and rigorous testing and explainability of AI systems. The biggest challenge of today’s AI is that it’s not fully explainable. If we solve that problem first, then ensuring that AI is used ethically and in a safe manner would be achievable.
What AI start-ups and labs are you keenly following?
In today’s environment of democratized AI, where a small set of people can work with open data sets, an open-source AI frameworks, and Cloud GPU/FPGA virtual machines to produce impactful offerings, there are a large number of exciting startups. It may be a bit too early to call out anyone in particular. Interestingly, given that data, more specific training data, is the source of power in AI, I foresee a concentration of competitive advantage in the already established Internet companies compared to startups. These companies are awash with incredibly valuable data. Likewise, even established large companies in other industries such as retail and banking have a huge advantage in their own domains. But many times, I find that these companies do not have full self-realization of the data-power they have.
As mentioned before, I keenly follow the work of Stanford researchers in AI. The Stanford AI Lab practically started the field of AI, when Prof John McCarthy founded it in 1963! We have sponsored research in Stanford Vision and Learning Lab under the guidance of Prof Juan Carlos Niebles in activity detection. I find the newly formed Stanford Human-centered AI institute led by Prof. John Etchemendy and Prof. Fei Fei Li to be very interesting as well. In addition, I follow the work of Prof Yadati Narahari and his team in the Indian Institute of Science Bangalore, particularly in the area of semantic search. I also follow the reinforcement learning work of Prof Balaraman Ravindran and his team in the Robert Bosch Center for Data Science and AI in Indian Institute of Technology Madras.
What technologies within AI and computing are you interested in?
Many things in this category, but I am most intrigued by the promise of reinforcement learning and more specifically in its potential to uncover actions and pathways that even a human may not normally think of. The team of Mindtree Faculty Scholar Prof. Silvio Savarese in Stanford is working on developing the intelligence of social robots – robots that can cohabit with humans. These robots may be around in settings such as a party. When the robot has to go from one end of the room to another when there are humans around, it has to follow rules of etiquette so that it doesn’t walk between two people having a conversation, for example. The team is trying out reinforcement learning techniques and I wonder if the robot will automatically discover a way to do what it wants to do so gracefully, one might be tempted to say, ‘that is awesome – wouldn’t it be great if humans acted like that’!
I am also very interested in new insights about natural phenomena uncovered by even traditional supervised learning techniques. For example, I recently heard about scientists discovering features of seismic waves that precede aftershocks of earthquakes by delving into the layers of a neural network being trained to predict aftershocks. The features that were thus uncovered were apparently completely new to established science. That’s quite amazing!
What’s your smartest work-related shortcut or productivity hack?
This would be choosing the right tool for the AI job on hand depending on what needs to be done and by whom. If I have a deeply technical team available to do a very specialized application that is not readily available anywhere, then I would prefer using one of the AI frameworks such as Tensorflow and hardware acceleration such as NVIDIA GPUs on one of the Clouds. If the team is not deeply specialized in AI techniques, but are good at the application level, my preference would be to use one of the many cognitive API services available on the Cloud such as the Microsoft Azure Cognitive Services. If the team is from the business with no engineering background, then our focus would be to create easy to use interfaces to help them train the system, for example, by tagging pictures of defective parts in a manufacturing line.
Tag the one person in the industry whose answers to these questions you would love to read.
I would tag Prof. Andrew Ng. I learned modern AI technologies from him. He has seen every development in the field of AI from close quarters and strongly believed in the power of deep learning even when many of the rest of us went away during the ‘AI Winter’. He has helped a number of companies become AI companies, Google and Baidu to mention the best known among them.
Thank you, Dr. Satya ! That was fun and hope to see you back on AiThority soon.
Dr. Satya Ramaswamy heads Mindtree’s Enterprise Reimagination Business, an entrepreneurial business unit in Mindtree, that is on a mission to reimagine businesses across industries by leveraging the next generation of Digital technologies. In a career spanning 25 years, Dr. Satya has rich leadership experience in engineering, product management, corporate strategy, and general management.
Prior to his current role, Dr. Satya was Senior Vice President and Global Head of Digital business in TCS, where he entrepreneurially built the Digital business from the ground up to a fully scaled global organization over a seven-year period and helped TCS win several recognitions.
Mindtree is a global IT consulting and services company which helps clients across 17 countries achieve business agility, competitive edge, and growth. We harness the power of Continuous Delivery, our digital expertise, industry knowledge, and research in emerging technologies to drive efficiencies and enable business innovation for over 340 clients. Mindtree is consistently regarded as one of the best places to work. This is a reflection of our entrepreneurial, collaborative and dedicated “Mindtree Minds” who embody the winning culture that defines our commitment to excellence, innovation, and co-creation.