‘Machine learning and AI a huge opportunity for India’
SRIRAM RAJAMANI, managing director at Microsoft Research India, who heads a 50-person computer science lab in Bengaluru, says he never believed machines could one day become intelligent. Now that they have, it has kicked off a massive restructuring in every sector. This country, he tells Alnoor Peermohamed & Raghu Krishnan, is placed well to benefit from this. Edited excerpts:
What does MSR India do? Which technologies did you develop that have been used in Microsoft products?
Microsoft Research worldwide is probably one of the biggest computer science and research lab networks; we do research in every area of computer science. The lab in Bengaluru focuses on four major areas — theory and algorithms, machine learning and artificial intelligence (ML & AI), systems, and technologies for emerging markets.
Every product or technology that Microsoft ships today has contributions from Microsoft Research. One of the more visible examples would be Kinect for the Xbox console. Until it came out, to play a video game, you needed controller in your hand. The Xbox began using computer vision, where you play with your hands and the computer sees you. This happened because various people at Microsoft Research thought you can have cameras to sense how a person moves.
Another would be Skype Translator, which can allow any two people on the planet to talk to each other in any language. All these things have been worked on by people in our lab, the China lab, the Redmond lab — they stitch it all together.
How many research papers does MSR India publish each year?
For a 50-person lab, we produce easily more than 100 high quality research papers a year and attend all the top conferences in the world. Every year, we win five or six best paper awards and the community awards for researchers who’ve done work over a long period of time, 10-20 years. This is how we value our people.
What is this ML and AI revolution all about?
When I graduated in 1991, I thought a machine is this dumb thing in that a programmer writes a programme and the machine only executes it. But, it has turned out that there are certain tasks that have been very difficult for even a programmer to write programmes on. For example, if I give you a picture and ask you to identify if there’s a dog in the picture, that’s something a human can do very effortlessly. But, if you ask a programmer to write a programme that identifies if there’s a dog in the photo, it’s almost impossible to come up with the parameters.
What they started doing, instead, was training of machines. We built computers in which you don’t describe what a dog is but just show it a picture of a dog and tell it, this is a dog. You show another picture and tell it this one isn’t a dog or is a cat. You show hundreds and hundreds of pictures like this to a computer and it learns the concept of the dog. For this to happen, we needed two things. We needed lots of photos labelled as cats and dogs, which the Internet gave us. People on Facebook have uploaded millions of images of their cats and dogs, and tagged these accordingly. The second thing we needed was very powerful machines to crunch all of that, which the cloud has given us. The machine learning algorithms have been around for a long time but this is suddenly why computers can recognise cats and dogs.
A recent survey said there was a requirement for nearly 4,000 ML & AI programmers in Bengaluru alone. What are these jobs for?
There are very few people in the world who can come up with ML algorithms, the people who come up with the mathematics behind everything. These things then get into packages, used for lots of applications. It’s like giving you a hammer — you need a special kind of person to build a hammer but once you build it, you can give it to many people to use it to build a house.