“Everyone doesn’t need to be an AI engineer, but everyone can try to figure how AI can solve problems in their areas of interest.”
A technologist who loves to try new things, Saurabh Tiwari has spent the last 17 years doing some amazing work at Intel India. From working on chipsets for mobile platforms and the earliest Intel® Xeon® processors to computer vision and other emerging technologies, he is now the principal engineer and technical manager for Intel’s Data Centre & AI Group. He and his team develop performance simulation and projection tools for future generations of AI training products.
We recently spoke with the AI enthusiast about his journey into the field and how emerging technologies can enrich the lives of every person on earth. Below is an excerpt from our conversation.
Tell us about your first brush with AI?
It was in 2013 when I got the opportunity to lead a small team of three engineers, which gradually transitioned into a whole new group within Intel. We were working on a new product range using AI to enable computer vision that could identify objects, people, and signboards. These were very early stages of AI, and it all seemed pretty much like sci-fi. We built a few iterations, made mistakes, rectified them, and eventually that work led to what we now know as NNP-I 1000, one of the first dedicated AI processors to make it to the market. I’m very proud to see it work exactly how we once envisioned it.
How would you describe AI to a non-technical person?
Well, AI is like mimicking the human brain. Our brains consist of neurons that sense signals. When we see or hear something, those signals are sent to the brain—or a network of neurons—that then process the information. Today’s modern form of AI, which we call deep learning, uses a similar concept. Each neuron makes sense of a bit of data and, together, the network makes unified sense of the data. For example, a neuron may sense a few pixels, and then the next set of neurons “sees” patterns and lines. As you go deeper and deeper, it can finally see the complete image. We model these neurons using mathematical formulations.
While traditional computers are also known to do math calculations, they are only known to do repetitive tasks efficiently. If I write a program, the computer will do exactly what I write—it cannot deal with unknowns, adapt, or learn. For AI, we design algorithms that can modulate themselves and become better with time and more data, deriving intuition or intelligence out of the data. Today, AI is already better than humans at image recognition and language translation.
In the technological world, what is the general perception of AI?
When I teach at Intel’s Data Centric Skills Academy, I often come across two types of people. The first set, the non-believers, think AI is limited to image recognition and cannot really solve any real-world problems. The other set is looking to become AI engineers instead of figuring out how to apply it in their fields of interest.
To begin with, everyone doesn’t need to be an AI engineer, but everyone can figure out how AI can solve problems in their areas of interest. AI applications can streamline processes across domains, be it medical sciences or civil engineering and beyond. Intel’s employee training program triggers a cycle where engineers attend these sessions and are then willing to incorporate AI in to their current projects or real-world situations. It is this approach of applying AI in existing situations that has led many former non-believers to trust the capabilities of AI.
What real-life problems have you and your team solved using AI?
As discussed previously, we built the first data-centric AI processors, which is now used by the world’s leading social media site. This social channel connects millions of people, families, and friends across the world and offers a platform for small businesses, like-minded communities, and more. I’m very proud to see how AI can influence and impact our day-to-day lives. Today, it is successfully deployed in the cloud and our work on it earned my team the prestigious Intel Achievement Award. We are also building AI training accelerators for another leading cloud service provider.
What’s exciting about your job at Intel?
If I must pick one thing, then it would be the risk-taking. At Intel, I am allowed to fail, to improvise, to make mistakes, to learn from those and try out new things. This flexibility and open-mindedness are crucial when working on advanced and emerging technologies. These technologies are fast evolving, so it is important to be agile, to adapt quickly, and be willing to backtrack and start over from scratch.
You hold a patent for what was an innovative weekend project. Tell us more.
Yes, I have been accredited with a patent in architecture for in-memory information embedding. While on a business trip to the U.S., I was working on a server product. During our discussions with architects and engineers, we identified a very specific problem with compressing the memory. Over the weekend, I couldn’t get my mind off the problem and decided to write a program. After some brainstorming and refinement, the program could solve the problem and that resulted in the patent.
What is the highlight of your 17-year-long career (and counting)?
Intel provided a fulfilling career path that has let me work on many interesting products. As part of my second project at Intel, I had to relocate to Israel for six months. It was a very interesting phase for me, both professionally and personally. That multicultural experience taught me to ask as many questions as possible and how to have healthy arguments wherein you do not carry the arguments out of the conference room—all in the name of creating the highest quality product possible.
Looking to impact lives with next-gen technologies like AI? Discover opportunities at Intel India today.