Killer robots are, if not hundreds, then tens of years away So that idea of computers taking over and all, that is not true. Huma Abidi is categorical when she dismisses the hype and mumbo-jumbo that abounds the general perception of her domain. But the Director of Machine Learning and Deep Learning Software Engineering at Intel, one of the world s largest tech firms, is certain that by 2020 artificial intelligence will generate $13 trillion across the world.
Abidi, who is originally from Lucknow, joined the chipmaker as a software engineer and is now responsible for deep learning framework software optimisation for Intel s top of the line Xeon processors. I was actually studying to be a doctor. But that was a long time ago and somehow it just ended there, says Abidi who followed up her BS in pre-med and chemistry with an MS in computer science from the University of Massachusetts.
Being responsible for optimising deep learning frameworks and workloads, Abidi now helps companies like Google, Facebook and Baidu iron out all areas where deep learning runs on Intel hardware. My team works daily with Google. The optimisations we do get upstreamed to Google Tensorflow. Anybody downloading Tensorflow gets all the work my team has done. On open source machine learning library, Tensorflow powers a lot of Google s machine learning products like image recognition.
Abidi s portfolio is broad, straddling both the hardware and software sides of what Intel is doing in top-end computing. On the hardware side, there are data centres and accelerators as well something like a compute stick. On the software side, which is my focus, we did so much work that it became at par with the competition, explains Abidi, whose team came in for recognition for their work improving deep learning capabilities by 278 times.
For someone who has been in the IT industry for well over two decades, the work on AI has been an eyeopener. What I liked the best about AI is how people are building over each other s breakthroughs. The models have been put on GitHub, papers are being shared, archive downloaded I have not seen this before. Unbelievable, she says, adding how everybody is coming together to standardise as much as possible. Abidi says her team s work on Intel One API a set of developer tools that provide a unified programming model is an attempt to offer a solution so that everyone can collaborate more across different architectures.
AI is nothing new, says Abidi, but now the coming together of data availability, cheap hardware and powerful algorithms have made this spurt possible. So it is moving in a very good direction and that AI winter and all is gone. This is phenomenal and I m really happy to be part of it. But if someone is hoping it to solve all those problems, then it is early.
But there are many projects that have given Abidi immense satisfaction. Like the work her team did to help Novartis process an image about 26 times larger than what the were used to in a matter of minutes and not days like before. Then there is the work on handheld devices that can detect diabetic retinopathy, as well as some image detection work that helped reduce poaching in Tanzania.
For companies looking to add an artificial intelligence layer to their work, Abidi has a simple message: Don t do AI for the sake of AI, but do it so that it helps your business. So bake it into your underlying core and see how you can use it. And don t reinvent the wheel. She cites the example of Mujeeb Kolasseri from Palakkad in Kerala, a high school dropout who set up a business of tagging data from camera for clients all over the world to help machine learning. That s what I said about the kind of jobs that can be created.
We have to be careful not to discard AI as a bad thing, Abidi warns, underlining the ability of this new technology to uplift mankind in all areas.