Igor Labutov designs artificial intelligence programs for smart phones and consumer devices.
What is machine learning?
Machine learning is a way that computer software programs understand patterns — like detecting a cat in a picture, without being specifically programmed to identify it.
So how does that apply to a personal assistant?
Right now, it’s more transactional: “Alexa, do this or that.” The key component for the next level is the ability for them to be taught new knowledge and skills by the end user, us. Like a lot of people, I forget stuff. I don’t think through details. I want the agent — just like when I chat with friends — to carry a conversation…to have this ability to ask questions and be proactive.
What made you pursue a career in artificial intelligence?
I was born in Russia, and I would build robots at home. My family moved to the U.S. when I was 11. In middle school I started getting into programming: I made a program for my teacher to keep track of grades. In college, I spent a lot of time in robotics competitions. The most interesting part wasn’t the mechanical part, but understanding the robot’s vision — the planning and the learning part.
How has machine learning changed since you started in 2010?
The field was more based on mathematical computations. But then researchers start making more use of neural networks, which are programs that mimic the way the brain synapses work. It’s a lot more exciting now, with a lot of crazy models that are more experimental. Now it’s more like being a scientist.
In middle school I started getting into programming: I made a program for my teacher to keep track of grades.
What will machine learning look like in five years?
Right now, I’m working on how we go from Siri, or Google Home, to more capable artificial intelligence, the way you imagine futuristic robots from 2001: A Space Odyssey. Machine learning has been solving smaller problems such as understanding questions, or how to generate questions. Now there is more and more work toward integrating intelligent systems that have memory, speech, and vision.
Take us through a typical day at work.
There’s a lot of thinking and modeling. I’m often not in an office. Maybe I’ll take my laptop and go to a coffee shop. I’ll sketch some models on paper, then write computer code to simulate an idea. Then try it on some real data.
What do you do when you’re not working?
That’s all I do (laughing).
I am taking off to New York City to start a startup. We are trying to create a conversational agent for law firms. We have seed funding for about a year and a half.