In this column, Daphne Koller asks Carlos Guestrin, Amazon professor of machine learning at the University of Washington, to share his thoughts on the growing value of data science. Courses and lectures from Carlos appear in the University of Washington's Machine Learning Specialization on Coursera.
Just 20 years ago, if you told your parents you were majoring in machine learning, they'd have suggested you'd been watching too much science fiction. Fast forward to today, and the ability to analyze and make predictions from big data is arguably the highest-demand skill one can list on a resume.
Nobody has a better perspective on this exploding field than Carlos Guestrin, Amazon professor of machine learning at the University of Washington and CEO of predictive app development platform Dato. Called "the information wrangler" by Popular Science (which named him one of its "Brilliant 10" in 2008), he's dedicated his career to making machine learning accessible to students, entrepreneurs and innovators everywhere.
We recently chatted about how his field became a magnet for the best and the brightest students around the world--and why tomorrow's most successful startups will have machine learning at their core.
When you and I were at Stanford in the late '90s, machine learning was pretty esoteric. What attracted you to it back then?
Yes, when we started, people weren't quite sure what machine learning was or what it could do for them. Honestly, it was mostly because I read a lot of Asimov and was really interested in building smart robots (laughs).
I had to laugh when Harvard Business Review called data scientist the "sexiest job of the 21st century." What's your answer when people ask why machine learning is so "hot" right now?
I remind them what the Amazon website looked like 20 years ago: it was just this grey page with text. And today, it has completely changed the way we shop and how merchants sell their products, upending the way we shop. Similar things happened with Google and Web search, Zillow with real estate, Netflix with video content and Uber with taxis. At the core, machine learning is what makes all of these applications special. It's the differentiator.
It's been a thrilling ride, and I think we're still in early days. Do you agree?
Definitely. These days, machine learning tasks are predicting things I didn't even expect to be possible on really huge data sets. The scale and magnitude of the problems has changed, and the complexity of the tasks has changed. And honestly, the accuracy of the underlying machine learning models has changed. Five years from now, I would say that every successful application out there is going to use machine learning at its core. That's an exciting thing--to go from being something theoretical to being something that's driving industry!
What gets you most excited about the future of machine learning?
I'm interested in ways it can improve our political process. Imagine an automated version of FactCheck.org where anytime a politician says anything, you're able to look it up and understand what's not quite right. The potential to personalize medicine is also extremely exciting. Today, if you and I both have the same condition, we'll get the same treatment despite the fact that we have different lifestyles, gender, DNA and ethnicities. That's barbaric! It's still nascent, but I want to think the field will eventually be looking at each individual and what's best for them.
Your online course introduces learners across the world to the fundamentals of machine learning. How do you convey that sense of excitement to a wide range of people?
My co-professor and I have been thinking about that for a long time. Typical machine learning courses start with probability rather than actual application. Our course is the opposite: We start with use cases, like how Zillow predicts house prices from data and how Pandora figures out what song you should listen to, and use those to explain the underlying concepts and algorithms. When you start with machine learning's impact [on society], it's hard not to be inspired.
What advice do you most often find yourself giving young people?
We've gotten to the point that it's possible to incorporate machine learning in ways that we never imagined. There are intelligent people who don't have that deep math background still having lots of transformational impact. And so what I say is to really unleash your creativity. Really think about what you could build and what's different and what's unique. The machine learning techniques are here to follow you.
And what about to entrepreneurs hoping to make a major impact?
I have a startup myself, and we talk a lot about how we're all really here to fulfill other people's needs. One thing that's changed a lot in the last 20 years--in the computing industry especially--is how the user comes first. Now, when I pull out my phone and it doesn't do what I hoped for, I think, "This thing doesn't love me." We've gotten to the point where the intelligence is expected.
The next "unicorns," I believe, are going to be startups that really, deeply understand their users and provide them with individual value. It's about adapting to what the marketer might call a "segment of one." And I believe you can only do that with machine learning.