Canadian Information Processing Society (CIPS)



Andrew Ng, Co-Founder, Coursera; Director, Stanford Artificial Intelligence Lab; World-renowned top-ranking distinguished researcher, innovator and entrepreneur

This week, Stephen Ibaraki has an exclusive interview with Andrew Ng.

Andrew NgAndrew Ng, Co-Founder, Coursera; Director, Stanford Artificial Intelligence Lab

Quoting extensively from his Stanford profile, ACM and Wikipedia, Andrew Ng is a co-founder of Coursera, the Director of the Stanford Artificial Intelligence Lab and a Computer Science faculty member where he is a distinguished researcher in artificial intelligence, machine learning, and deep learning with over 100 publishing credits.

In 2011 he led the development of Stanford University's main MOOC (Massive Open Online Courses) platform, and also taught an online Machine Learning class that was offered to over 100,000 students, leading to the founding of Coursera with his partners. Their goal is to give everyone in the world access to a high quality education for free. Today their platform partners with top universities to offer high-quality, free, online courses. With over 100 partners, over 500 courses, and 7 million students, theirs is the largest MOOC platform in the world.

Ng's recent awards include being named to the Time 100 list of the most influential people in the world; to the CNN 10: Thinkers list; Fortune 40 under 40; and being named by Business Insider as one of the top 10 professors across Stanford University. In 2008, he was named to the MIT Technology Review TR35 as one of the top 35 innovators in the world under the age of 35. In 2007, Ng was awarded a Sloan Fellowship. For his work in Artificial Intelligence, he is also a recipient of the Computers and Thought Award.

Outside of online education, Ng's research work is in machine learning. Ng’s Stanford research group focuses on deep learning, which builds very large neural networks to learn from labeled and unlabeled data. Recently, a Stanford team (led by Adam Coates) built the world’s largest deep learning system with over 10 billion learnable parameters trained via back propagation using inexpensive GPU hardware. This work was presented in ICML 2013. In 2011, Ng founded the Google Brain project which involved a neural network trained using deep learning algorithms that learned to recognize higher-level concepts, and is currently used in the Android Operating System's speech recognition system. His early work includes the Stanford Autonomous Helicopter project, the STAIR (STanford Artificial Intelligence Robot) project, and ROS, a widely used open-source robotics software platform.

More information can be found here.

To listen to the interview, click on this MP3 file link

The latest blog on the interview can be found in the IT Managers Connection (IMC) forum where you can provide your comments in an interactive dialogue.


Interview Time Index (MM:SS) and Topic

:00:34: In terms of your research what are the big research questions in machine learning and artificial intelligence and what are the ways and processes for answers?
"....I think the scene of AI and deep learning specifically (which I've been working in), has been taking off like crazy over the last few years. The number of academic research papers on deep learning and on the massive neural networks has been rising faster than almost anything on the scene and in industry too....Being at the center of some of this activity I feel like many causes of deep learning are still wide open and we don't really know where the future is going to go...."

:02:22: I had a discussion with Tom Mitchell and he talked about the Never-ending Language Learner (NELL) Project, and I guess Carnegie Mellon also has their NEIL Project and the ramifications of that. Do you think it's going to go in the direction where we are going to have really useful, almost human-like capabilities that are on the Web?
"....I think Tom Mitchell's work is really inspiring because his vision about continuous learning software is inspiring....There are some things that computers vastly outperform humans just in terms of the sheer amount of text they can consume, but in terms of reasoning about the world and real human level AI (despite getting far more data, text or images or audio than any human could possibly get in their entire lifetime), our software still falls short. So to me this is a sign that we still don't have the right algorithms yet because whatever humans do we learn to be much more intelligent with much less data...."

:04:17: What are some controversies in your field and why?
"....Deep learning is very exciting and one of the confusions in the discipline is that the term 'deep learning' encompasses really two ideas. The first idea is called Supervised Learning in which if you have a lot of labeled data, these algorithms are fantastic at soaking up the labels to make accurate predictions....But there's a second, not really unrelated body of ideas that also goes by the term deep learning that is very different, which is: 'can you get a piece of software to watch YouTube or read text on the internet or listen to audio for hours on end and without you telling it anything or tagging or labeling any data and have it figure it out for itself?'....I think the second unsupervised learning, learning from unlabeled or untagged data is maybe most human-like. I think most humans learn primarily from unlabeled data and I think that this unsupervised learning idea has tremendous potential for letting us make a lot of progress in machine perception...."

:07:36: Do you see some value being applied to the deep learning work?
"....The world is so complex my instinct is to try to select incredibly flexible learning algorithms and let the learning algorithms examine data from the world and to sort out what is true or what is not true about the world for itself, so my instinct is to steer away from that body of work. But it would be interesting to see how the more traditional, the logical reasoning aspects of deep learning algorithms could come together some day...."

:09:00: In the work that you do what are the practical applications for 2015 or 2016 and how do they impact things like business/government/media/education and society?
"....I think because of the work of the major tech companies using deep learning, many of us are already using deep learning algorithms....I think machine learning touches so many aspects of our lives, and I think what some of these organizations have done is create deep learning through the power engineered throughout their organizations in order to apply the deep learning algorithm to many different problems. I like to tell people that most of us use machine learning algorithms dozens of times a day without knowing it...."

:11:10: In your opinion, what will computers and robots look like by 2020?
"....It's only in the last two or three years that there have been far more robotics startups than in the previous 3 or 5 or 10 years, but I really don't know where it's going to go. It's really exciting work and people are producing low cost robots for manipulation which is fantastic for researchers....I think if we want to make progress towards truly intelligent robots we have to be careful not to overhype the science. In order to make a little bit of progress towards AI or towards truly intelligent perception, I think there is tremendous potential in deep learning algorithms, especially the uncertified versions of learning algorithms...."

:14:52: There are varying opinions of Kurzweil and his idea of singularity. Do you support it in any way or do you think it is too far out?
"....AI has tremendous potential that I think in the coming five or ten years we'll make tremendous progress in perception. I hope that deep learning will play a huge part of that but the discussion of the singularity I don't think is serving the science well, and I think it is far further out than the impression some popular media sometimes convey...."

:16:11: You are a founder of Coursera (with Daphne Koller). Can you overview your projections for 2015, the key KPIs of success?
"....A couple things we care a lot about, one is growth....We want to give everyone access to great education so there is a lot of room to grow this season in our work on mobile apps to let learners access Coursera....A second one is credential value. The thing is that mobile learners around the world are getting credentials that they are earning from taking Coursera MOOCs and this is on their resume and are using them to find better employment opportunities....Another thing we are tracking is the degree to which employers understand the value of these MOOC credentials and we're seeing that grow rapidly already and I hope that continues to grow throughout the next couple of years...."

:19:09: The concept of MOOCs is so controversial. There are people who rally behind it and say it's the greatest thing that ever happened and others that say it is an old story. What are your feelings about that controversy or do you think it's just a transition point?
"....The ability to deliver these highly scalable forms of education I think dramatically changes the economics of education. There is still a large upfront cost to producing the course content, then once you've done it the incremental cost of signing up new students is effectively zero so that's a change. Having said that I think the student experience isn't good enough. We haven't figured out the pedagogy, the software platform, the website needs a lot of work. I think it's still very early days...."

:21:40: Is there a value to other market segments such as business, government or society?
"....In the broadest sense I think an education gives you super powers. With an education you can learn to write software, teach other people, learn to cook healthier food for your children, with an education you will live a longer life....At a societal level I think that we can accelerate the progress of civilization. Governments are looking to MOOCs as a way to level up their populations' skill set....The world today changes so fast that all of us need regular infusions of knowledge in order to stay current. Even though we work with universities and this is a project that the five of us had launched out of university, the biggest impact MOOCs is having today is not on college students, it is on continuing adult education. I think many businesses, either individuals, the working professionals or often management are coming or sending others to come to Coursera to take MOOCs in order to continue to develop their employees...."

:25:17: What are the latest research findings on online education (MOOCs)?
"....By now we've had lots of things, dozens and dozens of studies done by Coursera or by our university partners. I will share one result that was surprising to me....this is just one example of the many dozens of studies that we and Coursera's university partners have done that I think are starting to allow us to more deeply understand student motivations and student learning...."

:28:10: Can you add additional insight to some of your biggest challenges and controversies in this field?
"....I sometimes get asked, 'Will this replace professors? Will MOOCs replace professors?'....I think the opportunity is for technology to free up that favorite teacher of yours from the more routine, repetitive aspects of teaching so that they can spend even more time in the future in conversations, mentoring and coaching students, just as they did with you and just as my mentors did with me...."

:30:06: Can you describe your most significant and influential research achievement and the practical outcomes seen today and forecast it into the future?
".....One of the things I'm quite proud of was the fact that we launched these MOOCs....I'm also proud of the deep learning work that my students and my Google team have done....It was really team work. I feel that because of my role I tend to get outside credit for a lot of the things that were really the work of my students or the work of my teammates....And one last thing. You asked about robots just now. One thing I didn't mention is that my Stanford students and I have been spending a long time looking at building self-driving cars using deep learning algorithms. I think that could be another economically important application...."

:34:05: Can you generalize a little about what you see as the most difficult challenges in research in general and then some lessons that you can pass on to the audience?
"....One of the books that has influenced my thinking a lot is a book called the Lean Startup by Eric Ries, and this is a philosophy to building startup companies, but it's a philosophy that I think applies well to research projects as well because I think there is a certain emphasis not on laying out grandiose multi-year plans that have huge assumptions and huge risks and may ever come to fruition, but an emphasis on staying humble and running experiments like crazy and learning and iterating....I think we should have a grand plan, we should have some idea of where we are going....But when we think about what happens day to day over those five or ten years I think there are little pieces of day to day learning. The faster you can learn the faster you can make progress in research because I think fundamentally research is learning about the unknown...."

:41:53: Andrew, what are the greater burning challenges and research problems for today’s youth to solve to inspire them to go into computing?
"....I think that computer science is fundamentally about scale. I think that in many professions that individuals could choose to go into, there are few disciplines as scalable as engineering or as computer science specifically....Computer science is impacting so many ideas. I hope more people will join me to work on AI, but even outside of AI I think the ability of computer science to change the world is almost unlimited...."

:45:31: There are a number of organizations, one of them is the ACM and it has a number of resources. How has the ACM and its resources supported your research?
"....I always look forward to receiving the ACM newsletter and the new ACM magazine in the mail and flipping through that in order to have another channel to learn about the latest computing trends. I've been grateful to the ACM for the events it organizes for bringing the technical computer together...."

:46:40: In a broader way, including but even looking beyond computing, what do you see as the top challenges facing us and how do you propose they be solved?
"....I think our planet faces a lot of challenges — I think that global poverty and inequality are some of the major challenges facing our society...."

:50:15: From your extensive experiences, speaking, travels and work, can you share any stories (perhaps something amusing, surprising, unexpected or amazing)?
"....A few months ago I was at a party at LinkedIn here in Silicon Valley and I met one of the students who had taken one of my Machine Learning MOOCs. He came up to me and said 'you must be Professor Andrew Ng'. He said one of the features he most liked was the ability to play video at 2x speed because it allowed him to blaze through the video and if he missed something he could just do an instant replay. But he said to me, 'I've listened to about 20 hours of video of you talking and all of these videos are of you talking at 2x speed, but now that I meet you in person, I'm really surprised that in person you'...."

:54:14: Andrew, with your demanding schedule, we are indeed fortunate to have you come in to do this interview. Thank you for sharing your substantial wisdom with our audience.