Interview with Professor Zhang Zheng

Can you tell me a bit about yourself and your background?

I’m a Shanghainese native. I graduated from Fudan University in 1987, and entered the Master’s program after that. After 1989, I went abroad and studied in Texas for a while and got my PHD at the University of Illinois. I worked in HP labs for five years before I moved back to Beijing. I have been a researcher at Microsoft Research Asia for 12 years, which is longer than the time I’ve been abroad.

Why have you chosen your field of study?

I like to build things. I’m an engineer by nature, although I am quite clumsy. I’m curious about science and its mysteries, for example how intelligence is possible. These puzzles have always intrigued me. When I was very young, I used to build radios. Perhaps people your age don’t have any such experience, and have probably never even seen radios with transistors. I tinkered with them when I was very young at Children’s Palace in Shanghai. I eventually moved onto computers. I remember when I was at Fudan University, my department had 500 students, which is more than the total number of students currently at NYU Shanghai. But there were only two PCs for all of us – two very old PCs. That was around the end of my undergraduate years in 1986, a very long time ago.

What are your recent research interests?

Recently, I have been interesting in “deep learning”, which has a computational format that is very close to the human brain. The brain has neurons, and this computational framework also has neurons. It’s good at recognizing objects and images, and understanding languages. It’s a giant step towards understanding how human intelligence operates. Human brains have this amazing ability to recognize objects instantly, like “hey, this is a phone” or “this is a pen”. Computers traditionally are pretty weak on these tasks. They’re very good at counting numbers and are much better at solving well-formulated mathematical problems than humans. Human beings are very bad at precise memory and reasoning, but we are good at fuzzy learning.

And conceptual learning?

Yes. We do not know how the brain represents concepts. You and I can talk, and we can understand each other instantly with the help of facial expressions and gestures. This is something very difficult for computers to do today, and this is also one of the things that motivates me, because it is a puzzle. Traditionally, these are two different disciplines. Problems I mentioned would be researched under machine learning, while my discipline has always been building systems. But my passion goes beyond system discipline alone. I used to say that I build systems for science, and I want to find the science of building systems.

What do you think are some possible societal impacts that may result from the advent of machines that can learn or perhaps replace people?

This semester I’m teaching Introduction to Computer Science, in which this topic is touched a number of times. Let me first introduce a bit about the course. Traditionally, this is a programming course, but I didn’t design it that way. It now has several parallel tracks, including programming and understanding algorithms. But I also want to show my students the broader landscape of computer science and how it came to be. In preparing this course, I learned a lot myself. We will follow the progression of computer science from a seed to an industry.

We will also cover the epochs in the evolution of computers where people have been worried about massive job loss from computers. There has always been this debate. From the very beginning, computers have been replacing people at menial tasks. That is a very real fear, but looking back at those menial tasks, no one wants to return to the kind of job where you sit there and count things all day long. 

This has always been a fundamental debate, because the flip side is that people are being freed from mechanical tasks to do other things. The trend has continued, but the question is when does it stop? Is there a limit? At the end of my course, I’m going to do something different. We start with history, so we’ll end with a puzzle. A forward-looking question – what is the future going to look like?

We are also going to discuss questions of privacy, once privacy moves beyond revealing personal information online to the revelation of your entire body and soul, when the problem of privacy gets taken to the extreme. We will examine the human relationship to the machine, which is a very deep philosophical question rather than a technical one. One of the things I really want to the students to take away is, “you are the builders of the future. More than making programs and fortunes, you have a lot of responsibility with the tools you have at hand.” I don’t know what that responsibility is going to be, but I want them to start thinking about it. 

What are some of the ways in which computer science is interacting with the humanities?

I have recently been working on the study of bias, and how it influences human recognition of objects. Computers can recognize digits on par with or even better than humans can. However, if you overlay two digits on top of each other, as in, if you write a 2 on top of an 8, the computer can no longer recognize either digit, but humans can. Why is that? It puzzles me. It’s a very simple scientific question. When we see a 2 on top of an 8, we begin to form a hypothesis about the digits. What does this mean? It means when you have an object embedded in a noisy environment, the human mind automatically picks out what parts it wants to see. The philosophical connection to that is “perception is reality” – a common statement that perfectly illustrates what is happening here.

There is no such thing as a noise-free environment in our lives; everything exists in the context of everything else, therefore the human mind must be able to “de-noise’ its environment, to form a belief that allows it to see part of the picture and ignore the rest. What we have done in this research project was demonstrate neuro-network circuitry that does precisely that in a computer. We were able to tell the computer to assume a hypothesis and remove the noise of the other digit. We demonstrated how this forming of bias can be done through neuro-network circuitry.

Going back to humans. We have biases to guide our daily life; these biases are important to   survival. When you have a high-noise environment, it means there’s no such thing as absolute truth. Data is embedded in noise. What we see and perceive may be biases. That is why we form such deep impressions of people in our daily life. You might have a friend who you like, and the more you like him, the more you form the belief that he is warm-hearted and kind. But you are only seeing one part of him. On the flip side, there are also people you don’t like, of whom you form an instant dislike that may not be based on anything objective. 

This is something most people know by heart, and this is psychology, but psychology will remain a pseudoscience unless we are able to somehow quantify and predict effects. This is just a small instance of how research in computer science is beginning to connect back to high-level humanities subjects like philosophy and psychology.

What are some of the greatest challenges facing the field of computer science?

One of my research interests is adapting technology to the diagnosis of cancer. There are some challenges in doing this, even though it is within the range of computer science. If we have sufficient data, we might be able to diagnose cancers much more effectively, but all of this hinges on whether you will be able to get the data. This is private patient data, but unless we get this data in massive amounts, we can’t create any models or do anything interesting. That’s one of the biggest challenges facing the field – access to data.

Advancing technology is about much more than money, data is essential. We have to find a way to access data to benefit technology and society as a whole, but maintain the privacy of the individual and navigate the pitfalls that can come with the use of private information. This is an eternally difficult challenge.

You have been doing research for a long time, how would you compare the environment now to when you began?

I started doing serious research 20 years ago in Illinois. The one thing that I can say for sure is that things are evolving much faster. A good friend of mine from MIT often says “technological evolution is happening exponentially”. Any discovery we make has its application in the tools, which makes our tools more powerful, which then accelerates discoveries. That’s why things have been happening so much faster recently.

Then, there are the aspects of doing research that will never change. You must always understand the problem and its context. You must always think of ways to innovate, demonstrate, and quantify your research. These things will never change, they just took longer in the past. Mendel was working on genetics for eight years. Darwin took 30 years to form his theory of evolution.

Research happens a lot faster nowadays, but of course there is a lot more noise to distract you. There’s a whole ecosystem through which you have to filter out and find the right piece of research to learn and to inspire you, and that ecosystem is getting more complex all the time.

I would say there are many more distractions for students. A number of the brightest students don’t go to graduate school and end up on Wall Street for instance. 

What would you tell those students who are thinking of going to Wall Street? 

Well, first of all I would tell them that if they are really clever and hardworking, then it’s better to stay in this field because there’s more innovation here, and there’s a much bigger financial upside to it too. And you get to have an impact on society. To paraphrase Steve Jobs, “Do you want to sell sugar water or change the world?” It’s the same thing, “Do you want to count money, or do you want to change the world?” If students decide to go, they go with all my blessings, but I want them to remember that they still have a responsibility because of the tool set they have acquired. Also, they can come back to study at any time.

Is that part of the reason you've come back to China?

I've come back to China for different reasons. I’m native Shanghainese. My roots are here. I need to take care of my family. Fundamentally, this is my country.

I didn't think about becoming a professor when I first came back to China, but in my job I found myself passing on my knowledge to my interns and staff. I've had over 30 students from the best institutions passing through Microsoft in my time there. I wasn't teaching in a classroom, I taught while doing research together. In many ways, when you do research, you’re a teacher and a student. You’re learning the technology and solving the problems together.

Here, teaching will be different. I think this is a really neat context. The new building is progressing really well, but what defines a school is not the building, it's the people. I look forward to talking to the students, and finding out more about them. I want to inspire them and lead them in a style of research that will help the field to evolve. What are the problems and what are the cool ideas that might solve these problems? What new problems do these ideas generate? I’m going to try to use a different style of teaching. I could have chosen an easier way to teach the course, but I’ve decided to go the harder way. My hope is that it will be more rewarding.

When your students leave your class, what are you are hoping they will take away?

Oh, I want them to think that computer science is cool.

In my lectures, I want to show them the interesting disciplines and interactions that have been taking place, the whole wide range of exciting and interesting developments. I want them to see how cool computer science is, and find an element that they would like to devote their careers to. Also, very importantly, I want to show them that computer science doesn’t stand alone, that it has a larger context within society.

If they end up understanding these things and choosing computer science, I will be happy, but if they don’t, I hope that they will leave my class free of the bias that I was talking about. I want them to know that computer science is not just for geeks. It’s much more than that. That’s the kind of bias I want to remove. I also want to see them develop a sense of passion and responsibility.