How AI can augment humankind
Back in 2014, Stephen Hawking warned that people should be careful about artificial intelligence (AI)—the full development of it could spell the end of the human race, he said. But its progression has not been slowed down by such warnings, and today AI has achieved some concrete advancements, such as translation and facial and speech recognition. In addition, combing it with hardware improves the machine’s performance, offering safer driverless cars, more capable drones and robotics of different kinds.
Brad Nelson, professor of robotics and intelligent systems at ETH Zürich, is optimistic about the technology’s development. To him, machines and robotics are augmenting instead of replacing the human workforce. In this interview with CKGSB Knowledge, Nelson talks about the state of AI so far, China’s advantages in this industry and, as an engineer, his insights into the relation between humans and machines, as well as how to find the balance between fundamental research and commercialization.
Q. Can you give us a brief overview of where we are today in terms of AI development? What are the most widespread applications of AI in our daily life?
A. From my perspective, AI has a long history. I first started doing research on AI in 1985—at that time everybody was convinced that AI was going to take over the world. It is a long time ago, over 30 years ago. And I think the excitement around the field was great at that time, and then people became more realistic about it. And even then in the late 1980s, people became interested in artificial neural networks, but people rapidly realized that a lot of claims weren’t living up to the reality. But what has become interesting in AI over the past several years is that there have been some very concrete performance improvements. We’ve seen in computer vision and machine translation that these technologies are improving, improving our state. That’s got a lot of people excited. We all have visions of the future—visions are a dime a dozen—but the question is how are we really going to get from our vision to there. I think it is an exciting time to be in the field, there are certainly a lot of people who have ideas on how to go forward. As an engineer, I fully believe that I’m going to see AI as one part of my toolbox. I will still have my numerical simulations that I am used to using, my design principles, I have my analytical and mathematical representations, but I fully expect that engineers in the future will have to have machine learning techniques, AI techniques as an integral part of the tool box that they are going to bring to solving a problem in the real world.
Q. What do you expect in the future, how well-developed will AI be?
A. We often get over excited about technologies and we get scared of technologies—people like to look further down the road. I am sure it is going to become more prevalent and it is going to be more capable, but I am not at all afraid that it is replacing people. We are just going to have to adjust the way we use these systems and we’re going to have to adjust the way we engineer things, but are we going to have everything planned by computers? I don’t think so. I think the world is much too complex and we do not understand that complexity.
Q. Machines might replace the human workforce. If that happens one day, what are the pros and cons?
A. I work in medical robotics, and I see robotics really augmenting the skills of surgeons—it’s not replacing surgeons, it’s making them even more efficient. We still have a tremendous amount of the world that just simply does not have access to surgery, for instance. There are many, many developing countries where if you really want to raise their standards of living, give them the access to surgery. Now what if we had machines that could be controlled remotely by surgeons to treat corneal diseases, to treat cataracts, to treat heart disease? That’s an area I am excited about—the potential for medical robotics to impact surgery in the developing world is exciting. Historically over the past 100 years or more, we’ve seen machines replace humans in certain ways, but it always opens up other opportunities, and that’s going to happen again. In medical robotics, it will be welcomed in many ways if we can bring this in to augment our medical professionals’ capabilities.
Q. Stephen Hawking warned people have to be careful because AI may threaten human beings one day. Do you think it is a valid concern?
A. I think machines are always a threat to humans, because they are dangerous. Cars get in accidents, airplanes crash, factory floors can be dangerous. We just have to engineer systems to be safe. And we have to understand test methods for that, and that’s what we do as engineers. I am not so worried about them hurting humans. The thing I most worry about is what happens in the financial markets. That’s the thing that scare me the most—every once in a while you see hints of the algorithms maybe causing some instabilities in the market, and we learn from those things. But that’s part of what we’re doing as engineers—to try to understand the limits of that, try to put bounds and proofs on that and understand how we can make sure these are safe systems.
Q. AI is really hot, so venture capital from all over the world, from Silicon Valley, to Beijing and Tel Aviv, are all pouring money into AI startups. Do you think there is bubble or are we just getting started?
A. I do think there are people pouring money into this without really understanding the limits of it. There was just NIPS, a big AI conference in Barcelona, and some of the researches got together and created a sort of artificial intelligence and immediately attracted investors to that, sort of pointing out that maybe people need a little more depth and understanding when they are putting money into this. It’s certainly a hot topic, I guess you could say it’s a bandwagon in a way, but you can’t argue with the performance improvements that we have seen in certain, limited fields. Now whether those extend to other fields, some of them I’m sure will and a lot won’t.
Q. What are the most important factors for a country to achieve advances in AI technology? Do you think China has them?
A. If you’re going to advance in technologies, you’ve got to be willing to take risks, for one thing, and I think China is well known for its risk taking. One of the things that the United States has always been good at is rewarding risk takers and not looking at it as a negative—“fail early”, is a phrase you hear in Silicon Valley a lot and I think that’s a good, healthy attitude. I think sometimes in Europe we don’t think that way, we are a little more conservative. But that is changing—the world is getting smaller, we are all communicating and learning from each other more rapidly. But I think China has a lot of going for it certainly: an openness, risk-taking and the promise of reward if you are successful. Those things drive a field.
Q. So besides the business environment, are there any other factors? Like technology or people?
A. My first trip to China was in 1998. You certainly see universities getting better and better. You see a lot Chinese coming back [from abroad]. I think it is interesting, if you go back to the United States 130 years ago, 90% of the Americans who got Ph.Ds., they went to Germany. They learned from the German system about research and then they came to the United States and transformed that system. You could almost see that kind of trend here—you see a lot of people coming back, you see excitement. Communication is so much quicker now and easier and everything is changing quickly.
Q. What is the biggest challenge you are facing in your research?
A. There are a lot of challenges in the research. I work in the field of medical robotics—micro- and nano-robotics—and I always found it challenging to communicate effectively with medical doctors, for them to understand what our capabilities are and for us to understand what their needs are. It’s always challenging [for doctors to] try to understand what your system does or your device is going to work in a hospital situation, in an operating room—those are challenges. It’s the interdisciplinary nature of the problem—people have different views of the world and bring those together. That’s always a challenge, that’s the challenge when you are a design engineer working with a customer.
One of the keys to our field in robotics has always been materials, coming up with better and more capable materials. When we come up with new material, that gives us a lot of capabilities. So I am a firm believer of the importance of doing material research, material science, and we do a lot of that within my group. So challenges are materials, understanding needs and then also in putting systems together that work together.
Q. So are your research results, to some degree, commercialized? Do you work with any companies?
A. We do. We have five companies now that have spun out of the group. So that is one of the things I think coming from Switzerland—the country really encourages entrepreneurialism, provides a lot of coaching for young students to help them understand the difficulties of really turning an idea into a product. So I think that is exciting, especially for young students trying to get into [the field] and take those risks, learn a lot and maybe even get a reward once in a while.
Q. How do you think companies should find the balance to both do research and commercialize the results of the research?
A. It depends on the industry—some industries require more research. I used to work with the disk drive industry and Seagate Technology, comparatively they put a tremendous amount of their revenues back into research. But that changes over years; people’s business models change. One strategy many companies use these days is clearly they wait for the startups, they let startups do the research and the development and then they acquire them after they get success. You could kind of understand why that’s attractive from large company’s standpoint. But I think we never have to forget about basic fundamental research—sometimes we get too far on the development side, and if you just develop things, all of a sudden you’ve run your well dry, and you have to prime the pump with research. We should never forget the importance of doing basic research.
Q. So in your industry, you have to focus on research a lot.
A. Fortunately in ETH Zurich we have a large enough group that we are able to do some more fundamental research where we can publish good, high-impact papers. And then along the way, some of those ideas we can develop more into an engineering prototype, and then when we feel good about it, we have the right team of students who are excited and they can grab that idea and spin off the company. We are fortunate that we’re in a country and at a university that can support a big enough operation to do that. But I think we kind of bridge it all—one of the challenges these days is connecting the basic research people to the engineering development side.