Smart machines and robots: An interview with the expert Ulrich Eberl

Ulrich Eberl: Well, of course, you could well ask whether my small NAO robot is a smart machine. I don’t think so. You can teach him a lot, for example, to quote from Hamlet, play soccer or buy a pretzel for me from the bakery – which is pretty amusing. But they’re modes of behavior you dictate to machines.

In my view, smart machines are more. They are becoming more and more like us humans, in all possible fields. They can walk, grip things, speak, listen, see, recognize objects and handle them. These are all things that NAO can also do already to a limited extent. What he can’t do – and that’s something smart machines are already capable of – is read and write, for instance.

Ulrich Eberl: There are numerous smart machines that can read natural-language texts. However, they are not robots like NAO, but computer programs. They can read specialist medical literature and patients’ files, for example, in seconds and then advise physicians. Or they can carry out research in legal texts for attorneys and help in making financial investments by analyzing business reports and correlating them with the latest news and stock market data.

Robot journalists are also able to write simple texts themselves, such as local weather reports or sports results. The machine creates a short text as soon as it’s known, for example, what the final score of a soccer game is and who scored the goals when.

Ulrich Eberl: Yes, and that is indeed the crucial aspect. Today’s smart machines constantly learn new things, for example, by observing and imitating. There are already robots in factories where you simply demonstrate a movement to them and they can replicate it right away.

Ulrich Eberl: The deep learning methods which have caused such a big stir for the past five years are based on the principles of the human brain. The machines often have billions of artificial nerve cells that are interconnected in a complex manner – they learn through reinforcement and feedback. We’re no different. When we’ve learned often enough that a red light means “Stop, danger!”, that association comes to our mind as soon as we see one.

If you show a deep learning network millions of pictures of animals or faces, it will become perfect in recognizing animals and people. The same goes for voice recognition, text analysis or translation programs. Google or intelligent loudspeakers like Alexa improve with every search query or voice command.

Ulrich Eberl: Yes, you can! And it’s already being done. My favorite robot is the iCub at the Istituto Italiano di Tecnologia in Genoa, which, incidentally, is a hot spot for robotics and artificial intelligence. In Genoa, the iCub learns like a child at kindergarten. It sits in a classroom or walks about there and constantly wants to explore new things. That way it learns how to clear the table or what it can do with its toys. For example, it picks up a beaker, looks at it and asks its human teacher: “What is that called?” And the teacher answers: “That’s a beaker and it’s orange.” Whenever the iCub learns something, it receives a reward.

Ulrich Eberl: Well, not by feeding it more power, for example. It’s a bit like good grades at school. The robot simply receives points to collect. That means it plays its life like a computer game. If it does something right, it’s awarded a point. While conducting research for my book “Smart Machines,” I also met such an iCub in Osaka, the world capital of robotics, in Japan. It won points for correctly predicting what the person in front of it would do next. I was sitting once at the table next to his teacher and a cup of tea was in front of them. The iCub said: “You will take the cup and drink.”

The only problem was that the cup was too far away. The human could not have reached out and picked it up – and the robot also saw that. What it then did was what was interesting: It simply pushed the cup toward the human. As if to say: “Now take it, I want my point.” The exciting thing was that no one had programmed it to do that. The robot had even taught itself how to push a cup. This kind of reward learning might be one of the best ways to make robots suitable for use in everyday life one day.

Ulrich Eberl: In the future we will definitely need something like teachers for machines. After all, you need to teach the machine what is sensible and how to solve problems. People will naturally assume that task, just like we bring up our own children. If we want machines that make decisions autonomously, we need to ingrain our ethical rules in them. That’s why there are now initial university chairs for machine ethics. How does a machine differentiate between good and evil? That’s an extremely important question.

Ulrich Eberl: An interesting question. I’ve encountered Roboy often, a nice little robot with a skeleton and an egg-shaped head. It can say something like: “Oh, I’m so shy.” Then it looks to the ground and blushes. That’s really sweet. But, of course, you need to be clear about one thing here: That’s only simulated. Machines don’t have real emotions. You can make them feel hungry for electricity when their battery’s running down, or feel pain when an engine runs hot, but they aren’t human feelings.

Ulrich Eberl: Researchers in Japan are building machines that you can hardly tell apart from humans. When you’re standing in front of such androids, you have to look closely to tell whether they’re a real person or a machine. Even their skin is soft and warm and has pores. The eyes are perfect. As are their mouth and hair. They speak like a human, flutter their eyelashes and smile. It’s pretty amazing.

A Danish professor commissioned a clone of himself in Japan. He then took it with him to Copenhagen and let it hold a lecture. And – this is no joke – the professor said that the students there only noticed at the break that it wasn’t their professor holding the lecture.

Ulrich Eberl: The cooperation will be very close. That’s already the case in part now. The company Glory in Japan manufactures automated teller machines, for example. Fairly humanoid robots sit there next to people and work together with them. They hand each other workpieces, assemble components together – and what’s funny: The robots even join in the morning sport session. Now you may well ask: What’s the point of that? The machines don’t need it. That may be the case. But it makes the robots more likable, say Glory’s managers.

Ulrich Eberl: Robots taking over the world is a very western myth. I’ve also learned that from my travels around the world. That’s completely different in Japan. We in the west do indeed have a tradition of conflict between human and machine. That goes back a long way. To the golem, to Frankenstein. HAL 9000, Blade Runner, Terminator, Transformers and all the rest of them, in movies, literature and plays. Human against machine – that’s what it’s about in the west.

It’s completely different in the east. In Japan, for instance, there’s a centuries-old tradition of helping machines. As far back as three hundred years ago, the Japanese built large mechanized puppets that served tea, for example. And so it continued. After the Second World War came mangas, video games and anime films with Astro Boy, a robot boy who helps people against evil. That’s a completely different tradition: It’s about harmony between human and machine. That’s why I believe there’s a lot of cultural influence behind the issue. And a lot of science fiction thinking. Super-intelligent machines which threaten humanity are far removed from reality.

Ulrich Eberl: Smart machines – in other words, robots and learning algorithms – will definitely change just about all jobs. Yet the crucial question is: Will they take our jobs away in aggregate terms or even create more new ones than they destroy? There are many studies on this in the meantime. One of the most well-known is the Oxford study, in which 700 occupations were examined to determine the likelihood that the main activities in them can be automated in the next twenty years.

Ulrich Eberl: Yes. That’s shown by a look at the three countries that now have the highest degree of automation in industry: South Korea, Japan and Germany. They also have the lowest unemployment. That’s actually counter-intuitive. You would think that people would become jobless where there’s a lot of automation. But if you do it right, industry will be more competitive and products can be sold well on the world market. That in turns means there’s enough work.

Ulrich Eberl: All in all, you can say that more has happened in the field of smart machines and artificial intelligence in the past five years than in the whole 50 years before. It’s truly a field that’s booming, and we’re just at the outset. In the next 20 to 30 years, the computing power, storage capacity and data transfer rate of microchips will increase thousand-fold – and their price will still be the same as it is now. At the same time, software is also becoming increasingly powerful – and that means a boost in performance in terms of the learning methods for smart machines and how knowledge is generated and processed.

That trend will impact all of us. Regardless of our age and occupation. These smart machines will become just as integral a part of our life as smartphones are now. The first smartphones came on the market ten years ago and many of us now can’t imagine life without them. The pace of advances in them is enormous. That will be similar with smart machines. We’ll see them everywhere, often without noticing they’re learning machines. We’ll live in a community with smart machines and work hand in hand with them. We’ll have them in the home, in the office, in factories and on the roads. Everywhere.

Last update: May 2018