Raúl Rojas González is a well-known expert in artificial neural networks and professor of computer science and mathematics. He was born in Mexico City in 1955 and moved to Germany in the early eighties. He teaches at the Free University of Berlin and gained a lot of attention with his projects like the FU Fighters (world champions in fobotic football 2004 and 2005) and the autonomous car project „Spirit of Berlin“.
For our readers who are not into science, what are generally artificial neural networks about?
Neural networks are abstract models of the way the elementary components of the brain are supposed to work. This research really took off only after the first computers appeared and people started thinking that there should be a common denominator for information processing in the brain and in machines. Artificial neural models are still too simple if we compare them with real neurons, but they give us an idea of what is happening inside the brain.
At the simplest level, you can think of an artificial neuron as a logical gate, that can combine a few inputs and produce a result. A neural unit could be one that just counts zeros and ones, and fires if the majority of the connected neurons are firing. This so-called “majority logic” has been proved to be extremely effective for building computers … or for wiring brains. More complex models of artificial neural networks can be used to recognize patterns, for example faces or speech. The good thing is that we know how to make them learn, so that we do not have to program everything from scratch. The system learns from experience.
When will machines be as intelligent as humans? Or will they ever?
I think that people living today will not see it. It depends on the definition of machine. If we are meat machines, we are already intelligent. If you ask about metallic machines, they will probably never reach the level of complexity and integration of biological material. So, I think the question will not be answered in this century, and I guess that genetic engineering is closer to that goal than computer scientists.
What would be the biggest challenges to achieve that?
The biggest challenge is how to achieve self-organization and learning. You do not program a child. It learns over many years. The brain is mostly a “learning machine”, it is built to absorb knowledge. That’s the job of the brain: to learn. That is what we cannot still do so elegantly with machines. We can program a machine to beat the chess world champion, but that machine will hardly develop insight or intuition. It will still be dependent on its programmers.
Concerning Artificial Intelligence there has always been one philosophical question: Are machines/computers allowed to be smarter than human beings? (like in the movie Space Odyssee 2001 etc.)
I think that they should be allowed … but as I said before, we will not see it this century. I think that Ray Kurzweil, to name a visionary, underestimates the power of biology and the effect of millions of years of evolution. The “singularity” is still very far away, hidden at the end of the rainbow.
Computers will be smarter than humans (as in Space Odyssey) when they learn to cheat and lie. That is the highest form of intelligence because it requires a “theory of mind”. I have to put myself in the shoes of the other people in order to lie effectively. Monkeys do not have a theory of mind, as has been proved with experiments. They are terrible liars. When computers could lie to us as effectively as HAL, those computers would pass the Turing Test and would be intelligent.
If we stick to the movies, is there a film which gives us a realistic scenario for the usage of robots in the future? (A.I., iRobot, Terminator, The Jetsons, etc.)
Most robot movies are not that good. Robots of the future will not necessarily have the form of humans. A laundry robot will look more like a washing machine with arms. The garbage truck will get also arms and will pick the garbage cans from the streets: no need of having humanoids running around the garbage truck. The function determines the form, that is why there are so many strange looking animals. Robots of the future will fill a complete zoo of different forms and purposes.
The movies iRobot and AI are therefore just too simple minded, not ambitious enough. They try to sell the cute robot helping at home that works like a humanoid slave. But the best robots are those you don’t notice. A escalator is a robot for moving people. The car-washing machine is a robot. They can be made better, but most of the robots of the future will not look like humans and will be special-purpose. Most of them will be integrated in the environment and you won’t notice them.
Do robots/computers learn the same way as we do? What are the elementary differences?
Computers can learn from Big Data. We learn from a few experiments and extrapolations. Our method is more flexible, but learning from Big Data can lead to impressive results, like for example classifying all the people in your pictures using face recognition.
We do not really know how people learn. Machine learning is at its infant stage. We will be able to do much better in the coming years.
You and your team at FU Berlin have built an autonomous car. A very good example where computers could drive safer than human drivers and make our society a safer one. Why are we so afraid of the capabilities of modern technologies? Or is it even a good sign to be sceptical?
The main problem with new technologies is who is going to lose his/her job next. This is what has been called the “end of work”. I remember the German banks when I first came here. There was a lady who would handwrite in your bank booklet how much money you were taking from your account, or depositing. Even in Mexico we had more ATMs than in Germany. Now the bank lady is gone – the complete bank is gone! We all do online banking and there is an ATM in every corner. I always joke that the next addition I would like for my home would be an ATM in the cellar. And once electronic cash takes off, not even the ATMs will be needed.
People displaced from banking could get a job somewhere else, but economists are starting to worry that we could be going too fast. People are not as flexible as machines. They must be trained and retrained.
In the case of autonomous cars I have always said that these will be the taxis of the future, able to move several persons, without having to park and waste space and time. We could move Berlin with one fourth of the total cars we now have. This would be good for the environment, less good for the automobile industry, who is already worrying that young people prefer to do car sharing. So, social aspects of technology are always a difficult issue.
How should we approach technologies as they are dominating our everyday life more and more?
I think we should try to move slower. People now expect you to be online 24/7 and complain when you have not answered an email within 30 minutes. Disconnecting and going slower is in fact something that some entrepreneurs are seeing as an opportunity. Your smart phone can give you a limited “budget” of calls or page views. Your computer can try to limit your screen time. It is great to have all these technologies available but we should try to avoid speeding up our pace at every new turn of the technology.
How important is it to question and/or hack the usage of technologies?
Questioning technology is very important. It is like the opposition in parliament. You need a government but you also need someone looking the government over the shoulder. Joseph Weizenbaum was a prophet of doom. He once came to our Lab and when he saw our small and cute soccer robots he immediately said: “that is military technology”. But he was right. Those are the two sides of the coin for every person working with computers today.
You are running the successful robot football team FU Fighters. What can we learn from ball kicking humanoids?
Robotic soccer is like a laboratory for the development of basic technology in a gaming setting. The robots have to be able to see, move, coordinate, plan and execute. These are all capabilities that we would like future service robots to have. There is a large community doing this kind of research. Several thousand persons take part each year in the RoboCup competition. We have been playing for 15 years now, and we have been twice the robotic soccer world champions.
Which contributions can AI and robotic make for a better social coexistence in the future?
Robots can be used in industry and at home. Robots should not be used for warfare. There is a general agreement that chemical weapons are brutal. The same can be said of some kind of robots and drones which lower the aggression threshold.
So, I am all for using robots for peaceful purposes and for helping people. However, not everything should be automatized. Sometimes cleaning your home is the only “sport” some people still do. Mowing the lawn can clear your head and some people actually like to do it. In the movie Wall-E all humans are just sitting in sofas and machines feed them, entertain them, move them around. That is not a future I would like to see.
You mentioned once that there are three main problems for our future. Energy, production/manufacturing and the population. What can we and the governments do to prevent a global collapse?
The three main challenges for the 21st Century are: moving to sustainable and clean forms of energy, redistributing manufacturing across the globe, maybe through the use of some kind of flexible manufacturing facilities, and eliminating poverty. These are three very difficult objectives and all countries should work towards them. Sometimes, when you read the newspapers, you can wonder if we are actually moving forward in this direction. Today, for example, we have already used all the earth resources, which can be restored in one year. And we still have four months ahead of us in 2013.
From your perspective, what potential does AI generally have to improve the world we are living in?
The curious thing is that many things we now have, starting from computers or smartphones that you can talk to, were called AI some years ago. The first computers were called “mechanical brains”. In Finland the word for computer is “knowledge machine”. But when you have it, it becomes just a gadget of everyday use. AI has already improved the world. It is just that when it materializes, we do not call it AI any more.
So working on AI is like walking towards the horizon. You never arrive, but on your way you find wonderful things. The horizon is still there to be explored. It could sound disturbing but actually I think it is a very good thing. Kurzweil’s singularity is that middle point at the horizon.