BLOG

Published 15 September 2016 by Stephanie Hanel

Artificial intelligence and neurorobotics

Some reach too far, some do not reach far enough. No, humanoid robots will not achieve world domination tomorrow. Yes, it is within the realms of possibility that one day, the ‘machine’ that stands in front of us will be a ‘person’. Or, as the philosopher and AI expert, Klaus Mainzer, describes it: He considers it possible that the crucial leap towards superintelligence is at a point where we are combining evolutionary and technical strategies and creating a ‘neuromorphic computer’, which links technical efficiency with evolutionary benefits.

What does this mean? Technical progress goes hand in hand with computer speed and an increase in memory capacity. But to automatically simulate the processes in the human brain requires a greater amount of energy – which is in complete contrast to the natural method of learning gains, in which the cross-linking density increases in a way that saves energy. Bringing these worlds together is no longer entirely a distant dream, but is illustrated by the work done by Tony Prescott, Professor of Cognitive Neuroscience at the University of Sheffield, on his iCub project, for example.

iCub_iit
iCub, Credits: Istituto Italiano di Tecnologia (“IIT”) 2016

iCub is a humanoid robot, which, in addition to seeing, hearing and feeling, can also use its 53 degrees of freedom in a coordinated way thanks to its sense of position, power and movement. It speaks, reacts to its environment and learns in a number of ways including by imitation. “Its control system is modelled on the human brain so that it ‘thinks’ in a similar way to you and I”, says Prescott. How is that possible? Was it not always the case that robots could only know whatever we had directly fed them, i.e. adult knowledge that is then retrieved for example in quiz shows. While a surprising level of success was achieved in this way, the process was also subject to the classic pitfalls in which the artificially intelligent machine gave the right yet also wrong answer, because it could not weigh up the difference between the factually correct answer and the answer that was actually required.

But new dimensions are opening up in the area of neurorobotics. Psychology and neuroscience are delivering key findings that are then picked up by the engineering sciences and programming. (Computer) linguistics can also make an important contribution. In the case of iCub, the crucial impetus came from disciplines that track the way the human brain works and deal with the composition of our conscious mind. For example, iCub’s control system mimics key processes in the (mammalian) brain. It also learns in the same way that an infant does, by executing small movements and observing the consequences of such movements. The robot thus acquires a degree of body awareness, as iCub must learn to distinguish between itself and the rest of the world.

Credits: A. Abrusci / Istituto Italiano di Tecnologia (
iCub was created and developed at Istituto Italiano di Tecnologia in Genoa (Italy) about ten years ago. Its software and hardware are completely open source in order to facilitate global knowledge about humanoid robotics. Credits: A. Abrusci / IIT 2016

The crucial next step is that iCub uses past experiences for the present and for events that are still to occur. To do this, scientists try to simulate the information processing activity that takes place in the relevant areas of the human brain that are responsible for autobiographical memories. By the way, expert teams from many different European countries, primarily Italy and Spain, are working on the iCub project, always at a certain interface. Their vision is that when all the components that make up human consciousness are perfectly simulated and compiled, the result will be a robot that can function side by side with humans.

For Klaus Mainzer, the crucial step in the discipline of artificial intelligence was the further development of sensors. iCub would not be able to collect any empirical knowledge if it did not have a sense of touch. Industry 4.0 would be inconceivable without sensors. Here, it is not just a case of, as has already happened, assuming that assembly line work will be supported by the partial deployment of robots. Instead, the world of things will be populated with, for example, sensors, and workpieces and workbenches will communicate with one another. The result: production processes that will be largely fully automated. Medical research in turn is driving the vision of nanorobots that travel through our bloodstream until they have fulfilled their respective mission and receive the order to self-destruct.

The future will show whether such “babies” like iCub and its successors will actually have been brought to life for the benefit or scourge of humankind. Or, as Nobel Prize Laureate Eric Betzig says: “Every technology is like a newborn baby – when it is born, you think it might become president or cure cancer. But in the end, you’re perfectly happy when it just stays out of jail.”

Stephanie Hanel

Stephanie Hanel is a journalist and author. Her enthusiasm for the people behind science grew out of her work as an online editor for AcademiaNet, an international portal that publishes profiles of excellent female scientists. She is an interested observer of new communication channels and narrative forms as well as a dedicated social media user and science slam fan.