Giulio Sandini, accompanied by a talk by Edgar Körner of the Honda Research Institute Europe, gave an insight look into recent developments on the way to learning robots at ESOF2010 in Turin. This thought provoking talk gave the determining factor for this interview.
Following Giulio’s fascinating session at ESOF 2010, EuroScientist asked him a few questions approaching the philosophy of robotics.
You mentioned that it is impossible for the human brain to “un-learn”. Could you describe what is the difference between unlearning and forgetting?
Once you have learned something you cannot voluntarily go back in time and try to replay the process that made you learn. When you have to solve an equation you can go through a set of processes until you get the solution and then, if you want to you can go back and follow all the steps again. With learning, once you have the solution all the intermediate steps are lost. As people say, once you have learned to ride a bicycle you never forget. Of course this is not true for all our memories (I do not remember the phone number of my first girlfriend). My point was more on the process of learning than on the memory itself.
Do you think, that robots might have an advantage because we can teach them how to “un-learn”?
This is not necessarily an advantage also because you could “delete” a memory but you cannot prevent the process of re-learning it (but it is an interesting point).
Does the field of Robotics, especially the discipline dealing with artificial intelligence, benefit from discoveries in medical research? Could a better understanding of Alzheimer’s disease or Dementia play an essential role in teaching Robots how to learn?
Yes, medical and neuroscience research are definitely important sources of knowledge if one is interested in building systems with human-like intelligence. I would not say that a better understanding of Alzheimer or Dementia could play a role in teaching a robot how to learn, but it could certainly give new information about how humans learn and this could be useful. The reverse is also true in the sense that building robots that learn may clarify which are the essential ingredients of the process and explain why the failure of some may cause Alzheimer-like dysfunction.
But dealing with the creation if intelligence is kind of a philosophic quest. Do you have to be a philosopher to work in the field of artificial intelligence?
Maybe so, the definition of a philosopher is fuzzy. In the past it was intended to study matter (so also physical phenomena) so, in some sense, also robots.
EuroScience is an international and interdisciplinary organisation: How is research in the field of robotics and artificial intelligence organised and handled, in respect to international co-operation and interdisciplinary research?
In the past, robotics was mostly in the field of engineering, but more recently some of the long-term objectives of robotics, such as learning, have expanded the field to include cognitive and social sciences as well as neuroscience. Interdisciplinary research is possible in some centers around the world, such as the IIT where I currently work, and is supported by specific programs such as EU research programs in the IST (Information Science and Technology) area.
Teaching robots to learn will affect our understanding of how we learn – and vice versa. This sounds very exciting.
Yes indeed! If it is not useful to both engineers and neuroscientists the collaboration does not work in the medium and long terms.
Giulio Sandini is Director of Research at the Italian Institute of Technology (iit) and full professor of bioengineering at the University of Genoa. After his graduation in Electronic Engineering (Bioengineering) at the University of Genova in 1976 he was research fellow and assistant professor at the Scuola Normale Superiore in Pisa until 1984. During this period, working at the Laboratorio di Neurofisiologia of the CNR, he investigated aspects of visual processing at the level of single neurons as well as aspects of visual perception in human adults and children. He has been Visiting Research Associate at the Department of Neurology of the Harvard Medical School in Boston where he developed diagnostic techniques based on brain electrical activity mapping. After his return to Genova in 1984 as associate professor, in 1990 he founded the LIRA-Lab (Laboratory for Integrated Advanced Robotics, www.liralab.it). In 1996 he was Visiting Scientist at the Artificial Intelligence Lab of MIT. (taken from the iit-webpages)