Tools


Siemens Worldwide

Pictures of the Future

Contact

Contact

sts.components.contact.mr.placeholder Sebastian Webel
Mr. Sebastian Webel

Editor-in-Chief

Tel: +49 (89) 636-32221

Fax: +49 89 636-35292

Werner-von-Siemens-Straße 1
80333 Munich

Pictures of the Future
The Magazine for Research and Innovation
 

Artificial Intelligence

Gateway to Intelligence

Professor Tomaso Poggio

Machines are not far from having the raw computing power of the human brain. What we do not yet have, however, are the algorithms to turn that power into something we would call intelligence, says Professor Tomaso Poggio.

What’s your definition of machine intelligence?

Poggio: The best definition was proposed by the British mathematician Alan Turing in 1953. He posited a situation in which you would talk with someone in a different room. If that “person” was actually a machine and you could not tell that it was a machine, you would essentially be dealing with a form of intelligence.

Is learning the gateway to intelligence?

Poggio: That is an article of faith and a reasonable claim. In evolutionary terms, in fact, primates and humans are the least hardwired beings on earth. Insects do learn. But a lot of their behavior is limited by evolution. Humans, instead, take many years to develop. For instance, before the age of ten, a child cannot recognize faces as well as adults can.

When it comes to learning from experience, how important are feelings?

Poggio: Emotions are certainly important in terms of explaining human behavior and the development of intelligence. In the context of biology, our feelings and their biochemical correlates are likely to be quite important for learning. In the context of developing machines that learn, I think feelings and emotions are not needed for learning. But if a machine is to pass the Turing test it will have to be able to simulate emotional intelligence. And this brings us to a gray area: A simulation system may be very different from a person; but if no one can tell the difference, should it make a difference to us?

What’s the biggest obstacle to machines becoming more like humans in their ability to learn?

Poggio: We don’t know! But I don’t think there are limits to the ability of machines to become as good as we are or better at learning. It will take a long time, but it certainly is not out of the question. Until about ten years ago it was easy to argue that human memory was much greater than that of any computer. But you can’t say that any more. Our memory storage cannot be much more than the number of synapses in the brain. So if we have 10 to the eleventh power neurons, then we have about 1,000 times more synapses, which adds up to about 10 to the 14th power. Now 10 to the 14th power bits — one hundred trillion — is a lot, but you can buy a terabit hard disc, which is about 10 to the 12th power bits — one trillion — for around $50. So machines are not far from having the raw computing power of the human brain. What we do not yet have, however, are the algorithms to turn that power into something we would call intelligence.

Why not? What is required here?

Poggio: At the moment, we don’t know what we need to do. If I knew, then the problem of intelligence — probably the most far-reaching challenge in science — would just be a matter of engineering. I feel that the core of the problem has to do with the integration of different aspects of intelligence — vision, language, common sense, etc. But to figure out how these elements relate to each other, we will need an effort in basic research that combines aspects of neuroscience, computer science and cognitive science. Only in this way will we gain deeper understanding of the problem and be able to move toward a solution.

Can knowledge about how the cortex functions help us to develop new learning algorithms?

Poggio: Yes. If we define intelligence as the ability to pass the Turing test, which is a test of human intelligence, then understanding the human brain is definitely going to help. And neuroscience is doing a good job of getting us there. It has been developing at an exponential rate over the last 20 years. At this point, I believe that it is just a question of time before our knowledge of how the brain works can directly help in engineering areas such as computer vision and machine learning.

Have you done any work along these lines?

Poggio: Yes. Most of our work has been with physiologists in recording signals from the brains of macaque monkeys using electrodes. This produces very precise information because it makes it possible to record data from single neurons. As a result of this work, we have been able to produce a mathematical model of the macaque visual cortex that simulates the learning activity of about one million neurons. We run this as a computer program and have trained it — using thousands of photographs — to recognize eight kinds of behaviors — hanging, running, sleeping, feeding, etc. — among mice that have been genetically altered to have autism, depression or schizophrenia. It simply marks a behavior as “hanging,” “running,” etc. on a video and enters the duration of the behavior into a statistical database. The program also detects transitions from one behavior to another, all of which adds up to a kind of behavioral fingerprint. By automating this process we have been able to objectively relate behavior to the genome.

How accurate is the system?

Poggio: We have compared the system to the output of human annotators and have found that it is at least as good, or better. And it works 24/7 without getting bored!

Could such a technology lead to surveillance systems capable of providing descriptions of human activities?

Poggio: In principle, yes. But of course such a system would need an immense amount of training. And human behaviors are far more complicated than those of mice.

Being able to show a picture to an intelligent artificial system and get a description of what is happening — is that something you are also working on?

Poggio: Yes. But we are not there yet! I think we are getting very close to having systems that can automatically tell what is in a picture, whether it is a pedestrian, a car, a bird or whatever. But there are much more complex questions, such as being able to understand what people are doing in a picture. There is no computer that can do anything like that today. So that is the next challenge.

Why is that so difficult?

Poggio: Humans benefit from a huge amount of knowledge and experience. We know how to identify cues that tell us, for instance, that one person is involved in a conversation while another is not. If you think about it, when you look at an image and interpret what is happening, that requires far more than vision — it requires intelligence.

Will machines achieve that kind of intelligence in ten years?

Poggio: The ability to describe the content of an image would be one of the most intellectually challenging things of all for a machine to do. We will need another cycle of basic research to solve this kind of question — of telling a story from an image. I think it will take at least 20 years before we have such a technology.


Tomaso Poggio is Eugene McDermott Professor in the Department of Brain und Cognitive Sciences at the Artificial Intelligence Laboratory at the Massachusetts Institute of Technology. He is also Co-Director of MIT’s Center for Biological and Computational Learning. Poggio joined the MIT faculty in 1981, after ten years at the Max Planck Institute for Biology and Cybernetics in Tubingen, Germany. He received a PhD in 1970 from the University of Genoa. Poggio is a Foreign Member of the Italian Academy of Arts and Sciences.

Arthur F. Pease
Picture credits: Kent Dayton/MIT