Researchers have built a simple silicon chip that mimics the operation of brain cell networks
By Yanai Ofran
What is in the brain that allows it to perform tasks that no computer can do? Many models have been proposed in the past to explain what is so special about the way the brain processes information. This week there was a significant breakthrough when a group of Swiss-American researchers reported that they succeeded in developing a silicon chip based on the operating principles of brain cell networks. This simple processor, it turns out, manages to easily perform tasks for which ordinary computers need a large amount of resources, time and energy. The researchers expect the new chip to beat all of its electronic competitors, at least in certain types of calculations.
Electronic computers, for their part, are expected to retain the advantages they have over the brain in other types of calculations. Calculating the product of two 100-digit numbers, for example, is a matter of seconds even for an old-fashioned home computer, but an impossible task for a person without paper and pencil.
But the brain can casually solve problems that sophisticated computers fail to solve even after prolonged effort. Facial recognition, for example. It is not difficult to recognize Haim Yavin on the street, among dozens of anonymous faces - even if you see him, let's say, only from the profile, an angle from which even the most dedicated Mbat viewers have not seen him before. The mind can deduce within a fraction of a second, even from this angle, how this face looks from the front and link it to the face of Haim Yavin, stored in memory. At the same time, the brain ignores all the huge flow of information that is being fed to it from all the other faces, the strangers, on the street.
The ability to recognize faces is not unique to humans. Chimpanzees, for example, can determine with great success whether two foreign chimpanzees presented to them in a picture are mother and son. But when it comes to computers, it just doesn't work. There is no computer today, however sophisticated, that can do something similar, neither in terms of accuracy nor in terms of efficiency and speed.
Not only is the brain able to sort out only what is relevant from a mass of data, but even if the information being fed to it is imperfect or inaccurate it is sometimes able to fill in the gaps. If he does not have the tools to give an exact answer, he manages in most cases to provide a fairly good approximate answer, and above all, he is able to learn and improve in performing tasks. Each of these capabilities is a difficult problem for a computer.
How does the brain manage to do this? The difference probably lies in the architecture - the structure of the calculation units, which is designed in a completely different way in the two systems. The anatomical structure and the basic mechanisms of action of the brain cells have been known to researchers for decades, yet it is still not clear how they enable the performance of these complicated tasks. The brain is a system of millions of billions of cells. Each of them can receive and transmit electrical currents to thousands of other cells. When such an electrical current reaches the target cell, it can cause it to fire electrical currents into the cells it is in contact with, or alternatively silence it. The current can also not affect the cell at all - it all depends on the strength of the current and the type of cell from which it was received.
Theorists have been trying for years to explain how such a network can perform such complicated calculations. Richard Henlozer from Switzerland and a group of American researchers working with him decided to test these theories. They tried to imitate the working principles of the brain with a simple system of electrical circuits on a silicon chip. The system built by the group has 16 electrical cells representing 16 brain cells. Each of the cells can flow electric currents to itself and to the cells around it, so that if it receives a signal, that is, if an electric current is flowed to it, it transfers currents to its neighbors, who also transfer currents to their neighbors and so on. In the center of the system they put a suppressor cell - a cell that can receive current from any of the cells, and in response send them signals that reduce the strength of the current they transmit to their neighbors. Such types of connections are very similar to the types of connections that brain cells have between them.
When the researchers examined the way in which this simple system processed electrical currents, they discovered characteristics that are very reminiscent of the brain's behavior. When many electrical currents were fed into the system to many cells at the same time, it ignored most of the currents and processed only the strongest current, similar to the attention mechanism of the brain - the mechanism that focuses attention on the important signals and ignores noise and information that is not relevant at that moment. There are also computerized systems that are able to select the strongest signal from a pool of signals, but this system is much simpler and more efficient.
In the latest issue of the journal "Nature", where the researchers report on the system, they also talk about calculations and measurements they made, which show that the way the system processes information is very similar to the activity of brain cell networks in other respects. For example, they discovered a great similarity between the behavior of the system and the brain activity, as measured by physiologists, when learning and using memory.
Researchers around the world consider this research to be one of the biggest breakthroughs in recent years, which can revolutionize both brain research and computer science. The system, they hope, will pave the way for the formulation of a theory that will finally explain how the simple basic units that make up the brain - the cells and the connections between them - work in coordination to create a sophisticated calculating machine. Such a theory could not only help in understanding the brain, but perhaps also provide new tools for computer science - tools that will provide computational ways to solve problems that until now seemed intractable.
{Appeared in Haaretz newspaper, 26/6/2000}
https://www.hayadan.org.il/BuildaGate4/general2/data_card.php?Cat=~~~318315066~~~207&SiteName=hayadan
