Calculating and reasoning are very different processes. Computers calculate complex problems with a speed and accuracy that even human prodigies cannot begin to compete with. On the other hand, no computer can read this sentence, understand it, and accept or dispute it. For these operations we need a brain, and computer scientists have tried for years to make computers (fast, accurate calculators) reason like a brain. They have even given their field the audacious title of Artificial Intelligence, or AI. Nonetheless, after many years, many billions of dollars, and some notable partial successes, AI, though well established, is widely agreed to be hopelessly inadequate for its ultimate goal of humanlike reasoning. It appears that, after all, the brain is not a computer. But if it is not a computer, what is it?
In the past we have sought to understand the brain in terms of current technology. We regarded the brain as a switchboard, or a computer or a hologram. Today many scientists take a revolutionary yet simple viewpoint: A brain is a brain. Nothing else is like it. To understand the brain, we must understand brain structure and functioning, not computers, switching networks, or holograms. To make something that behaves like a brain we must give it brainlike structure. Nothing else will work.
Current efforts to develop brainlike processors are focused on what are called artificial neural networks or, often, simply neural networks. Yet studies of the human mind and brain continue to raise new challenges to the brain modelers.
It appears that human thought involves more than just the brain. The brain is the "mechanical" substrate that permits human thought. In order to perform tasks based on intention, emotion, purpose, and so forth, much more than that substrate is needed. Goals, emotions, and aesthetics must themselves enter the brain as signals, yet explaining how this happens is well beyond current science. It seems most unlikely that artificial and inherently predictable neural networks will spontaneously solve these problems and develop "wills of their own." This is the stuff of science fiction, not of current science.
The subject of this discussion is optical neural networks. In this article we will show how light can be used to make and operate powerful brainlike structures that are both faster and more energy-efficient than their creators--human brains. These optical neural networks are also much faster and more energy-efficient than any present or future electronic systems.
These dramatic claims are being substantiated in labs in the United States, Japan, the Soviet Union, and Europe. The Japanese call neural networks the sixth generation and have an extensive electronic and optical effort underway. They see neural networks as the logical step beyond the highly touted fifth generation of artificial intelligence for which they have already mounted a major research offensive.
Neural networks seem to be particularly promising for applications to the recognition of signals or patterns. This could range from the problems of speech recognition and translation to the monitoring and control of such a complex system as a factory.
Neural
...
Read Full Article
|
