Building a Brain 

By Lesleyanne Banks

One of the goals of science is to better understand ourselves. Part of that understanding has been the search for the basis of our intelligence. We have looked, and in some cases found, intelligence in other animals, such as dolphins, primates and parrots. But however clever these creatures appear to be, capable of learning and communicating, they have not surpassed our intellectual abilities. Researchers have tried to build Artificial Intelligence; making a computer or machine think or do what people can do already. Many approaches have been taken, such as Fuzzy Logic which models the uncertainty of language, and Expert Systems which are very powerful predictors in their fields. However, the best model for how we think and learn is the Neural Network, based on the structure of the brain.

Frank Rosenblatt of Cornell University proposed the first neural network in 1957. While his first machine was incapable of learning, he proved that simple, man-made networks similar to neurons could read perform recognition tasks and had memory. Rosenblatt continued his work, and in 1960 demonstrated a machine that could learn. The basic unit he proposed was one which acts similar to a biological neuron. Rosenblatt s work sparked interest in neural nets and artificial intelligence, both scientifically and publicly. Rosenblatt felt that the importance of his work lay in the possibility of non-human systems to show human cognitive functions not possible through digital computers.

However, much of the progress stopped in 1969 with the publication of by Marvin Minsky and Seymour Papert. They proved that perceptrons could not perform computational tasks. With theorems, they demonstrated perceptrons limited use and shortcomings, such as their inability to perform simple logic. During the 1970 s research into neural nets was abandoned as unimportant and without application.

The criticisms of perceptrons slowed the development of artificial intelligence for almost fifteen years. Finally, in 1982 and 1986, neural nets were reconsidered as advancement of Artificial Intelligence. The Hopfield Networks in 1982 could store and retrieve patterns similar to the human brain, and recall with only partial information. In 1986, NETTaIk could learn to read and translate into sounds with a 80% success rate. Since then, neural net research has grown rapidly, which many different kinds in use. Today, we use them for pattern recognition, optical character recognition (reading), predicting outcomes and problem classification. Limitations, such as slow speed are being overcome with the use of more interconnected nets. With a success rate of 80% or higher, and a capability for learning through training, they are becoming a viable form of artificial intelligence.

In this decade, improvements in neural networking have increased the possibilities of its use. In 1997, sales reached $1 billion worldwide, and spanned such industries as medicine, engineering, energy and financial industries. Popular science fiction has adopted the neural network as the basis for artificial intelligence. In many movies, robots and human-like androids have very complex neural nets that enable themto take on human characteristics and intelligence. The perceptron is the basis for all current neural networks, and has been expanded into other use in Artificial Intelligence. All these advancements began with a simple neuron-like device that was criticized and proclaimed useless only twenty years ago. 

Questions:

1. Why was research abandoned on neural networks? What might have stimulated its comeback?

2. What are some current applications of neural networks? 

3. What traits would a machine have to exhibit for you to considerit intelligent?

4. How could you tell if a machine or network was learning 

5. How would your life change if machines had a human equivalent of artificial intelligence?

References:

Dewdney, A. K. (1997). Yes, We have no Neutrons: An Eye-opening Tour Through the Twists and Turns of Bad Science. New York: JohnWiley and Sons Inc

Rosenblatt, Frank (1962). Principles of Neurodynamics: Preceptrons and the Theory of Brain Mechanisms. Washington: Spartan Books.

http://ei.cs.vt.edu/~history/Preceptrons.Estebon.html

http://www.ee.umd.edu/medlab/neural/nn1.html

http://www.geocities.com/ResearchTriangle/Lab/8751/index.html