Introduction to the special issue on goal-directed neural systems

What does genuinely intelligent behavior require? All definitions of intelligence include not only successful achievement of predetermined goals but being able to set one’s own goals. Thus far, in the development of both artificial and biological neural networks, there has been more work on the achievement of predetermined goals than on setting goals. This is true both because achieving existing goals is easier than setting new goals and because of the influence on researchers of psychological notions such as instinct and stimulus-response. Yet as the brain has become more complex, the architectures for satisfying evolutionbased drives have gradually expanded into architectures capable of evaluating both the external environment and the body’s internal states, and using their evaluations to select courses of action. The same capabilities have also gradually entered into artificial neural systems used for diverse applications including pattern recognition, robotics and navigation, prosthetics, and video game design. This special issue is an offshoot of a post-conference workshop on Intentional Neural Systems at the August 2005 International Joint Conference on Neural Networks in Montreal. The workshop was followed a year later by a two-day conference on GoalDirected Neural Systems with many of the same speakers, held in October 2006 at Arlington, Texas, and jointly sponsored by the Texas SIG of the International Neural Network Society (INNS), the Metroplex Institute for Neural Dynamics (MIND), and the University of Texas at Arlington. After the editors-in-chief of Neural Networks approved our proposal for a special issue based on updates of presentations given at the Arlington conference, we invited several additional authors who did not attend the conference but had made important contributions to the theme of the issue. The change from the term ‘‘intentional systems’’ to ‘‘goaldirected systems’’ was made to avoid potential concerns about whether or not artificial neural systems could possess conscious intentions. However, intentionality remains an important consideration in many of the articles in this special issue. In much of