History [of ideas about thinking]
Ever since antiquity immense amounts have been written about human thinking. Until recent centuries most of it was in the tradition of philosophy, and indeed one of the major themes of philosophy throughout its history has been the elucidation of principles of human thinking. However, almost all the relevant ideas generated have remained forever controversial, and almost none have become concrete enough to be applied in science or technology. An exception is logic, which was introduced in earnest by Aristotle in the 4th century BC as a way to model certain patterns of human reasoning. Logic developed somewhat in medieval times, and in the late 1600s Gottfried Leibniz tried to use it as the foundation for a universal language to capture all systematic thinking. Beginning with the work of George Boole in the mid-1800s most of logic began to become more closely integrated with mathematics and even less convincingly relevant as a model for general human thinking.
The notion of applying scientific methods to the study of human thinking developed largely with the rise of the field of psychology in the mid-1800s. Two somewhat different approaches were taken. The first concentrated on doing fairly controlled experiments on humans or animals and looking at responses to specific stimuli. The second concentrated on trying to formulate fairly general theories based on observations of overall human behavior, initially in adults and later especially in children. Both approaches achieved some success, but by the 1930s many of their positions had become quite extreme, and the identification of phenomena to contradict every simple conclusion reached led increasingly to the view that human thinking would allow no simple explanations.
The idea that it might be possible to construct machines or other inanimate objects that could emulate human thinking existed already in antiquity, and became increasingly popular starting in the 1600s. It began to appear widely in fiction in the 1800s, and has remained a standard fixture in portrayals of the future ever since.
In the early 1900s it became clear that the brain consists of neurons which operate electrically, and by the 1940s analogies between brains and electrical machines were widely discussed, particularly in the context of the cybernetics movement. In 1943 Warren McCulloch and Walter Pitts formulated a simple idealized model of networks of neurons and tried to analyze it using methods of mathematical logic. In 1949 Donald Hebb then argued that simple underlying neural mechanisms could explain observed psychological phenomena such as learning. Computer simulations of neural networks were done starting in the mid-1950s, but the networks were too small to have any chance to exhibit behavior that could reasonably be identified with thinking. (Ironically enough, as mentioned on page 879, the phenomenon central to this book of complex behavior with simple underlying rules was in effect seen in some of these experiments, but it was considered a distraction and ignored.) And in the 1960s, particularly after Frank Rosenblatt's introduction of perceptrons, neural networks were increasingly used only as systems for specific visual and other tasks (see page 1076).
The idea that computers could be made to exhibit human-like thinking was discussed by Alan Turing in 1950 using many of the same arguments that one would give today. Turing made the prediction that by 2000 a computer would exist that could pass the so-called Turing test and be able to imitate a human in a conversation. (René Descartes had discussed a similar test for machines in 1637, but concluded that it would never be passed.) When electronic computers were first becoming widespread in the 1950s they were often popularly referred to as "electronic brains". And when early efforts to make computers perform tasks such as playing games were fairly successful, the expectation developed that general human-like thinking was not far away. In the 1960s, with extensive support from the U.S. government, great effort was put into the field of artificial intelligence. Many programs were written to perform specific tasks. Sometimes the programs were set up to follow general models of the high-level processes of thinking. But by the 1970s it was becoming clear that in almost all cases where programs were successful (notable examples being chess, algebra and autonomous control), they typically worked by following definite algorithms not closely related to general human thinking.
Occasional work on neural networks had continued through the 1960s and 1970s, with a few definite results being obtained using methods from physics. Then in the early 1980s, particularly following work by John Hopfield, computer simulations of neural networks became widespread. Early applications, particularly by Terrence Sejnowski and Geoffrey Hinton, demonstrated that simple neural networks could be made to learn tasks of at least some sophistication. But by the mid-1990s it was becoming clear that—probably in large part as a consequence of reliance on methods from traditional mathematics—typical neural network models were mostly being successful only in situations where what was needed was a fairly straightforward extension of standard continuous probabilistic models of data.