What if some of these human faculties cannot be imitated by operations on symbols? That need not be the end of the road for AI. From the very earliest days nonsymbolic machines have been used for computing. Until the end of the 1940s analogue computers were as promising as digital ones. An analogue computer has no store of symbols waiting to be pushed around by instructions, but instead carries out computation by representing values directly, as electrical voltages, or lengths, or turns of a gear wheel. (A slide rule (see figure 1.3) is a simple example of an analogue computer, where numbers are represented as lengths along a piece of wood or plastic.) Analogue computers fell out of fashion largely because each one was designed to do just a limited range of tasks, such as solving differential equations, unlike the more versatile digital machine.
[IMAGE ]
Figure 1.3: A slide rule.
Another strand of AI, which gained strength in the 1950s and 60s, is neural modelling. The human brain does not seem to have one central processor controlling passive memory cells, but a complex network of active neurons (brain cells). The mystery is how a ``large number of highly interconnected elements [the neurons] which apparently send very simple excitatory and inhibitory messages to each other'' (McClelland, Rumelhart, and the PDP Research Group, 1986) can support the intricacies of human reasoning. There is now a great revival of interest in neural modelling, both as an attempt to understand the fine detail of human cognition (those processes that go on below our awareness, for example, the recognition of visual images, coordination of hand and eye, and learning of concepts), and more generally as a simulation of the human brain and a competitor to the symbol-processing tradition of artificial intelligence.
Most of the work in neural modelling is carried out by simulating the neural networks on a conventional computer, but machines are now being designed with not one complicated central processor, but hundreds of simple distributed processors, connected together somewhat like neurons in the brain. The hope is that not only can these model the subsymbolic actions of the brain, but that they can also be used as the building blocks for machines that mimic the many levels of human thought, from the stimulation of a single neuron to the sophistication of conscious reasoning. That is still only a hope; we are still a long way from being able to design `electronic brains'.
![[IMAGE ]](img5.gif){kind=link}