9. Epistemology and the Digital Computer

During the second half of the twentieth century computers enter into the picture.

It is natural to think of digital computers as ``electronic brains''. Whereas previous machines automated physical activities, the digital computer performs tasks which are normally considered mental. That they required detailed prescriptions in terms of simple elementary steps suggests at first that their capabilities will fall short of those of intelligent human thinkers, but results in the theory of computation obtained even before the appearance of digital computers, suggest that any information processing task can be accomplished by these means. The argument was made by Alan Turing as early as 19509.1 that there was no obstacle in principle which prevented appropriately programmed computers from any mental task which an intelligent human might accomplish.

In practice this proved harder to accomplish than many hoped, the rest of the century passing without quite cracking the nut. Though computers still fall short of human brains in many areas (and excel them in others), there is no question that they can now be seen as cognitive agents, at the very least appearing to have knowledge and perform inferences on that knowledge to obtain further knowledge.

The great bulk of what is done by computers is most naturally thought of as information or data processing. Computers store large amounts of data and obtain yet more from that data by computation. But from very early in the history of computing, some of that data has been deliberately crafted to represent propositions on which computations are performed which constitute sound deductions. Though the contexts in which computers are intended to be seen as storing propositions and performing inferences are relatively few, the data in general is intended to be accurate data about some aspect of the real world. The data files when understood as intended can be understood as representing propositions, and the computations as inference. The distinction between data and knowledge may be just a matter of interpretation. The perception as data is easier, the bit patterns in the memory can be manipulated blindly accorinding to some algorithm. The perception as propositions requires more information, one needs not only the bit pattern but some idea of its significance, and this extra information (semantics) is crucial to an appreciation of what inference is accomplished by some computation.

Roger Bishop Jones 2016-01-07