The brain-machine disanalogy.

The comparative study of information processing in brains and machines leads to a picture in which disanalogies are more fundamental than analogies. The major dichotomy is between evolvability and programmability. Brain models, to be tenable, must pass an extended Turing test in which the capacity to self organize through the Darwinian mechanism of variation and selection is a key element. Programmable machines that simulate the type of structure-function relations that allow evolution to occur are, however, too inefficient in their use of resources for problem solving to support cognitive abilities comparable to those of biological organisms. Furthermore, real evolutionary systems are open in that it is always possible for them to tap previously unexploited physical interactions for computing. Nevertheless, computer simulation provides a powerful tool for studying brain function; and non-programmable designs that exploit the high efficiency, high adaptability domain of computing are in principle possible.