Reply to Krellenstein on Parallel Computation

In replying to my argument that the speed furnished by parallel processing is highly relevant to the understanding of intelligence and mind, Krellenstein says that “the serial simulation of parallel processes and the portability of software are computational principles and everyday empirical realities.” While it is true that in principle parallel processes can be simulated on serial machines, the practical reality is that such simulations can be painfully slow. Ten years ago graduate students in artificial intelligence could be told not to worry about parallel computation because you could always simulate it on a serial machine. But now that speed-ups achieved in serial processes are running up against physical constraints such as the speed of light, increasing amounts of research in computer science in general and also in AI are concerned with parallel computation. According to Denning (1986), current supercomputers are within a factor of 10 of the processor speed limit determined by the speed of light in a chip. Once this limit is reached, the in-principle serial simulation of parallel processes will be no more of an “empirical” reality than the in-principle simulation of any computer by a Turing machine. The developers of connection machines and other parallel architectures (see references in Thagard, 1986) have important practical reasons for seeking the speed afforded by parallel processing that should not be ignored for abstract philosophical reasons. As for portability, at the practical level it is noteworthy that a great deal of non-trivial work is required to port programs between such simple machines as the IBM PC and the Apple Macintosh, let alone between more complex machines.