| |
Given the overwhelming use of computation in modern science and the continuing difficulties in quantifying the results of complex computations, it is of increasing importance to understand its role in the essentially Popperian scientific method. There is a growing debate but this has some distance to run as yet with journals still divided on what even constitutes repeatability. Computation rightly occupies a central role in modern science. Datasets are enormous and the processing implications of some algorithms are equally staggering. In this paper, some of the problems with computation, for example with respect to specification, implementation, the use of programming languages and the long-term unquantifiable presence of undiscovered defect will be explored with numerous examples. One of the aims of the paper is to understand the implications of trying to produce high-integrity software and the limitations which still exist. Unfortunately Computer Science itself suffers from an inability to be suitably critical of its practices and has operated in a largely measurement-free vacuum since its earliest days. Within CS itself, this has not been so damaging in that it simply leads to unconstrained creativity and a rapid turnover of new technologies. In the applied sciences however which have to depend on computational results, such unquantifiability significantly undermines trust. It is time this particular demon was put to rest.
|
|
