Application of Artificial Intelligence Techniques to Economic Planning

The relevance of computer science to economic planning is defended. An algorithm for constructing a balanced economic plan is presented and found to be of time order Nlog(N) in the complexity of the economy. The time taken to perform a complete plan optimization in natural units for a whole economy is estimated. Plans and computational costs This is a paper by a computer scientist that disputes an economic hypothesis. The hypothesis rests upon certain premises about computation and is therefore open to criticism from outside economics. The hypothesis is that: the complexities involved with performing the calculations required to optimize an economic plan in natural units are so great that they are beyond the realms of feasibility and that in consequence all rational economic calculation must proceed by means of the intermediary of prices. The hypothesis is an old one. It was originally proposed by Von Mises (1936) between the wars. In recent years it has been ably restated by Nove (1983) and has gained circumstantial support from recent developments in the USSR. It is argued here that computer science has something relevant to say on the topic. It is then argued that the difficulties experienced in socialist economies with planning in natural units are partly a consequence of the particular formalism used to express the problem. Finally, it is argued that techniques developed in artificial intelligence can be applied to solve planning problems with economically acceptable computational costs. In what follows it is assumed that planning authorities start out with a pre-given objective in terms of net output. That is to say they wish to achieve a specified bundle of output flows at the end of the plan period. If possible they would like this bundle of outputs to be exceeded, but excesses of some commodities are not desired if they result in shortfalls in others. In other words these are the classic assumptions proposed by Kantorovich (Ellman 1972). We do not concern ourselves with how these plan requirements are arrived at. It is further assumed that the plan authorities have effective property rights over all means of production and can allocate them between different productive activities in order to achieve plan goals. Relevance of computer science Computation is always a physical process. It is always performed by real physical mechanisms. These may be humans, humans aided by pen and paper, humans aided by calculators or electronic computers. At some point in the future these may be replaced