Little Models, Big Results
Artificial intelligence and artificial life are nothing if not ambitious. The creation of an artificial intellect or life form is a daunting task, and daunting tasks seem to call for pulling out all the stops and using the biggest, baddest analytical tools on the block. Indeed over many, many years, AI and A-Life have thrown a plethora of sophisticated mathematical and computational techniques at solving the important problems of AI and AL, but the track record is mixed, and many of the knotty problems are problems still. This talk suggests a simpler approach to penetrating the complexity of AI and AL. In particular, a methodology of little models, using facetwise analyses, dimensional analysis, and a procedure of patchquilt integration are suggested to construct models that are especially useful in the design of AI and AL that works. The little modeling methodology is illustrated with a case study drawn from the development of competent and efficient genetic algorithms, including models of population size, run duration, and market share, and the race, and other examples are given from current work in the development of a simplified quantitative organizational theory (SQOT). The talk concludes by suggesting specific ways to adopt these techniques to advance the agendas of AI and AL.