Simulated induction & its application to botanical key generation

The dissatisfaction expressed by taxonomists with the results obtained from automatic key-generation methodologies employing deductive logic led to an examination of interactive key generation methodologies employing inductive logic. Philosophical and psychological aspects of induction were examined to ensure that the resulting methodology would be philosophically and psychologically acceptable, and a case was made that such methods would in fact be more widely understood in the community than deductively-based methodologies. The effects of the discussion on the debate about the existence of artificial intelligence and the problems of obtaining rules for expert systems were noted, and a computerised methodology implemented. The results of applying this methodology to Tasmanian data obtained from measurements of specimens of the Acaena complex and Danthonia genus were compared with several competing methodologies, namely clustering, neural networks, discriminant analysis, a paper-based key produced by a domain expert and entropy-based methodologies. The results obtained were either similar or superior to the competing methodologies; perhaps because the methodology implemented combined the strengths of each of the participants, i.e. the tireless calculating ability of the computer with the background knowledge and common sense of the domain expert. With some types of data, the methodology was also less computationally intensive than some competing methodologies.