Copy the page URI to the clipboard
Cheng, Peter C-H
(1990).
DOI: https://doi.org/10.21954/ou.ro.0000fc64
Abstract
Traditionally, the Philosophy of Science has examined the nature of scientific discovery. In recent years, Cognitive Science has gathered together work in Artificial Intelligence (AI) and Cognitive Psychology that attempts to understand scientific discovery. However, at present, there is no generally accepted account of scientific discovery in any of these disciplines.
This thesis aims further to explore the nature of scientific discovery from an AI perspective, but does so within a clearly defined Framework, designed to structure cognitive science research on scientific discovery. The framework proposes a minimum set of components as a guide to the construction of acceptable accounts of scientific discovery. The focal concept is the Research Programme; a body of research that investigates a delimited set of phenomena using a Theoretical component and an Experimental component. The framework posits: three types of theoretical knowledge; three levels of experiments; inferences to apply and generate new theoretical & experimental knowledge; criteria for assessing the acceptability of theories & the reliability of experiments; and multiple levels of communication between the components.
Previous computer models and empirical studies of scientific discovery are reviewed. They tend not to offer complete accounts of scientific discovery, as defined by the framework. In particular, many completely ignore the crucial role of experiments.
The STERN computational model of scientific discovery is introduced. It instantiates all the components of the Framework. STERN currently models discoveries made by Galileo in the domain of naturally accelerated terrestrial motion, although it may be applied more generally. STERN has four main strategies that are used to make discoveries: (i) confirming existing hypotheses; (ii) generalizing experimental results to form new hypotheses; (iii) generating new hypotheses from known hypotheses; and (iv) generating new experiments.
STERN is more complete than previous computational models. As such it allows novel heuristics at the level of research programmes to be investigated and high level abilities to emerge from its complexity.