A Coupon-Collector Model of Machine-Aided Discovery

Abstract

Empirical studies of scientific discovery---so-called Eurekometrics---have indicated that the output of exploration proceeds as a logistic growth curve. Although logistic functions are prevalent in explaining population growth that is resource-limited to a given carrying capacity, their derivation do not apply to discovery processes. This paper develops a generative model for logistic \emph{knowledge discovery} using a novel extension of coupon collection, where an explorer interested in discovering all unknown elements of a set is supported by technology that can respond to queries. This discovery process is parameterized by the novelty and quality of the set of discovered elements at every time step, and randomness is demonstrated to improve performance. Simulation results provide further intuition on the discovery process.