Converging for effective exploration : how to learn across unique successes

Abstract

Organizations are often advised to engage heavily in exploration in order to succeed - casting a wide net for diverse solutions that are superior to what it currently exploits. However, what is the organization to do when the fruits of its exploration are not commensurate with one another? If all solutions appear beneficial, but each recommends differenth decisions for the same organizational choice, how should an organization learn from them? Unfortunately, such a mixed bag of learning opportunities is likely to be the case on a rugged learning environment, where solutions succeed not because of specific individual choices, but due to the complementarities between these choices. By applying the learning mechanisms in March (1991) onto an NK landscape, this paper is able to show that such a challenge is surprisingly surmountable - with March's algorithm performing almost as well on a rugged landscape as it does on a smooth one. While the rugged learning opportunities may initially stymie the organization, March's inherent process of mutual learning enables explorations to grow progressively similar, so as to converge upon a smooth locality. On this smoothed locality, valuable explorations then become salient enough to learn from. The counterintuitive takeaway is thus that in order to capitalize on diverse explorations, an organization must first engage in convergence.