How strategies are learned.

I propose that both individuals and organizations can learn strategies in complex or ill-defined situations more effectively, using the process of inductive versus deductive reasoning. To support this proposition, a theory of strategy learning is developed, where a strategy consists of a set of moves which can be combined many ways to respond to business situations. Strategy learning is captured by three basic processes: variation, adaptation, and selection. My theory is implemented using a computer-simulated game among artificial agents, with each agent represented by a computer program capable of intelligent learning. This game can make more significant predictions for a reputation game than traditional game theory. Computer simulation results are then compared with results from corresponding economic experiments. They indicate that sophisticated strategies such as reputation building can be learned inductively. Using computer simulation of the genetic algorithm, Part II explores different hierarchical levels of organizational evolution. In the case presented, selection at the population level of analysis influenced learning capabilities at the organizational level, while the learning capabilities influenced the selection process. Thus, it is crucial to understand how strategy learning at these different organizational levels interact with each other. Part III discusses the importance of long chains of strategy moves. A mathematical model demonstrates that specific strategic instances can be identified effectively with a "bottom-up" hierarchical search, which first looks for the smallest, stable parts of an action chain and then combines them into larger and larger sub-strategies until the final strategy is identified. This model can predict how humans learn specific strategic instances when provided with intermediate feedback. Empirical evidence from a pilot experiment indicates that humans use hierarchical search to discover action chains. This work provides a theoretical foundation for the evolution of intelligent organizations. If individuals can learn sophisticated strategies inductively, then so can organizations. Organizations should be able to function like large, parallel-distributed information systems, capable of learning strategies more effectively and able to leverage their strategic decision-making capabilities through the contribution of many intelligent individuals.PhDBusiness AdministrationUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/103715/1/9409643.pdfDescription of 9409643.pdf : Restricted to UM users only