10. Exploration & Exploitation and Organizational Mistakes
Updated: Jun 10
By Jerker Denrell, Warwick
Li, D., Raymond, L, and Bergman, P. “Hiring as Exploration”, working paper, MIT.
Denrell, J. and G. Le Mens (2020). ”Revisiting the Competency Trap”. Industrial and Corporate Change, 29 (1), 183-205.
Denrell, J. (2007). “Adaptive Learning and Risk Taking.” Psychological Review, 114 (1): 177-187.
Denrell, J. and J. G. March (2001). “Adaptation as Information Restriction: The Hot Stove Effect,” Organization Science, 12 (5): 523-538.
What is the impact of generalization? Bandit models tend to assume that alternatives are independent and that if we observe an outcome for alternative A, we update our belief only about alternative A and not about another alternative B. However, in many realistic settings, we do generalize to other alternatives, because we believe they are similar. What are the implications of such generalization?
How should individuals be organized to balance exploration and exploitation when everyone only has access to part of the information and there is a cost to communication? Formulated differently, what is a good, decentralized algorithm for exploration and exploitation?
How can we ensure fairness? Biases, such as the hot-stove effect, can be eliminated by exploring more. This is costly, however, but important to ensure justice. Can more cost-effective ways to explore by found, by sharing information?
What are practical algorithms for estimating the optimal trade-off between exploration and exploitation, when the parameters of the model (uncertainty etc) are not fully known? Can we know (before it is too late) that we explore too much / too little?