Multi-agent learning and the descriptive value of simple models
Artificial Intelligence
Organizational Learning: From Experience to Knowledge
Organization Science
| Hi-index | 0.00 |
Individuals and social systems are often portrayed as risk averse and resistant to change. Such propensities are characteristically attributed to individual, organizational, and cultural traits such as risk aversion, uncertainty-avoidance, discounting, and an unwillingness to change. This paper explores an alternative interpretation of such phenomena. We show how the reproduction of successful actions inherent in adaptive processes, such as learning and competitive selection and reproduction, results in a bias against alternatives that initially may appear to be worse than they actually are. In particular, learning and selection are biased against both risky and novel alternatives. Because the biases are products of the tendency to reproduce success that is inherent in the sequential sampling of adaptation, they are reduced whenever the reproduction of success is attenuated. In particular, when adaptation is slowed, made imprecise, or recalled less reliably, the propensity to engage in risky and new activities is increased. These protections against the error of rejecting potentially good alternatives on inadequate experiential evidence are costly, however. They increase the likelihood of persisting with alternatives that are poor in the long run as well as in the short run.