|Ph.D Student||Kinneret Teodorescu|
|Subject||On the Decision to Explore New Alternatives|
|Department||Department of Industrial Engineering and Management||Supervisor||Full Professor Erev Ido|
|Full Thesis text|
The decision whether to explore new alternatives or to choose from familiar ones is implicit in many of our daily activities. How is this decision made? When will deviation from optimal exploration be observed? My thesis connects different lines of research under a unified umbrella of exploration decisions, in an effort to shed light on the basic properties of the decision to explore new alternatives. The experimental approach includes a context-free framework, in which participants are faced with a simplified multi-alternative task. In this simplified framework, different payoff structures can be easily manipulated and the optimal exploration level can simply be derived.
In the first part, the coexistence of under- and over- exploration is examined and sufficient conditions for suboptimal exploration behavior are outlined. Results revealed higher sensitivity to the frequent outcome than to the average outcome from choosing new alternatives. That is, the implicit decision whether to explore a new alternative reflects underweighting of rare events: Over-exploration was documented in “Rare Disasters” environments, and insufficient exploration was evident in “Rare Treasures” environments. These results can be captured with a model that distinct between “data collection” and “outcome-driven” decision modes. Under the data collection mode, the decision maker collects information about the environment, to be used in future choices. Under the outcome-driven mode, the decision maker relies on small samples of previous experiences with familiar versus unfamiliar alternatives, before the selection of a specific alternative. The predictive value of this model is demonstrated.
Implications of the main findings to the famous learned helplessness phenomenon are further investigated in the second part. After experiencing an environment in which they cannot control their fate, individuals tend to give up too early. Exploration efforts remain low even when the environment changes and control is possible. Based on the previous part, the frequency of rewards from exploration is suggested to account for this effect. Using the simplified exploration task, for the first time, both absence of control and the difficulty of obtaining rewards are manipulated. Contrary to traditional accounts, results suggest that the difficulty of the task is a better predictor of exploratory behavior than perceived controllability: The classic learned helplessness effect emerged only in a moderate difficulty condition and not when the task was easy or too difficult. However, this pattern was not reflected in perceptions of control. The suggested model is shown to explain both behavioral and perceived controllability findings.