Dynamic decision making (DDM) involves making decisions in a changing environment while collecting information about it. In dynamic tasks, decision makers remain sub-optimal even with extended practice, unlimited time, and performance incentives -- but it is unclear why. We aim to improve our theoretical understanding of decision-making behavior in dynamic tasks: 1) how hypothesis are generated from cues in continually evolving situations and 2) how different feedback types change dynamic decision making behavior.
Hypotheses generation and feedback are important decision making mechanisms in multiple contexts, such as, 911 operators determining relative urgency to deploy resources, drivers trying to find the best route in heavy traffic, or military crews deciding how to respond to a threat. Our investigation will focus on medical decision making. An example of the kind of DDM situations we will investigate might concern a patient that exhibits symptoms of high blood sugar (e.g., blurred vision and reasoning disturbances). Tests indicate high blood sugar and low insulin (i.e., hyperglycemia). The physician's goal is to stabilize the patient's health. As the symptoms (cues) develop over time, the patient may be diagnosed with diabetes and given treatment, for example, to take insulin. Insulin often takes a moderate amount of time to have an effect. If the amount of insulin is not well calibrated to the amending physical state, it may result in too much insulin, low blood sugar, and a hypoglycemic crisis. At that point an exigent solution is required, such as to take some sugar by mouth or drink some orange juice which often would have a fast effect on the body. Ideally, a doctor should use the diagnostic hypothesis (i.e., type-I diabetes) and feedback about the patient's state to keep the system under control with correct dosages of insulin and/or sugar.
We will conduct this research using laboratory studies employing a DDM learning tool (MEDIC) and computational cognitive modeling using and extending architectures previously developed by our team. Laboratory studies will help us answer questions regarding hypothesis generation and feedback effects on human learning and performance in DDM. Computational cognitive models will propose representations of the cognitive processes and behaviors involved in learning and performance in DDM and will help us make predictions for follow-up experiments. We will test our theoretical developments in laboratory and realistic settings with medical professionals at all levels.
