Friday, June 22, 2012 - 9:00am
Rhim's research uses the tools of signal processing and detection theory to study models for human decision making. When deciding between alternatives based on uncertain or noisy observations, people are known to act approximately as Bayesian decision makers; it is as if they assign costs to incorrect decisions and attempt to minimize the expected cost. Such a minimization gives a decision rule that depends on prior probabilities for each of the alternatives being correct. However, it is implausible for a human to use a different decision rule for each decision, as the previously-studied models require.
In Rhim's work, the natural human tendency to form categories for decision problems is identified with quantization of the prior probabilities. Rhim is able to obtain intriguing qualitative conclusions about decisions made by voting. For example, even when each vote counts equally and each voter has an observation with the same accuracy, the voters collectively make better decisions when they form their categories differently. Rhim and his colleagues interpret this as a mathematical justification for preferring diversity when forming teams and committees. Their earlier work showed that differences in preferences lead to a quantifiable penalty of team discord, even when the team shares the common goal of making correct decisions. They are working to draw other new conclusions from their unique way of modeling human behavior.