Metacognitive Strategies in Support of Recognition

Recent research on decision making has focused on complex real-world domains and experienced decision makers, in contrast to artificial tasks and novice subjects (e.g., Klein et al., 1993). This research has produced a revised view of decision making, one that stresses knowledge-based, recognitional processing rather than general-purpose, analytical methods. Nevertheless, a full understanding of recognitional processing and its implications for the design of decision aids has not yet been achieved. In particular, the active, controlled aspects of recognitional processing have not been fully accounted for in approaches that stress relatively automatic pattern recognition as the basis for responding. Early decision aids implemented formal, general-purpose methods for inference and choice. In inference, the ideal decision maker was thought of as generating a set of exhaustive and mutually exclusive hypotheses, making numerical assessments of the probabilistic relationships among prior beliefs, evidence, and hypotheses, and combining the numerical assessments to arrive at probabilities for the hypotheses. Once a hypothesis was accepted, the decision maker was thought of as generating all feasible options, assessing probabilities for each possible outcome of every option, evaluating each outcome in terms of every relevant evaluative dimension, and then combining the probabilities anc! values to arrive at a measure of the desirability of each option (e.g., Raiffa, 1968; Keeney & Raiffa, 1976). Many empirical studies focused on comparisons of actual behavior in artifi.cia1 tasks with formal models of this sort; the conclusion was often that human decision making is fundamentally flawed and irrational (Kahneman et al., 1982; Cohen, 1993). The most obvious decision aiding implication of this literature involved the attempt to reshape human reasoning directly in the mold of normative models, such as decision analysis (e.g., Edwards, 1968).