Cognitive biases in a geospatial intelligence analysis task: an ACT-R model

An ACT-R model of sensemaking in a geospatial intelligence task was developed based on Instance-Based Learning Theory (IBLT). The model (a) maintains hypotheses about the probability of attacks by insurgent groups, (b) seeks new information based on those hypotheses, and (c) updates hypotheses based on new evidence. The model provides a functional account of how these sensemaking processes are carried out in a cognitive architecture, and model performance can be compared to normative (Bayesian) standards. Simulations exhibit two well-known cognitive biases that are frequently identified as problems in intelligence analysis: (1) anchoring in the weighting of new evidence and (2) confirmation bias in seeking new information.