Modeling the confidence of predictions: A Time Based Approach

Everyday life demands explanations and predictions from everybody all the time. Using experience based knowledge, the human mind is well suited to draw the required causal inferences. However, due to failures in the past, such inferences are usually drawn under uncertainty and come along with different degrees of confidence. We present an ACT-R model describing the cognitive processes of induction and deduction for a prediction task in a simple, simulated technical environment. While ACT-R provides excellent mechanisms to capture causal learning and causal inferences, no process has been defined yet to account for the trust humans put in their predictions. Based on the availability heuristic by Tversky and Kahneman (1973), we propose an approach for modeling different levels of trust by using a temporal module from Taatgen, van Rijn and Anderson (2007), thus relating availability to retrieval time and confidence judgments. The forecasts of our model are compared with the results of an empirical study and nicely fit the experimental data.