Virtual ICCM III

In the fan effect, reaction time (RT) increases as a function of fan size (i.e. the number of associations of a fact). Spreading activation in ACT-R provides a good account of the fan effect at low fan size (i.e., 1--4). However, little is known about the predictions of ACT-R at ecologically valid scales. We developed a general guessing mixture model (GMM) within ACT-R in which a guessing process is triggered by retrieval failures, and analyzed the predictions for fan sizes much larger than those used in laboratory experiments. Our analysis revealed the following properties of the GMM: RT increased as a function of fan size, but stays within a plausible range (< 2 seconds) as long as the retrieval threshold is not excessively low, and, in the limit, accuracy asymptotes at the value of the guessing bias parameter. We discuss practical challenges with testing the predictions at larger fan sizes.

Many decisions we face in life are sequential, where alternatives appear over time. We often must decide whether to take the opportunity and stop searching or to continue evaluating potentially better future alternatives. Research suggests that humans are notoriously poor at stopping optimally in sequential decision-making tasks. These sequential decisions are difficult because they involve the consideration of how past, present, and future decisions affect the outcome. Recent research suggests that the wisdom of the crowd (WoC) --- that is, aggregated decisions of many people that outperform most individuals --- can be applied to sequential decision tasks and potentially help improve stopping decisions. However, current models rely on a process of fitting human data, making it difficult to understand how those individuals would behave in new problems. Furthermore, these models do not account for the learning process that humans experience while making these decisions. In this work, we demonstrate how simulated agents using a cognitive model derived from Instance-Based Learning Theory (IBLT) can produce WoC that is similar to WoC from human participants in two sequential decision tasks. We demonstrate that the WoC performance from simulated groups of agents is better than the performance of most agents and that the Instance-Based Learning (IBL) crowd behavior is similar to the human crowd behavior. Thus, cognitive models that account for learning and experience can be used to inductively predict the behavior of human crowds in sequential decision tasks.

In previous work, we provided a neurally-based Actor-Critic network with biologically inspired grid cells for representing spatial information, and examined whether it improved performance on a 2D grid-world task over other representation methods. We did a manual search of the parameter space and found that grid cells outperformed other representations. The present work expands on this work by performing a more extensive search of the parameter space in order to identify optimal parameter sets for each configuration using one of four representation methods (baseline look-up table, one-hot, random SSPs and grid cells). Following this optimization, the baseline, one-hot and random SSPs methods did show improvement over the previous study, in some cases showing performance as good as grid cells. These findings, combined, suggest that whilst the baseline and one-hot methods do perform well once optimized, grid cells do not necessarily require optimization in order to produce optimal performance.

Accurately fitting cognitive models to empirical datasets requires a robust parameter estimation process which is often arduous and computationally expensive. A way to mitigate this challenge is to integrate participant-specific and efficient mathematical models such as a drift diffusion model (DDM) into the parameter estimation process of cognitive modeling. In this study, we exhibit a clear mapping of the parameters outputted by DDM onto the declarative memory parameters utilized in the cognitive architecture, ACT-R. We show a fairly consistent recovery of simulated ACT-R parameters using DDM and a successful application in using this method to optimize ACT-R simulated fit to an empirical dataset. Notably, we show that the DDM-derived estimated parameters are individualized to the original participant, providing a unique opportunity for parsing out individual differences in cognitive modeling. This method outlined here allows one to estimate ACT-R parameters without the need to manually build and run an ACT-R model while also allowing for neural contextualization of DDM parameters.

There have been increasing challenges to dual-system descriptions of System-1 and System-2, critiquing them as being imprecise and fostering misconceptions. We address these issues here by way of Dennett’s appeal to use computational thinking as an analytical tool, specifically we employ the Common Model of Cognition. Results show that the characteristics thought to be distinctive of System-1 and System-2 instead form a spectrum of cognitive properties. By grounding System-1 and System-2 in the Common Model we aim to clarify their underlying mechanisms, persisting misconceptions, and implications for metacognition.