Using natural language processing to measure cognitive load during use-of-force decision-making training

Abstract

PurposeFew studies have tested the efficacy of instruction based on cognitive load theory in police use-of-force (UoF) training due to limitations of existing cognitive load measures. Although linguistic measures of cognitive load address these limitations, they have yet to be applied to police UoF training. This study aims to discuss the aforementioned issue.Design/methodology/approachOfficers’ verbal behavioral data from two UoF de-escalation projects were used to calculate cognitive load and assess how it varied with officer experience level (less-experienced, experienced). The verbal data were further analyzed to examine specific thinking patterns that contributed to heightened cognitive load across officer experience levels.FindingsAcross both studies, responses from less-experienced officers contained greater usage of cognitive language than responses from experienced officers. Specific cognitive processes that contribute to cognitive load in specific situations were also identified.Originality/valueThis paper enables police trainers to facilitate the development of adaptive training strategies to improve police UoF training via the reduction of cognitive load, and also contributes to the collective understanding of how less-experienced and experienced officers differ in their UoF decision-making.

Measuring Latent Trust Patterns in Large Language Models in the Context of Human-AI Teaming

“…Research into the application of machine learning as an analysis tool in areas of psychological research (e.g., trust in automation, human-automation teaming) is just beginning to emerge (Jeong et al, 2019; Li et al, 2022; Santander-Cruz, 2022; Ta-Johnson et al, 2022; Tausczik & Pennebaker, 2010). LLMs represent words or sentences as ordered sequences of numbers (i.e., vectors) similar to x, y coordinates; however, LLM vectors have dimensions in the hundreds or thousands.…”