Measuring the role of mental model complexity on individual behavioral and neural differences in adaptive decision making

点击翻译按钮获取中文摘要

PROJECT SUMMARY To make good decisions in uncertain environments, humans build and update ‘mental models’ of relevant environmental statistics that can be used to make predictions and guide decision-making. When the environment changes, these models need to be adaptable to retain their predictiveness. This kind of adaptability typically involves key information-processing trade-offs that are well understood theoretically but have yet to be applied substantially to our understanding of human brain function and behavior. Here I examine systematically how these trade-offs, measured both from behavior and brain-imaging data, relate to the considerable variability in decision-making abilities that are typically evident across subjects and task conditions. My focus on behavioral, computational, and neural mechanisms of individual variability in decision-making abilities is particularly relevant to long-term research in mental health. Decision-making is severely disrupted in a number of mental illnesses including anxiety, schizophrenia, and addictive behaviors, but the exact mechanisms underlying these disruptions have yet to be fully elucidated. My central hypothesis is that individual and task-dependent differences in adaptive decision-making reflect systematic variability in the complexity of the mental models upon which the decisions are based. In the fields of statistics and machine learning, predictive models compress past observations into representations that can generalize to the future. A model’s complexity determines the flexibility with which this compression can account for new information. Complex models are more adaptive (low bias) but can overfit spurious observations, leading to more behavioral variability. In contrast, simpler models tend to have higher bias but lower variability. This tradeoff between bias and variance is well described in statistics and machine learning, but its influence on human mental models and decision-making behavior is not well known. The two primary aims of this project are: 1) to develop a principled measure of mental complexity that can be applied to human behavioral data; and 2) to identify the influence of mental model complexity on neuromodulatory brain networks involved in the mental exploration required for adaptive decision-making, and how activity in these networks differs across individuals. By linking a strong theoretical framework with methods from information theory, psychology, neuroscience, and computational modeling, the current proposal will provide a novel lens with which to examine behavioral and neurobiological sources of individual variability in human decision-making. Moreover, the results of this research will provide crucial insights for interventions aimed at understanding and improving decision-making processes affected by mental illnesses.