Computational Primitives for Cost-Benefit Decision-Making

Abstract

Cost-benefit decision-making is a critical process performed by all organisms, including humans. Various factors, including risk1,2, uncertainty3, age4, sex5, and neuropsychiatric disorders6, can alter decision-making. To explore cost-benefit decision-making in humans, we developed a comprehensive task and analysis framework that presents participants with a series of approach-avoid trade-offs across a variety of contexts. With this system, we found that cost-benefit decisions in humans are made using a set of computational strategies that may be used for integrating costs and rewards, which we term decision-making primitives. We further show that these decision-making primitives are used by rodents performing a similar decision-making task7. We find that utilization of these primitives in both rodents and humans shifts based on factors like hunger and sex, and that individuals use primitives differently. We additionally demonstrate that using a naturally-inspired neural network architecture generates output that overlaps with human and rodent performance over a non-constrained neural network. This novel conceptual framework, by isolating discrete decision-making primitives, has potential to help us identify how different brain regions give rise to decision-making behavior, as well as to facilitate better diagnosis of neuropsychiatric disorders and development of naturally-inspired artificial intelligence systems of decision-making.