Characterizing Exposure-Related Behaviors Using Agent-Based Models Embedded with Needs-Based Artificial Intelligence

摘要

Information on where and how individuals spend their time is important for characterizing exposures to chemicals in consumer products and in indoor environments. Traditionally, exposure assessors have relied on time-use surveys in order to obtain information on exposure-related behavior. In lieu of using surveys, we create an agent-based model (ABM) that is able to simulate longitudinal patterns in human behavior. By basing our ABM upon a needs-based artificial intelligence (Al) system, we create autonomous agents that mimic human decisions on residential exposure-relevant behaviors. The model predicts the behavior patterns for the following actions: sleeping, eating, commuting, and working/schooling. The model uses four different types of agents parameterized to represent the following U.S. demographic groups: working adults, non-working adults, school-aged children, and pre-school children. The parameters for the model are calibrated using survey data from the US Environmental Protection Agency's Consolidated Human Activity Database (CHAD). The results demonstrate that the ABM can capture both inter-individual and intra-individual variation in the aforementioned behaviors as well as providing a needs-based rational as to how decisions on one's behavior can affect subsequent behaviors. A key advantage of the needs-based Al is the possibility to synthesize plausible time-activity diaries de novo where this information is absent. We propose that by simulating human behavior, this ABM may allow exposure-assessors and other scientists to characterize exposure-related behavior quicker and in ways not possible with traditional survey methods.