An official website of the United States government.

We've made some changes to EPA.gov. If the information you are looking for is not here, you may be able to find it on the EPA Web Archive or the January 19, 2017 Web Snapshot.

EPA Researchers use AI to Mimic Human Behaviors that Could Affect our Exposure to Chemicals

Published December 13, 2018

EPA’s new method allows scientists to mimic human decision-making for four behaviors (sleeping, eating, commuting and working) that might affect our exposure to chemicals.EPA’s new method allows scientists to mimic human decision-making for four behaviors (sleeping, eating, commuting and working) that might affect our exposure to chemicals.Where and how we spend our time – like eating breakfast around the table at home or riding a bike on the commute from work – plays a major role in the types of chemicals we’re exposed to each day.

For EPA scientists, understanding exposure to chemicals is an important consideration in how they assess risks to human health. One piece of data required for studying human exposure is a record of an individual’s daily activities, including where they’re spending their time. Traditionally, researchers have relied on surveying people about their daily activities but collecting enough data this way can be a challenge. Surveys also can’t capture information that people aren’t aware of (like consumer products being used by people nearby) or information that’s too time-consuming to report (like the amounts of all consumer products used in a given day).

EPA researchers have proposed an alternative to relying on human activity surveys by creating a method that models human behaviors using artificial intelligence (AI).

Their new method incorporates an “agent-based model” into a “needs-based” AI program, which allows scientists to mimic human decision-making for behaviors that might affect our exposure to chemicals. The AI program can simulate the behaviors over extended periods of time, making the framework and models based on it able to generate human behavior data suitable for use in exposure assessments.

EPA researcher Namdi Brandon says it’s important to consider human behavior over time because chemical exposure can have a snowball effect, slowly accumulating across days or weeks.

“When someone is exposed to a chemical from a consumer product on a given day, it’s possible that a fraction of the chemical might remain in their body for the following days,” Brandon says. “Meanwhile, there may be exposures to the same or other chemicals during the same time, causing the chemicals in their body to accumulate over time to possibly hazardous levels.”

EPA’s model currently addresses four exposure-related behaviors – sleeping, eating, commuting and working – but doesn’t account for more complex actions, like raising children or unscheduled group activities. Though, Brandon believes the framework could be extended to include new behaviors or interactions.

“We could improve the model to have multiple agents within a household,” he says. “With multi-agent households, we could have agent-to-agent interactions, in which one agent may do an activity that may affect the needs of the other agent such as: child rearing and communal responsibilities like cleaning, food preparation, yard work, etc.”

Beyond modeling exposure to chemicals, this research has many possible applications. “Because this method allows us to simulate people’s daily routines in and outside of their homes, scientists can use the human behavior patterns generated by the model to better understand human exposure to both indoor and outdoor air pollutants,” Brandon notes.

Likewise, Brandon suggests the method could be used to better understand exposures related to residential water use. “By simulating behaviors related to water use – like bathing, using the faucets for cleaning, drinking or cooking – we may be able to obtain more detailed estimates of exposure due to use of residential water sources,” he says.

Learn more about this research in the article “Simulating exposure-related behaviors using agent-based models embedded with needs-based artificial intelligence.” Exit