A new study suggests that ChatGPT could be used to prevent future epidemics. For the study, researchers developed a model that simulates the spread of an illness similar to Covid-19.
According to the Daily Beast newspaper, the fake model developed by a team of researchers at Virginia Tech, was a Covid-like deadly airborne pathogen called the “Catasat virus”, causing symptoms that range from a light cough at best, to a fever and moderate cough at worst.
For their model, the researchers created 100 different personas with names, ages, personality traits, all living in the fictional town of Dewberry Hollow.
ChatGPT predicted the behaviors of those personas after the outbreak of the virus in three different experiments. In the first, the personas were given no additional health-related information like how much the virus is spreading in the town and how Catasat was affecting them. In the second, the personas were given information about their own health—allowing them the ability to self-quarantine if it chose to do so. In the third, the personas were given information about their own health and the town’s growing number of cases.
The first experiment resulted in the epidemic spreading until nearly every citizen of Dewberry Hollow was infected. However, once the personas were informed of their own health situation in the second experiment, there was a sharp decline in generative agents leaving their house and number of overall Catasat cases.
Armed with the full gamut of information and context about the virus, though, the generative agents in the third experiment were able to reduce the number of cases and bend the curve much more quickly than the previous experiments.
“We coupled an epidemic model with ChatGPT, in a unique and innovative way to predict human behaviors during epidemics,” said Ross Williams, a doctoral student in industrial and systems engineering at Virginia Tech.
“We think generative AI has the potential to provide us with synthetic data on human behavior so policy makers can make more informed choices,” Williams added.
According to the team, their findings indicate that AI could significantly help prevent future epidemics by providing a fairly accurate picture of how we’d respond to a potential disease outbreak.
