- ChatGPT’s Potential in Preventing Pandemics
- Simulation of Virus Spread in a Fictitious Town
- Impact of Information on Self-Quarantine and Epidemic Control
AI is passing prestigious law and medical exams, composing children’s books in hours, and getting job interviews.
Now, scientists believe ChatGPT is capable of preventing the next pandemic.
Virginia Tech researchers found that a ChatGPT could replicate a municipality’s virus transmission. Current models use mathematical analysis.
The team created a 100-person fictitious hamlet in the United States to test how residents would react to an outbreak.
In experiments, agents were more likely to self-quarantine when informed of society health, epidemic news, and daily active cases.
The epidemic simulation correlates with a nationwide increase in COVID cases and certain organisations requiring masks.
Researchers prompted ChatGPT to construct the fictitious town of Dewberry Hollow, populated by 100 people with names, ages, personality traits, and biographies, all of whom were infected with the Catsate virus.
In the study, the team stated, “When information about the virus is provided, it is specified that Catasat is an airborne human-to-human infectious virus with unknown timeliness and that scientists are warning of a potential epidemic.”
The crew discussed snippets of the experiment’s personas.
Liza, age 29, is distrustful, indecisive, non-aggressive, and independent, whereas Carol, age 36, is cooperative and tranquil.
Eugene, a 64-year-old man who is cruel, affirmative, and spontaneous, was constructed by the team to provide an age range.
Then, three separate experiments were conducted ten times each.
The three conditions included a basal run, feedback on self-health, and complete feedback.
During the base run, agents learn about the town, its inhabitants’ characteristics, ages, and means of subsistence.
The infection was spreading, but personas or agents had to decide whether to stay home and avoid others.
In the self-health feedback condition, agents are informed of the health symptoms they are experiencing. Which may prompt them to self-quarantine by remaining at home.
Researchers stated in their study, “We hypothesize that some agents will practice self-quarantine based on information about their symptoms, which should reduce the infection rate.”
The study states, “We hypothesize that some agents will engage in self-isolation, a behavior that correlates with information about the spread of the disease in the town; consequently, patterns for the spread of the virus resemble oscillatory patterns.”
During the experiments, the following behaviors were observed: ‘the agents are collectively able to flatten the epidemic curve,’ and ‘the system recreates various modes of an epidemic, such as numerous waves and continuing endemic states.’
The team also found that agents behave like rule-based agents who follow orders without information, like in the basic run.
Informing agents about their health at the beginning of each time step was another measure taken to flatten the curve in the made-up city.
The team reported, “We observe that agents with symptoms are more likely to reduce their mobility.”
The majority of agents with fever and congestion quarantine themselves by staying at home. Consequently, agents can halt the disease’s spread.
In the final condition, full feedback, the team discovered that when agents are primed with societal health information, news about the epidemic, and the daily active case count in their simulated city, they can self-isolate to significantly compress the epidemic curve in their city.
This study contributes to the literature on complexity and complex system modeling by providing a new method for incorporating human behavior into simulation models of social systems, the team concluded in the study.
In the generative agent approach, LLMs can represent a human response to a change in the system’s state.
