US study finds that ChatGPT has limitations in providing environmental justice information for rural counties.
Researchers at Virginia Tech University in the US have identified limitations in ChatGPT’s ability to provide location-specific information on environmental justice issues. These limitations may indicate potential geographic biases within the chatbot. The researchers conducted a study where they prompted ChatGPT to address environmental justice concerns in all 3,108 counties across the United States. ChatGPT, developed by OpenAI, is a powerful artificial intelligence tool trained on vast amounts of natural language data. These large language models, also known as AI tools, can understand and generate textual responses based on user prompts.
The researchers found that while ChatGPT demonstrated the ability to identify site-specific environmental legal challenges in large, densely populated areas, the tool had limitations when it came to local environmental legal issues.
They said the AI model was only able to provide location-specific information for about 17 percent, or 515, of the total 3,018 counties it was queried about. Their findings have been published in the journal Telematics and Informatics.
“We need to explore the limitations of the technology to ensure that future developers recognize the potential for bias. That was the motivation for this study,” said Junghwan Kim, an assistant professor at Virginia Tech University and lead author of the study.
The researchers said they chose environmental law as a research topic to broaden the range of questions typically used to test the performance of generative AI tools.
The US Department of Energy describes environmental justice as “the fair treatment and meaningful participation of all people, regardless of race, color, national origin, or income, in the development, implementation, and enforcement of environmental laws, regulations, and policies.”.
Asking questions by county allowed the researchers to measure ChatGPT responses to sociodemographic parameters such as population density and median household income, they said. The counties they studied ranged in population from 1,00,19,635 (Los Angeles County, California) to 83 (Loving County, Texas).
The team found that in rural states like Idaho and New Hampshire, more than 90 percent of the population lived in counties that could not access local data.
On the other hand, in states with larger urban populations, such as Delaware or California, less than 1 percent of the population lived in counties that cannot access specific data, the researchers said.
“Although further research is needed, our findings reveal that the ChatGPT model currently has geographic biases,” said Kim, who teaches in the Department of Geography.
According to study co-author Ismini Lourentzou, who teaches in the Department of Computer Science, the findings hinted at problems with “reliability and flexibility of large language models.”
“This is a starting point for exploring how programmers and AI developers can anticipate and mitigate knowledge gaps between large and small cities, and between urban and rural environments,” Kim added.
