by Researchers from Virginia Tech have recently uncovered limitations in the capability of ChatGPT, a large-language model developed by OpenAI Inc., to provide location-specific information regarding environmental justice issues. The study, which was published in the journal Telematics and Informatics, suggests the presence of potential biases within current generative artificial intelligence (AI) models. ChatGPT is widely used for various applications, ranging from content creation and information gathering to data analysis and language translation.
Assistant Professor Junghwan Kim, an expert in geography and geospatial data science at the College of Natural Resources and Environment, emphasized the considerable potential of generative AI. However, he also stressed the need to investigate the limitations of such technology to ensure that future developers recognize and address possible biases. This motivated the researchers to conduct the study.
To assess ChatGPT’s performance, the research group used a list of the 3,108 counties in the contiguous United States and prompted the ChatGPT interface to provide insights on the environmental justice issues in each county. Environmental justice was selected as the topic to expand the range of questions usually employed to evaluate generative AI tools. By asking questions based on counties, the researchers aimed to compare ChatGPT’s responses against sociodemographic considerations, such as population density and median household income.
The research team surveyed counties with diverse populations, ranging from Los Angeles County, California, with over 10 million residents, to Loving County, Texas, with a population of merely 83. The generative AI tool demonstrated an ability to identify location-specific environmental justice challenges in densely populated areas. However, it exhibited limitations in identifying and providing contextualized information on local environmental justice issues. As generative AI is becoming increasingly utilized as a source of information, assessing potential biases in its output is crucial for improving programs like ChatGPT.
Assistant Professor Kim, who teaches in the Department of Geography, commented, “While further study is necessary, our findings highlight the presence of geographic biases within the ChatGPT model.” He added, “This marks the beginning of investigating how programmers and AI developers can anticipate and mitigate the informational disparity between large and small cities, as well as between urban and rural environments.”
Notably, Kim has previously published a paper that explores how ChatGPT comprehends transportation issues and solutions in the United States and Canada. His research group, Smart Cities for Good, focuses on using geospatial data science methods and technology to address urban social and environmental challenges.
Assistant Professor Lourentzou, who teaches in the Department of Computer Science and is affiliated with the Sanghani Center for Artificial Intelligence and Data Analytics, stressed the issues concerning the reliability and resiliency of large-language models. He expressed hope that their research can guide further enhancements to the capabilities of ChatGPT and other similar models.
Ravina Pandya, Content Writer, has a strong foothold in the market research industry. She specializes in writing well-researched articles from different industries, including food and beverages, information and technology, healthcare, chemical and materials, etc. With an MBA in E-commerce, she has an expertise in SEO-optimized content that resonates with industry professionals.
