A new study finds that when AI models are asked to illustrate small, rural towns, the results are not always realistic. The technology appears to have an easier time producing images of large cities, like Washington D.C. As more city planners turn to AI, this bias may cause more problems for rural America.
Virginia Tech geography researcher Junghwan Kim set out to test how an artificial intelligence image generator could produce an illustration of the town where he lives, Blacksburg, population 44,000.
“We also asked local residents, do you think these AI generative images looks realistic?” Kim said.
These residents agreed, the images AI produced were off. Kim’s study found that AI has an easier time producing images for larger cities than it does for small, rural towns.
“Typically this AI model performed poorer for the areas like low density areas, rural areas, or low income areas that have already suffered from the limited resources,” Kim said.
This raises ethical issues, if urban planners and policy makers increasingly look to artificial intelligence to solve problems. Areas that have historically been neglected may continue to see less authentic representation.
“It’s like a vicious cycle,” Kim explained.
Kim thinks one of things causing AI’s limitations is related to news coverage.With the rise of news deserts, there are often fewer images and news stories online that show rural communities. AI programmers also may be less aware of online publications for small towns, or news sites that aren’t owned by large media companies.
Kim noted that AI may have biases, depending on who programmed the models, and where they source the data they use. He said this can sometimes lead to gender and political bias in AI, as well as geographic bias.
