UB expert in urban analytics says AI can take urban planning – and society – into the great wide open
Li Yin, associate professor of urban planning at UB, is conducting research at the forefront of urban AI research. Photo by Lukas Iverson
By Rachel Teaman
Published April 8, 2025
For more than two decades, Li Yin, UB associate professor of urban planning, has applied the tools of technology and spatial modeling to generate insights on the interplay of urban space and human activity, spanning topics as diverse as walkability, property abandonment and residential segregation.
Her dissertation for the University of Colorado at Denver in 2004 was among the first to apply agent-based modeling – computational simulations of complex systems – to understand land use dynamics in the then-sprawling exurbia of the American Intermountain West. After arriving at UB later that year, she began exploring the potential of satellite imagery as a source of street-level data to study the built environment. After being told for years that it wasn’t possible, she and a team of engineers and computer scientists published a pioneering paper in 2015 that applied machine learning algorithms to Google Street View images, combing the pervasive portrait of urban space to count pedestrians as an indicator of walkability. A following publication in 2016 took the research one step further, extracting from Google Streets a key variable of streetscape visual enclosure – called the “sky view factor” in urban design – which previously was capturable only by field surveyors. The research helped pave the way for large-scale, objective assessments of walking behavior in urban environments.
Today, the urban scientist is helping to carve out new territory within the rapidly emerging field of urban AI by applying Large Language Models like ChatGPT to generate more nuanced intelligence on urban dynamics.
According to experts, AI signals a radical transformation of the smart city in its capacity to understand how humans engage with the material and social fabric of our increasingly digitized urban environment. AI is also generative, allowing for emergent and anticipatory intelligence that could reconstitute urban space and bring cities closer than ever before to the ideals of resilience and equity.
“Cities are complex systems, making them challenging to understand,” says Yin, who is trained both in urban planning and architecture. “Technology and computation have always been useful in helping us describe cities and how they behave and operate. AI, however, will change how we plan and design cities. It’s revolutionary, really.”
Yin is currently exploring ChatGPT, which enables planners to generate large-scale analyses of public discourse and opinion as captured by social media and the interconnected world of Google Maps, urban sensors and IoT devices. The release of GPT-4 in 2023 introduced image reading capabilities, allowing for even more comprehensive sentiment analysis.
“The progression and extensive training of Large Language Models has brought them in greater alignment with human language behavior, allowing for more complex and reliable analyses,” she says.
Yin’s most recent publication tests ChatGPT’s ability to discern a more granular picture of human emotion in its assessment of public response to the COVID-19 vaccine in New York State. The study ran more than 10,000 tweets posted on Twitter (now known as X) through a series of chatbot queries on vaccine literacy, awareness, receptivity and behavior.
The paper, “Unlocking Blended Emotions and Underlying Drivers: A Deep Dive into COVID-19 Vaccination Insights on Twitter Across Digital and Physical Realms in New York,” was published last year in the Urban Science journal. Yin’s co-authors are Xuanyi Nie, UB associate professor of urban planning, and Mo Han, an expert in machine learning.
In designing their model, the team engaged in a back and forth with ChatGPT, tweaking their queries with more precise and layered qualifiers. The resulting matrix of 10 interrelated emotions sheds light on the public’s complex progression through assumptions, education, influence and, ultimately, action.
For example, “relief” was the most prevalent emotion detected, but it was also closely tied to “concern.” Readings of “trust” were also followed by “hopefulness.” “Anger” was among the emotions that lacked a positive association, most often linking to “concern,” “skepticism” and even “sarcasm.”
“Cities are complex systems, making them challenging to understand,” says Yin, who is trained both in urban planning and architecture. “Technology and computation have always been useful in helping us describe cities and how they behave and operate. AI, however, will change how we plan and design cities. It’s revolutionary, really.”
Yin says the paper is part of a foundation of research testing the validity of AI-based methodologies for wider application across urban planning and the social sciences. Among the team’s conclusions, however, is that the research will always require collaboration between AI and humans.
In effect, AI needs to be steered by human sensibilities and knowledge, says Yin. “LLMs can provide incorrect answers if we’re not asking the right questions. So, we need to learn how to respond effectively. This technology needs human oversight. AI will never replace us; it’s a tool that can help us explore and understand data more effectively.”
Yin is in the final stages of an additional LLM study on public perception of the recently legalized cannabis distribution centers in New York State. Her comparative scan of social media and traditional media has so far revealed stark discrepancies in sentiment – comments on Twitter lean more positive while, in the case of the New York Times, the tone is more anti-cannabis.
“This kind of intelligence can change the policymaking landscape,” says Yin. “We’re building models that can help inform the way decisions are made about policies that shape our social and built environment.”
The technology is also changing the education of future planners, says Yin, who was invited by the Association of Collegiate School of Planning to participate in its AITask force. At UB, Yin has for years taught core courses in Geographic Information Systems for the Master of Urban Planning and BA in Environmental Design. She says AI can potentially transform how GIS is used in planning, at once enabling more complex and large-scale analyses and widening access to the sophisticated geospatial software.
“We have had this big black box around technology in the planning profession. Traditionally, students in planning have a steep learning curve with GIS and other technologies. AI will smooth this curve and open new opportunities for learning and scholarship.”
Graduates of UB’s PhD in urban and regional planning, founded in 2013, are also taking this expertise farther into the field. Yin herself has chaired the dissertation committees of three doctoral students, who today are applying urban analytics to such questions as inclusive transportation, disaster response, property abandonment and the revitalization of historically disinvested neighborhoods.
While ethical concerns remain around AI and urban planning – as they do in every discipline – Yin says the prevailing sentiment in the field is one of optimism.
“The potential of technology in planning, in some ways, has always been there,” says Yin. “The field of planning needs to catch up. Now, with AI, we can step into waters we’ve never touched before.”