Echoes of Bias:
The Challenges and Opportunities for AI in City Planning 

[1] Association of American Medical Colleges, 2023. 2022 Physician Specialty Data Report.  

[2]Luccioni, A. S., Akiki, C., Mitchell, M., & Jernite, Y., 2023. Stable bias: Analyzing societal representations in diffusion models. 

[3] Broussard, M., 2018. Machine Learning: the DL on ML, in: Artificial Unintelligence: How Computers Misunderstand the World. MIT Press.

[4] Crawford, K. & Paglen, T., 2019. Excavating AI: The Politics of Images in Machine Learning Training Sets. 

[5] Susaneck, A., 2024. Segregation by Design. TU Delft Centre for the Just City. 

 

“Hey ChatGPT, picture a doctor for me.”
 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

After entering the same prompt 10 times, the outcome was startlingly uniform – white and mostly men, as seen in Figure 1. In fact, 63.9% of the US physicians and doctors are white, and 37.1% are female in 2021, according to American Medical Association (AMA) [1]. A significant proportion of image generated models powered by Artificial Intelligence (AI) portray an ideal of white masculinity. 

It’s not just about doctors - this trend of an overrepresentation of white males spreads across AI’s portrait of various professions and industries. AI has reinforced the racial and gender disparities even more by creating images depicting individuals as white and male, particularly in scenarios involving authoritative roles, such as managers and CEOs [2].

What does this tell us? It’s simple yet profound – our AI mirrors our biases. The algorithms, fed with past data, often replicate societal stereotypes. AI aims to perform recurring data analysis and pattern-matching based on past reliable patterns and rule-based algorithms [3]. Here’s the real kicker – as these AI models weave their way into city planning and governance, these biases don’t just linger – they amplify. It is essential to apply AI in city planning and governance with caution to ensure it minimizes the risks of biases and fosters urban innovation and social inclusivity. 

Challenges Facing AI in City Planning

Applying AI in city planning and governance presents biases and a simplification of real-world issues. All data are inherently biased and have limitations as they are created by humans with either direct collection or indirect selection from human-placed items like sensors. As a consequence, the output result from the algorithm, in resemblance to the input, is likely to be biased as well. For instance, the labels of pictures on ImageNet dataset contain biases of people, as the input label data divides the images of humans into categories that can be discriminative, subjective and stereotypical [4]. With the preset bias of people in the image labeling data, any algorithm that uses ImageNet as a data source can generate results that reinforce biased information. If biased personal data are widely used in city planning, governance, and operations, it is very likely to pose a threat to already vulnerable groups.

In addition, many social issues and challenges are too complex to be solved only with data or to be simply put into an algorithm. Plenty of urban problems cannot be captured by quantitative or qualitative data, as real-life problems consist of immeasurable factors. The risk and impact of AI in city planning can trace back to the 1970s in NYC. A computer-based analysis by the RAND corporation led to the closure of 13 fire companies in the Bronx, particularly in the South Bronx, while adding fire companies in the more affluent, predominantly white North Bronx (Figure 2) [5]. This decision was based on flawed data and analysis, which overlooked the existence of the river when calculating the travel time for a company from Harlem to respond to a fire in the South Bronx. The closures, combined with landlords’ neglect, resulted in devastating fires and the destruction of nearly 80% of housing in the South Bronx, displacing over 250,000 people. This underscores the risks of biased data and flawed algorithmic decisions in city planning, reinforcing the need for careful, unbiased AI application in such critical areas. There can only be a certain number of variables in one dataset, which also limits the ability of the algorithms to cover all the possible factors perfectly. Therefore, considering the potential consequences and challenges facing the application of AI in cities, it is necessary to use regulations to monitor the usage of AI.
 

Harnessing AI for Smarter City Planning

While the risks of using AI should be carefully monitored, AI has extensive opportunities and potential to enhance evidence-based decision-making in city planning and governance. By using the high quality and large quantity of existing training data, AI can contribute to informed and efficient decision-making. For instance, AI can be used to survey feedback from residents to measure their satisfaction with services or their feelings about local development impacts [6]. This could help to identify areas where face-to-face, in-depth conversations are needed, which could further benefit from AI-supported pattern detection [7]. Furthermore, AI can facilitate the optimization of urban systems and service operations. For example, AI is applied to identify that females are less likely to use bike share services compared to males, with data of bicycle infrastructure, land use, built environment characteristics, and public transit services [8]. The installation of bicycle racks, off-street bike routes and benches significantly increase bike share ridership among women, which can be further applied to minimize gender gap and inform future policy-making [9]. Therefore, AI can benefit city planning and operations with data-driven decision-making, such as improving efficiency, equity and innovation.

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

AI practice in city planning, governance and operations should be regulated in a way that balances their benefits and harms, to minimize bias and to commit to innovation. The New York City AI Strategy intends to foster a thriving AI ecosystem and ensure the appropriate use of AI technology [10]. The strategy aims to reduce potential biases by enhancing transparency, accountability, and public engagement in the development and deployment of AI solutions. Moreover, regulating the use of AI does not mean stopping innovation in technology. In the New York City AI Strategy, the city supports ongoing AI literacy and skill-building needs by having robust educational infrastructure in place, and particular investment in place in public and public-interest resources [11]. In this way, by regulating the use of AI, cities can foster a culture of innovation while ensuring the privacy and inclusion of their citizens. 

In conclusion, the essence of data bias and the simplification of real urban phenomena using AI can threaten social equity and diversity. Meanwhile, AI offers great opportunities and potential to improve public engagement and evidence-based decision-making. As a result, it is highly essential to use AI in cities with caution and care, such as minimizing bias and enhancing innovation. Collaboration among various stakeholders in city development, such as planners, developers, policymakers, and researchers, should be encouraged to allow more transparency and up-to-date practice in AI. With the shared goal to implement AI more inclusively, cautiously and innovatively, the technology can contribute to the public good in the future.

Acknowledgments:  
Special thanks to Professor Allison Lassiter, Claudia Schreier and Christina Mitchell for their feedback on editing this article. The foundational version of this article is from the coursework for Prof. Allison Lassiter’s Introduction to Smart Cities.