Artificial Intelligence and Planning Practice
Cite as: Wasserman, D. Flaxman, M. (2022). Artificial Intelligence and Planning Practice. PAS Memo 111. American Planning Association. Retrieved from https://www.planning.org/pas/memo/111/artificial-intelligence-and-planning-practice/
Download at: Planning Advisory Service Memo 111
Contact me if you want to request a copy.
Abstract
The term "artificial intelligence" (AI) conjures images of autonomous vehicles maneuvering through streets, smartphone assistants that answer your questions, or androids exploring final frontiers.
At a basic level, however, AI can be understood as the multidisciplinary endeavor to approximate human reasoning with computation. For planners, it represents an emerging toolbox that enables a range of new capabilities. Whether AI primarily benefits entire communities or narrow interests, though, depends on planners' abilities to engage with the challenges and opportunities surrounding its civic applications. Naively applied, these technologies can automate discrimination, create unaccountable processes, and create a false certainty about what the future holds.
This PAS Memo intends to equip planners with an understanding of AI concepts and their potential uses for practice. And because planners have a responsibility to understand the implications of the technologies they choose to deploy and help to ensure that those technologies are used responsibly, it discusses important considerations regarding AI applications and their roles in larger trends connected to digital governance and civic data in planning.
More Than Just Machine Learning
This memo discusses up front that there is more to consider for planners than just the applications and considerations for machine learning in planning practice. In many cases, digital governance requires us to think about how we manage, use, and visualize data. The memo reviews civic analytics problem typologies that can assist with a broad spectrum of challenges within operations, policy, and planning for cities and regions. In many cases, we are still looking to walk before we run.
Excerpt from Artificial Intelligence & Planning Practice
CONSIDERATIONS FOR AI AND PLANNING
AI opens new possibilities for planning practice, but it requires awareness of the technological foundations underpinning planning methods and practice and acknowledgement of the risks associated with AI and related emerging technologies . Responsible and effective applications of AI in urban planning practice will depend on planners' understanding of these issues.
…
When Past Should Not Be Prologue
Planning requires thinking about how policies and public investments shape potential pathways for community futures (Wright 2019). Statistics and machine learning will develop predictive models for the future based on past data, and by so doing the insights created from them use the past as prologue (ITF 2019; Mayson 2018). There are two major concerns, however, regarding the application of these models:
Cementing past mistakes. Basing decisions on predictions from historical data is likely to repeat and reinforce the outcomes of the past (Mayson 2018). This can often follow the use of metrics or data that are convenient or at hand, with outputs reinforcing historic values or creating unintended outcomes (Crawford 2021). Predictive models can mirror how we have historically addressed problems rather than reflecting the lens we bring to them now (Mayson 2018). In other words, we risk automating processes that were problematic to begin with because they will “inherit” the analytical frame of the system they originate from.
Managing change. Predictions based solely on historical data will not adapt to changing conditions. For example, a planning challenge likely to define the 21st century will be planning for a changing climate. As of 2019, carbon dioxide concentrations not seen for two million years are clear examples of how pure machine-learning models may not provide as much value in an environment where conditions change (IPCC 2021).
For these reasons, care should be taken when these algorithms are applied in situations with high degrees of uncertainty or unprecedented circumstances, or where they are likely to reinforce undesirable historical outcomes.
