Talking Headways Podcast: Planning for Generative AI
During this Talking Headways Podcast Dr. Michael Flaxman and I talk about generative artificial intelligence and its intersections with planning practice. We chat about what generative AI is, high-level descriptions how it is trained (tried to avoid the word transformers), and some of the ways it could be used or misused in a planning and transportation context. Michael and I have been collaborating on this and related topics for the last 8 years since meeting at an Esri Geodesign summit. We have been working to push the field forward together ever since.
Key links to share from the discussion are below:
Augmented: Planners in an Era of Generative AI - APA
Open Letter on Generative AI & Ethical Planning - APA Technology Division
Datasets for Datasheets - Cornell Arxiv
Automating Zoning Data Collection - Urban Institute
Bias in, Bias out - Yale Law Journal
Can Language Models Be Too Big? - ACM Digital Library
Supplemental Thoughts About the Podcast
This is a topic where there is a lot to say with only so much time to say it. Michael and I recognize it is early days, but we both see the potential the technology has for scenario planning, making massive reams of text into useable data, and the possibility of merging our understanding of the world today with how our text heavy zoning codes and plans articulate our futures. We also recognize the risk as we clearly identified in the links we provide in the show notes, and the papers we identified in the discussion. Pretty heavy hitting articles by Bloomberg identified how generative AI’s biases could exacerbate stereotypes, and even the conceptual problems of value lock for a profession whose mission to think about improving outcomes and confront tradeoffs as its relates to community futures.
There are already examples of planning applications of AI (generative or not) that give me pause. I think my key point for planners would be to think clearly about problem definitions and the analysis frame you might be inheriting when you are attempting to use prediction to approach problems in planning.
Are you cementing the intellectual frames of the past through automation?
Is the tool replacing critical thinking and good judgement?
Is the data going into a model representative of your community?
Who does a model benefit or harm in its application?
What voices were in the room to define the “need” for AI to begin with?
Do you have the data, process, and funding to evaluate this application in the future?
These and other questions related to the strengths, weaknesses, opportunities, and threats of generative AI are ones to consider carefully.
