Could Scranton Be The Next San Jose?
How AI Might Change American Geography.
The Intelligent Friend - The newsletter that explores how AI changes our daily lives, through scientific research.
The producers made an unusual move. Despite the palpable dimension of serenity, it was not charismatic enough. The city did not have the character to be iconic. The place that millions could become attached to. It didn't have the uniqueness of New York or the glamour of Los Angeles. On the surface, there wasn't anything really exciting to be found either.
Yet, Scranton has struck and sunk the hearts of viewers. Michael, Jim, Pam, Dwight and all the others work in the small town in the state of Pennsylvania, where Dunder Mifflin tries to survive the various crises that a company like the paper one can face. As of today, it has a population - decreasing compared to last year - of about 76 thousand people1. It has a long history tied to the coal and energy industry, so much so that it is also called the "Electric City". It is a small city compared to big urban centers - it is the sixth largest in Pennsylvania - that is very different from many other realities, near and far.
If you try to draw a line slightly sloping to the left and meet the corresponding western shore of the United States, you will discover that at that point of the line you will find a city that is truly separate from the home of The Office. It is about 2,840 miles that separate Scranton from the famous place you will see on the map: San Jose.
These cities probably could not be more different. San Jose has a population more than ten times that of Scranton (about 970,000 inhabitants2). Instead of being the Electric City, it has been named - and probably still considered today - the “Capital of Silicon Valley”. Inside, you can find the headquarters of companies that you find every day to work or buy something: eBay, Cisco, PayPal. While Scranton is a city mostly of Americans, San Jose has a share of almost 40% of people born abroad. Among all, Vietnam certainly stands out: in the Northern Californian city there is in fact the largest community of Vietnamese outside of their home state3.
In short, these two cities are profoundly different. Detached in economic terms. Culturally. Demographically. And comparing them, it would seem obvious to say that recent times could favor San Jose. The city has everything it needs to flourish further. Scranton, at first glance, could appear as decadent. As defending the economy against the unstoppable advance of technology. Above all, of AI. And yet, this is not, necessarily, the case. Even Scranton has its chances. As do several small towns. And not despite AI. But precisely because of it.
A shock with unexpected effects
Generative AI tools could completely replace almost 15% of all tasks performed by human workers4. AI could therefore have major economic effects, both positive and negative. However, the effects are not only the immediate ones that one might imagine. There are crucial downstream effects (i.e. secondary, indirect, cascading). Among these, according to Scott Abrahams and Frank Levy5, there are consequences that concern Scranton. But Scranton is only one of the many cities involved: Rochester, Savannah, Toledo. All potentially protagonists.
These urban centers could in fact become the destination of migrations of college graduates due to the shock created by AI. This is one of the three central predictions that economists advance in their work (from which this issue takes inspiration for the title, and taken up in a recent article in the New York Times).
But why should this happen? And above all, could these small towns benefit from AI and be targets for travel, unlike metropolises like San Jose, New York or Chicago?
A significant parallel
To understand this, the authors started by considering AI as a shock to the economy. This conception allowed a parallel with the post-80s decline of US manufacturing employment. During that period, the collapse of manufacturing jobs displaced millions of workers, particularly in the South, Midwest and Northeast. An important unit of their analysis are the Commuting Zones (CZ), which allow a visualization (literally) of the effects analyzed. As you can see in the image, in red there are the areas most strongly affected by the crisis of the 80s. These are areas that, according to the authors, found it particularly difficult to "reinvent themselves".
If you put yourself in the situation for a moment - or if you have lived it - you might find it not hard to believe that one of the effects of this crisis has been the migration of college graduates.
This category in particular has seen a strong desire to emigrate to the cities over the years. This, in turn, has also caused cities to have an intense difficulty in adapting. The authors summarize the concept very effectively:
Places - local economies - were not adapting to the manufacturing shock as much as the people who lived there were adapting, both through acquiring more education and through migration.
This shift was also reflected in political ideologies: the transition in political affiliation was gradual, with college graduates shifting from the Republican to the Democratic Party, while non-graduates moved from the Democratic to the Republican Party.
Both transitions were influenced by cultural and economic factors, yet both were tied to the impact of the manufacturing downturn.
The LLMs-driven Migration
These categories of effects are the starting point for scholars to understand how the advent of AI - specifically the authors refer to Large Language Models or "LLMs" - could actually impact the geography and demography of states. Concretely, the reference measure is the occupation-level exposure to LLMs (Elondou et al., 2023). This represents, as the authors clearly explain, "which occupations are most likely to experience some form of direct labor demand impact from the shock [of AI]". In parallel to what they did for the crisis of the 80s, they then built a map of the areas potentially most impacted by the arrival of LLMs from an employment point of view.
Now, for big fans of the game of finding the differences, what do you notice different compared to the first map I showed you? As you will notice, the highlighted areas are very different. They are mainly large cities, like Washington, but also Boulder and our dear San Jose.
Assuming that local production structures remain largely unchanged over time, these regions are likely to see the segment of college graduates facing diminishing opportunities due to a lack of the specific skills needed to leverage AI tools effectively.
In essence, it's a reversal. If college graduates have left small cities to move to metropolises flourishing with development opportunities, now advanced and explosive technological developments leads them to reconsider previously underestimated urban centers.
Furthermore, to identify potential destinations of interest for the “migration”, the authors focused on mid-sized metropolitan areas with specific requirements:
A strong presence of college graduates;
A healthy job market;
A relatively affordable housing.
These cities need to strike a balance: they could be large enough to offer opportunities for skilled workers, but not so expensive or overexposed to AI-driven disruptions that they would mirror the challenges of major urban hubs. By filtering for these criteria, they pinpointed places that could attract professionals seeking stability and new opportunities in a changing economy. You can see below the result of this analysis, which represents the insight that really fascinated me about this work6.
Overall, the South and Midwest appear to be able to benefit from the downstream effects of LLMs on the economy.
If we would like to further narrow our eyes on the appetizing destinations, also looking at the areas that are actually already growing, Augusta, Savannah, Greenville, Chattanooga, Oklahoma City, Columbia, and Houston are the ones to mark on your notebook, according to the authors.
So, if AI could on the one hand lead to a disruptive technological development in our daily lives, some people could actually decide to relocate their life. To build one in line with the expectations that work and education might bring. To move, for example, from Washington to Toledo. From New York to Rochester.
From San Jose to Scranton.
Takeaways 📮
A changing geography. AI could influence migration trends, driving college graduates from large cities to smaller urban centers.
What requirements? Mid-sized cities with affordable housing, strong job markets, and educated populations could benefit from AI-driven economic shifts.
What areas? The South and Midwest are poised to gain more from these dynamics.
An effort to highlight 📣
For 2025, I will try to feature a nonprofit or socially engaged organization in each issue of The Intelligent Friend. I discussed this in this Note.
As the first organization, I am happy to spread the word about what First Therapy does. It is an organization that aims to positively impact mental health. It was also founded by a Substack writer, Vaaibhav, and is part of a larger effort to raise awareness of mental health issues and related solutions.
In particular, First Therapy "breaks the barriers of costly, commitment-based, and inaccessible therapy options by providing free therapy sessions".
If you know of a nonprofit or organization that I would like to feature, DM me, comment, or reply to this email. I will be happy to share their efforts in an upcoming issue of The Intelligent Friend.
Other enlightening references
Acemoglu, D., Autor, D., & Johnson, S. (2023). Can we Have Pro-Worker AI?. CEPR Policy Insight, 123.
Acemoglu, D., & Johnson, S. (2024). Learning from Ricardo and Thompson: Machinery and labor in the early industrial revolution and in the age of artificial intelligence. Annual Review of Economics, 16(1), 597-621.
Acemoglu, D., Kong, F., & Restrepo, P. (2024). Tasks At Work: Comparative Advantage, Technology and Labor Demand.
Alam, M. F., Lentsch, A., Yu, N., Barmack, S., Kim, S., Acemoglu, D., ... & Ahmed, F. (2024). From Automation to Augmentation: Redefining Engineering Design and Manufacturing in the Age of NextGen-AI.
Autor, D. (2024). Applying AI to rebuild middle class jobs (No. w32140). National Bureau of Economic Research.
Autor, D., Patterson, C., & Van Reenen, J. (2023). Local and national concentration trends in jobs and sales: The role of structural transformation (No. w31130). National Bureau of Economic Research.
Autor, D., Salomons, A., & Seegmiller, B. (2023). Patenting with the stars: Where are technology leaders leading the labor market?.
Capraro, V., Lentsch, A., Acemoglu, D., Akgun, S., Akhmedova, A., Bilancini, E., ... & Viale, R. (2024). The impact of generative artificial intelligence on socioeconomic inequalities and policy making. PNAS nexus, 3(6).
Thank you for reading this issue of The Intelligent Friend and/or for subscribing. The relationships between humans and AI are a crucial topic and I am glad to be able to talk about it having you as a reader.
Has a friend of yours sent you this newsletter or are you not subscribed yet? You can subscribe here.
Surprise someone who deserves a gift or who you think would be interested in this newsletter. Share this post with your friend or colleague.
https://www.census.gov/quickfacts/scrantoncitypennsylvania
https://www.census.gov/quickfacts/fact/table/sanjosecitycalifornia/PST045223
https://www.sanjose.org/pdf/san-jose-at-a-glance
Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs: An early look at the labor market impact potential of large language models. arXiv preprint arXiv:2303.10130 .
Abrahams, S., & Levy, F. S. (2024). Could Savannah be the next San Jose? The Downstream Effects of Large Language Models. The Downstream Effects of Large Language Models (June 23, 2024). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4874104
In their working paper the authors analyze two other potential effects of primary importance: the impact of LLMs on educational attainment and on political opinions. In this issue I have focused on the second "avenue" related to geography. But the work is illuminating in all its predictions.