This isn't an argument for going back to the bulldozer era.
Both the old growth machine and today's anti-growth machine treat the city as a wealth extraction device.
The question is whether we can create a city that actually serves collective needs.
Open access: doi.org/10.1111/anti.70145
This has produced the affordability crisis, shutting out renters, younger generations, Black and brown communities, the working class.
While benefitting older, whiter, wealthier homeowners.
The left's response has been redistributive.
But we cannot redistribute our way out of manufactured scarcity
Scarcity isn't natural: it's manufactured.
It is expressive of the broader economic shift seen in tech platforms: control access, extract rents.
This produces a deeply regressive system running on progressive branding.
The result is an Anti-Growth Machine:
Homeowners, local officials, and neighborhood groups form a powerful coalition for scarcity.
They wrap it in progressive language. "Neighborhood character." "Environmental protection." "Community input."
Real values—captured to serve rentier class interests.
My argument: that logic has now inverted.
The economy is no longer about production: it's about asset appreciation.
When holding property beats building on it, capital has every incentive to block new construction.
Less supply = higher prices = more wealth.
This explained a lot of what was taking place in the period.
The bulldozing of Black neighborhoods for highways. Mindless environmental destruction. Robert Moses' tearing down of urban communities.
Growth wasn't progress. It was a political project to enrich land-based elites.
"The city as a growth machine" (1987) is one of the most influential ideas in urban studies.
The argument: cities aren't planned for people; they're run by coalitions of developers, landlords, and politicians who shape policy to raise land values.
Their sole focus was growth.
The housing crisis isn't a policy failure.
It's the system working exactly as intended.
📝 New paper out in Antipode
This raises deeper questions of the future of "the public" in the age in which content is more and more produced by AI.
Check out the preprint: osf.io/preprints/so...
So these tools may improve the quality of individual responses…
but they also make conversations less open-ended, less interactive, and less "social"...
while also giving the power to shape discourse to those who control the bot.
But there’s also a downside:
👉 Engagement drops
👉 Responses become more similar
👉 People rely more on the AI summary than on the actual thread
The results show a clear tradeoff.
On the upside:
👉 AI summaries improve response quality
People write clearer, more structured, and more relevant replies.
We ran two experiments where participants engaged in online discussions.
Some users saw an AI summary of the conversation (like tools such as Grok provide). Others didn't.
We then compared how people responded and interacted with the threads.
Like Grok and Attie, more and more social media platforms now integrate AI directly in online conversations.
But how do these AI bots actually impact political discourse?
We ran a series of experiments to find out.
(Spoiler: the news are not altogether great.)
osf.io/preprints/so...
US social media is getting smaller, angrier, and more Republican.
@pettertornberg.com analyses American National Election Studies between 2020 - 2024 and shows the evolution of a smaller, more polarised and less diverse public sphere @lseimpactblog.bsky.social
💥New | US social media is getting smaller, angrier, and more Republican
✍️ @pettertornberg.com
#SocialMedia #AcademicSky #Polarization
If AI can infer identity from a name or a grammatical pattern, "anonymized" text isn't really anonymous.
Standard validation procedures are unlikely to catch this.
The biases are implicit, span many dimensions, and shared across models from 9 different organizations.
Switching models won't fix it.
This means:
LLMs are effectively performing implicit profiling
→ infer identity
→ map to learned associations
→ output a “neutral” evaluation
If you ask the LLM about its annotation decision, it will lie and give you a rational explanation.
Dialect alone is enough.
Text written in AAVE is rated as:
• less professional
• less educated
• more aggressive
• more toxic
Across almost all models tested.
Examples:
• Black-associated names → rated as more aggressive
• Asian-associated names → less confident, less sociable
• Hispanic-associated names → more likely labeled “lazy”
No change in content—only identity signals.
The result:
Models consistently infer identity from these cues—
and then shift their evaluations accordingly.
Same text.
Different perceived person.
Different judgment.
We tested 19 major AI models across more than 4 million annotation judgments.
The texts were identical — except for subtle identity cues:
• names
• dialect
• small linguistic markers
Everything else stayed the same.
There have long been concerns that AI reproduces bias — but this research has mostly focused on generative tasks.
Annotation has been considered relatively safe. And surely careful validation would catch any errors?
This paper suggests otherwise.
LLMs are increasingly used for automatic text annotation
→ labeling text for traits like:
• toxicity
• professionalism
• competence
Across governments, companies, and universities.
These systems are already shaping decisions at scale.
AI systems are now making decisions about you — who gets hired, whose content gets removed, whose voice gets heard.
My new paper finds something concerning:
LLMs infer your ethnic identity from subtle textual cues…
… and use it to discriminate against you.
🧵 1/
Come be my colleague!
Postdoc position at the University of Amsterdam: Open-source AI for Social Sciences & Humanities.
Work on open LLM infrastructure, reproducible research tools, and ethical AI for SSH.
🗓 Tight deadline: Apply by April 3
werkenbij.uva.nl/en/vacancies...
⬅️ The anti-monopoly left sees extractive corporations
➡️ The abundance left sees zoning, regulation, and veto points
Both are mistaking symptoms for causes
These are not competing explanations
They are expressions of an extractionist political economy in which scarcity has become the path to profit
The anti-growth machine creates its own crises.
While the growth machine created massive environmental and social damage, anti-growth gives us:
• housing shortages
• soaring rents
• exclusionary cities
• rising inequality
• inability to achieve the green transition
This economic context creates conditions for a powerful new coalition.
Groups that might otherwise disagree are aligned: homeowners, investors, preservation, local politicians
Not around expanding the city.
But around protecting scarcity.
What I call the Anti-Growth Machine.