What If AI Designed All New Cities?

The blueprint is perfect. Every street width optimized for pedestrian flow and emergency vehicle access. Every green space placed for maximum psychological benefit per resident. Every building orientation calculated for passive solar gain. Schools positioned within 400 meters of every family home. The commute from any residential zone to any employment district takes under 20 minutes by public transit. The stormwater system handles 500-year flood events. The city will house 3 million people and produce net-zero carbon emissions from day one.

Thought experiment. This article explores a hypothetical scenario for educational and speculative purposes. The events described are not predictions and should not be interpreted as warnings, advocacy, or instructions.

It has never been lived in. And that might be the problem.

Key takeaways

• AI urban planning tools already exist and are being used for traffic modeling, zoning optimization, and infrastructure placement — the question is how much authority they should hold.
• Optimized cities can be efficient without being livable — historical precedents from planned cities suggest that what gets measured isn't always what matters.
• AI could solve real urban failures — housing shortages, traffic congestion, inequitable service distribution — that human political processes have failed to address.
• Cultural identity and organic urban character are hard to specify as optimization targets, and cities without them tend to feel wrong to the people who live in them.

The optimistic scenario

Modern cities are the product of accumulated historical decisions, many of them terrible. American cities were often designed around the car in the 1950s and 1960s, a choice that created suburbs requiring automobiles for basic errands, contributed to pollution, fragmented communities, and made public transit economically marginal. European city centers preserved for historical reasons often have inadequate housing density, pushing the poor to distant peripheries with poor services.

An AI that could design a city from scratch, unconstrained by political deals, historical accidents, and the preferences of incumbents, might make dramatically better decisions. It could place housing near jobs, schools near families, parks near pollution sources. It could design for climate — maximizing shade in Phoenix, maximizing sunlight in Stockholm. It could integrate utility infrastructure intelligently so water pipes, power lines, and transit corridors don't create maintenance conflicts.

Cities in rapidly urbanizing regions — sub-Saharan Africa, South and Southeast Asia — will need to house hundreds of millions of new urban residents over the coming decades. Much of this housing will be built fast, on limited budgets, with limited expertise. AI design assistance could help ensure that rapid urbanization doesn't reproduce the informal settlements and infrastructure failures that have characterized previous waves of urbanization.

There is also an equity argument. Human urban planners, working in politically charged environments, often design cities that serve powerful interests. Highway projects routinely destroyed Black neighborhoods in American cities in the mid-20th century because the people affected lacked political power to resist. An AI system with equity metrics built into its objective function might resist those political pressures — or at least make the tradeoffs explicit.

The realistic complications

The history of cities designed entirely from scratch by planners with complete authority is not encouraging.

Brasília, built in the 1950s as a planned capital, is architecturally stunning and practically frustrating. Its rigid separation of zones — designed by Oscar Niemeyer and Lúcio Costa on rationalist planning principles — means that walking anywhere meaningful is nearly impossible. It was designed for cars. The spontaneous street life, mixed-use vibrancy, and human-scale neighborhood character that makes cities pleasant emerged instead in informal settlements outside the planned zones.

Chandigarh in India, Canberra in Australia, Naypyidaw in Myanmar — planned capitals almost universally share the same quality of feeling somehow wrong to be in. They are functional in narrow senses and alienating in ways that are difficult to specify precisely but are immediately apparent to anyone who walks their streets.

The problem is that cities are not just logistical systems — they are cultural artifacts. The slightly irregular street grid of an old European city center, the chaotic mix of residential and commercial uses in a Vietnamese neighborhood, the unexpected park created when an old factory was demolished — these things were never designed. They emerged from human decisions, conflicts, compromises, and accidents over generations. No objective function can specify the value of an unexpected alley that becomes a night market.

AI systems optimize what they can measure. They cannot easily measure the feeling of belonging, the vitality of an informal economy, the identity that comes from sharing particular streets with generations of predecessors.

The unintended consequences

An AI-designed city might be so thoroughly optimized that it becomes rigid. Cities need to adapt — to demographic shifts, to economic changes, to cultural evolution, to technology change. A city whose every element was precisely calibrated for 2026 conditions may fit 2046 conditions poorly, and because its design is integrated and deliberate, modifying it may be harder than modifying the organic mess of an unplanned city.

There is also a social sorting risk. If AI city design optimizes measurable welfare metrics — commute times, green space access, school quality — it may implicitly sort people by characteristics correlated with those metrics. Cities that optimize purely for efficiency may become deeply exclusionary of the poor, the elderly, the disabled, and anyone whose needs don't fit the dominant model.

The precedent question matters enormously: whose data trains the AI? Urban planning AI trained primarily on data from wealthy Western cities will embed those cities' assumptions about what good design looks like. The needs of a household in Nairobi, where most of daily life happens in informal markets and extended family networks, are different from those of a household in Singapore.

AI-designed cities might also be susceptible to unexpected cascading failures. A city where every system is tightly integrated and optimized creates tight coupling — when one element fails, effects propagate rapidly. The redundancy and inefficiency of organic cities may actually be a resilience feature.

What science actually says today

Urban planning AI is a legitimate and growing field. MIT's Senseable City Lab has developed AI tools that model pedestrian flow, optimize service placement, and simulate the effects of design changes on urban behavior. Singapore uses predictive modeling extensively in urban planning and has one of the world's most analytically managed urban environments.

Research from the Santa Fe Institute on cities as complex systems suggests that certain urban properties — innovation rates, patent production, creative output — scale superlinearly with population and appear to be products of density and mixing rather than planning. These emergent properties are extremely difficult to engineer deliberately.

A 2022 paper in Nature Computational Science demonstrated that AI systems could generate urban layouts significantly more efficient than human-designed ones on specific metrics — walkability, energy use, transit coverage. The same paper noted that none of the metrics fully captured livability, and that human evaluation consistently rated AI-generated layouts as less appealing than organic alternatives.

The emerging view among serious urban researchers is that AI should be a powerful tool in the hands of human planners — handling optimization of specific systems, modeling consequences of design choices, identifying inequities — while humans retain authority over the values embedded in design. Full AI authorship of cities is premature not because the AI can't optimize, but because we haven't figured out how to specify what we actually want from a city.

FAQ

Are any cities being seriously designed by AI today? Several master-planned new cities — including NEOM in Saudi Arabia and some new urban developments in China — are using AI tools for aspects of their design. But none has handed full design authority to an AI system; humans remain in charge of key decisions.

Would AI cities be more sustainable? Probably yes on measurable metrics — energy use, water consumption, transportation emissions. Whether they'd be sustainable in the deeper sense of adapting and thriving over generations is much less certain.

What would happen to informal neighborhoods in AI-designed cities? This is a key question. Informal settlements often emerge because formal planning fails to accommodate real human needs — especially for the poor. An AI that could not account for informal economic activity and extended-family housing patterns might design out the very features that make cities work for their most vulnerable residents.

Could AI fix the housing crisis? AI could likely optimize zoning, reduce planning approval times, and identify underutilized land — all of which could help with housing supply. But housing affordability is fundamentally a political and economic problem, and AI design alone cannot solve it.

Do people actually want to live in perfectly planned cities? The evidence from planned cities suggests people appreciate certain elements — good transit, reliable services, accessible parks — while often finding the overall experience sterile. People consistently choose older, messier, more organic urban environments when given a choice, even when the infrastructure is objectively worse.

---

Read next:

• What If AI Controlled All Financial Markets?
• What If AI Could Cure All Diseases?
• What If AI Could Predict Every Natural Disaster?