Can computers design cities?

Can computers be creative?...

Can computers be taught to design? What is the process of teaching computers to design? Can computers design cities?


We have started a startup, Vivacity, that wants to provide answers to these questions above. In the past few years, these questions about creativity, learning, and imagination have been raised by companies like DeepMind and OpenAI. DeepMind’s defeat by AlphaGo of Lee Sedol at the game of Go and advances by OpenAI on machines learning how to play video games have proven that the field of machine imagination is still in its infancy and is ripe for exploration. At Vivacity, we are interested in exploring problems around machine imagination. Our focus is not on imagination and games, but rather on how imagination is formed in the field of design and the built environment and how can it be applied to practical problems. 

But why focus on the built environment? 

With more than 50% of the population of Earth projected to live in urban environments by the year 2020, we believe that any advance on the field of machine imagination for city planning will benefit the quality of life of all citizens. 

World Health Organization's Urban Population Growth Chart

World Health Organization's Urban Population Growth Chart

At Vivacity, we have decided to focus particularly on problems related to city planning and explore the endless possibilities that exist in the fields of architecture, urban design, and real estate development. We believe this focus allows us to address questions around providing better housing conditions, figuring out how to best deliver affordable homes, and provide feedback for  ergonomic design of cities. For our team, providing design solutions for the 21st century comes with a sense of excitement for new technology and responsibility towards shaping a favorable outcome for future cities.

Right now Vivacity is helping to solve one of the biggest problems urbanists will face in the coming decades as we transition into a more interconnected world: the transition of cities to become smart cities. We have identified that a city’s zoning legal text will play a crucial role in this transition.

Length of NYC Zoning Text Over Time

Zoning text length over time has increased exponentially; posing a unique challenge to real estate developers, architects and lawyers. This graph was made by looking at the length of PDFs documents available at the NYC Planning Department website.

 

Today, many cities' zoning text is archaic, cumbersome and convoluted; and the inability to interpret and understand this text has slowed governments, stalled developers, challenged architects and compromised the ability of cities to meet citizens needs. As cities transition to become “smart”, or data enriched, they will need to address head-on the challenges posed by old zoning texts. 

Why is that the case? Couldn’t a city just discard the legal text and start from zero? Well, the answer is no. 

The main reason is that any city’s zoning text will still be around us probably for the next 50 years. A city’s zoning text serves as a connector and mediator between the different stakeholders that shape any given city. City councilmen, real estate developers, urban planners, architects, community activists, all use zoning text as the legal text to debate, discuss and resolve problems around housing, transportation, and economic activity. But given the rapid pace of change and growth of economic activity, this discussions cannot be carried out with the cumbersome tools of the 20th century. Often there is disagreement over the interpretation of the zoning text; or even worse, the direct economic impact the development allowed by the zoning allows is not clearly understood. All of this discussions happen with the use of archaic tools for interpreting zoning. This needs to change if we want to guarantee a better standard of living.

Tools for Smart Cities


At Vivacity, we are building new tools for the smart cities of the 21st Century. For us, a smart city is not simply a city of the Internet of Things, or a City built from the internet up; a smart city is a city that has the tools that directly impact the discussion around how cities will be imagined, planned and built in the future.

Building a SimCity Doppelgänger

simcity-2000.jpeg

Cities are built with different overlapping layers, from physical infrastructures like subways and sewer systems and the skyscrapers that sit on top of them, to information systems that include geographic information, political demarcations, and statistical profiles. Any modern city exists on top of all these layers. All these layers often make it difficult for a local or big real estate developer to build on a given plot of land.

We are currently building two layers that will help cities transition into becoming smart cities. We are building a:

a) Machine learning layers for understanding how policies (in this case zoning) affects the built environment.

NYC Zoning Text: 4,000 pages long

NYC Zoning Text: 4,000 pages long

A network segmentation of the NYC Zoning Text

A network segmentation of the NYC Zoning Text

b) Machine learning layer that is capable of building a city’s digital 3D doppelgänger based on the As-Of-Right capacity the zoning text allows.

A machine learning how to solve zoning

We think these two layers are crucial to help solve problems around housing delivery; especially in trending real estate markets with a complicated regulatory and community engagement framework. In cities like New York, Los Angeles, Boston, Chicago and the Bay Area we are already seeing how these complicated and interconnected problems can stall housing delivery to a growing population. This has lead to the rising cost of owning or renting a house in metropolitan areas like New York City or the Bay Area.

One of the main reasons the process of acquiring city permits for developments takes so long is that a scalable solution to interpreting a zoning text and converting that into a design of a building does not exist.

Interpreting legal text to deduce what is possible is not a simple problem to solve.

Lawyers become experts at interpreting zoning text after performing the same repetitive task for many years for their clients. Hence, for every new property a lawyer is asked to interpret, he accesses his trained memory (aka his neural network) of many years to give the developer a satisfactory answer based on years of interpretation.

If his interpretation as a legal text aligns with the city’s interpretation of the text, the text proceeds to become the regulatory framework of what can be built in 3D. This problem is compounded with the difficulty of establishing if a building is buildable from a market perspective.

In similar fashion to the lawyer, after years of experience of construction, architects and developers take the legal interpretation of the zoning text and design buildings that fit both the city’s 3D regulatory frame and the financial constrictions. A building becomes buildable only when the interpretation of the zoning text aligns with the market-driven design.

A different approach to zoning

A natural language processing solution for the zoning text would benefit both the planning offices and private developers.  A planning office looking to scale how a zoning text is interpreted can use an NLP approach to teach its staff the “ABC’s” of the zoning text it just rolled out.

A private developer, whether an individual or a real estate fund, could also use an NLP approach to quickly minimize the risk that comes from misinterpreting a legal text and hedge risk against the possibility of losing rights to the development too late in the development process.

The key to the NLP approach for zoning text is summarizing a zoning text without losing the inferred meaning. This lossless conversion has proven to be difficult in the past, but advances in NLP frameworks such as Google’s SLING and spaCY are opening the door for lossless zoning text summary to become a reality.

A rule-driven approach to how the interpreted text becomes buildable would also immensely benefit both planning offices and private developers. Instead of manually generating a 3D doppelganger for each lot as a request comes in, this process can be automated using parametric design software. CityEngine, SketchUp, Autodesk BIM and  Rhinoceros3D all have the capabilities to be able to due rule-driven design.

The key to choosing the correct software is scale. It is one thing to do this for a single parcel or a city block, it is a very different thing to do it for a whole area, like Mid-Town Manhattan or the Tenderloin in San Francisco. Thankfully, programs like CityEngine with its CGA code and Rhinoceros with Grasshopper have the capabilities to do just that.

Example from CityEngine of Procedural Modelling with CGA code below.

Example from CityEngine of Procedural Modelling with CGA code below.

/**
 * File:    building_01.cga
 * Created: 02 Jun 2017 10:47:50 GMT
 * Author:  andi
 */
 
version "2017.0"

attr minheight = 10
attr maxheight = 30
attr floorheight = 3
attr windowwidth = 2

Lot --> extrude(rand(minheight,maxheight)) 
        Components

Components --> comp(f){top : Roof. | 
                       side : Facade}

Facade --> split(y){~floorheight : Floor}*

Floor --> split(x){~windowwidth : Window}*

Window --> i("modern_window.obj")

With tools like the one mentioned above, the process of navigating future options for communities and cities can become more streamlined and can help lower the friction to deliver housing and remove friction from having sustainable and community-oriented urban development.

Some cities like Boston and Boulder, Colorado have started to take that step towards digitalization. Their attitude is very promising and it shows that the realm of new technologies in the real estate and urban planning are just beginning to be explored.

You might be saying to yourself, this sounds a lot like SimCity. And you are right!

We would be lying if we did not say that ultimately what we want to achieve is a version of SimCity for all cities of the world.

simcity-2000-real-city.gif

What Next?

If you liked this post, be sure to subscribe to our blog! We will be posting our thoughts on this and many other subjects on a monthly basis. Also, don’t forget to visit our website and subscribe to our mainling list or try our FREE Property Platform. We are currently beta testing the two layers discussed in this article in the City of New York.