Connectivity of green and blue infrastructures: living veins for biodiverse and healthy cities

Imagining the future of nature

On a recent visit to Milan I got to see two shows put on by the Fondzione Prada, one at their space Osservatorio Milano and the other at Fondazione Prada itself.

The first show, put together by Trevor Paglen and Kate Crawford, assembled a number of images from the emergence of facial and voice recognition and artificial intelligence (AI) methods, by which automated detection programmes are meant to learn to correctly recognise individuals (see:
The images included many, many, many faces, vast rows and tables of them, smiling or frowning, lit from different angles, famous people taken from the internet, repeat offenders’ perp photos showing them grow older and more desperate, selfies stolen from Instagram. The need to recognise individuals is of course related mainly to law enforcement and security. The mere premise of some of the image databases is troubling, such as the images of Japanese women displaying the ‘five universal emotions’. Why would you need this dataset, other than to police emotion in public and private spaces? The image datasets with tags are also disturbing. The tags, which identify the content of the images, are a strange mix of the absurdly random or senseless, and the most basic and offensive stereotypes. What does it mean for authorities to interpret the world through the pronouncements of these most stupid of oracles?

The second show was curated by Wes Anderson and Juman Malouf, and consisted of a selection of objects and paintings from a history and natural history museum in Vienna ( The objects were placed within a kind of purpose-made mini museum with glass cases lined in velvet of carefully chosen colours. The rooms were thematic: paintings of children, green things, cases and containers, things made of wood, etc. It was charmingly arbitrary, visually beautiful and rich in suggestion.

Collections of interesting natural objects and beautiful artworks are how the powerful used to demonstrate the reaches of their power and the extent of their curiosity, the cosmopolitanism of their understanding. Today, power no longer consists of collections of singular exquisite objects with particular interpretations, but of huge datasets of variations on trashy images with low-quality descriptive labels.

Its not that in the past, wunderkammer collectors knew and ‘correctly’ understood everything, or that states did not seek to surveil and gather information before the existence of AI. But today, the object of unique interest, the type specimen, the rare item, the beautiful example, cannot speak to power: it has nothing to teach a machine learning program because the sample size is 1 and computers are too stupid to learn from sample sizes of 1. (More specifically, human collectors also do not learn from unique items in isolation, but are able to compare flexibly and creatively across things that are not standardized in such a way as to constitute a dataset).

Power in the AI paradigm is a function of the size of your training set, but also of its selection. As with any data analysis, a very large data set with a very strong selection bias will give you a clear signal: the stronger the bias and the bigger the sample size, the more likely the bias will be detected. People talk about improving the performance of AI surveillance systems—in the sense of removing stupid, harmful, and counterproductive biases—by creating “neutral” training sets. Of course, it is unclear what neutral means: does it represent reality as it really is (almost all samples or subsets of reality do not represent reality as it really is, as anyone who knows anything about statistics knows)? Does it represent reality as we think it ought to be? Even if we could imagine a truly neutral dataset with no biases incorporated into its selection, it is not clear to me that machine learning would be able to learn anything from this. Without selection biases to guide its pattern forming, it seems to me it would just produce nonsense. You could easily end up with meaningless but strong patterns like these:

Ecologists sometimes use a dataset to create reconstructed, imaginary datasets, which are variations on the real dataset. They then ask how many of the imaginary reconstructed data sets closely resemble the real one. From this they claim to assess how likely it is that the real dataset looks like it does, rather than looking like something else. If the real dataset is a sufficiently low-probability arrangement of its own data, then it is considered to be a good representation of reality. This is called Baysian statistics.

I have always thought that Baysian statistics is a fraud. Because the likelihood that a data distribution represents reality is based on millions of rearrangements of that same dataset, rather than established distributions such as parametric distributions, no two Baysian analysis results can ever be compared. (This doesn’t stop people from implicitly comparing them within the scientific literature, however.) It is as though if you wanted to measure the length of a rug, you made a thousand variations on rugs of the same weight (or some other arbitrary feature of the rug) and then saw that only 3 of them were the same size as the original rug: therefore, you conclude, the rug is indeed, in reality, the size of that rug. You then do the same thing with a second rug. In the end, both rugs are the size that they are, that is, the sizes of rugs. However, you have no way of knowing which rug is longer (all you know is that they are both the sizes of rugs), or whether either will fit in your hall. If you use a meter stick, you are not claiming that any rug is exactly one meter long, but you can find that one rug is 1.3 m long plus or minus some error, and the other 2.7 m long plus or minus the same error, and that your hall is 2 meters long plus or minus the same error.

Scientists love Baysian statistics because they allow you to do statistics, and thus make claims, about datasets that are incomparable to standard distributions like parametric distributions—the standards that act like metersticks and allow between-dataset comparisons and sense-making. Datasets that are incomparable may be datasets that were badly selected, and that incorporate many unknown biases rather than known methodological biases. They may be datasets that were well-selected, but that represent a reality that doesn’t fit well with the assumptions of standard distributions. Either way, rather than having a discussion about interpreting levels of uncertainty and error in standard statistical outputs, many researchers opt for Baysian statistics.

The artist Anna Ridler talks about how the outputs of AI resemble dreams: the training set is our waking experience of life, and our dreams are the brain reprocessing this and making new narratives and meanings with it. Her animated work The Fall of the House of Usher (1) ( is a visual example of this idea. Her work with tulips ( and ( also shows a machine-learning program recreating dynamic images of tulips blooming. The way they move is clearly not right, however, and indeed look like something from a strange dream.

Another very interesting artwork using AI is Terre Seconde (Second Earth), an installation by Grégory Chatonsky that uses AI to produce images of a duplicate Earth ( A large number of images of birds, for example, are used to produce a pattern for birds, the essence of birds, an imaginary bird that summarizes all the bird-imagery it was given. What he appears to have done, however, is to put together images selected to not necessarily be similar and strictly comparable (the selection bias is something else), resulting in 3D-printed structures that look marginally organic but certainly imaginary: a machine imagination. He also trained the machine learning on a database of human dreams, to produce machine-learned dreams. In the work, Chatonsky proposes a fictional machine that is carrying out this task of learning and reproducing life-forms and geological features, while asking itself what it is doing and why.

As I discussed in a previous post (, technologies of large-scale nature surveillance exist and are increasingly discussed as the solution to biodiversity monitoring and conservation. Autonomous machines that would automatically perform nature restoration have been proposed. We may not be far from developing the big-data datasets and the autonomous capacity to act on them, that would result in our living in a machine’s dream of nature. Perhaps, though, not one with morphing tulips and hallucinatory antediluvian organic structures. I am afraid the future of nature will be more Paglan-esque than Chatonsky or Ridler-esque. I fear we will live in a future with an absurd, trashy, stereotyped plan for how forests and wetlands, urban parks and wildlife corridors should be constructed. Maybe one where autonomous drones armed to detect and kill invasive cats also kill the last hidden population of Tasmanian tigers because they fit the ‘cat’ target image ( One where only things that have been seen before can be seen, and new discoveries become impossible. One where Baysian statistics are the only kind of statistics because reality itself has been contaminated by the incorporation of millions of runs of recreated bias, and there are no longer any valid reference distributions. One where no sense can be made of the unique, except as a specific case of an imaginary collective. Does that make any sense at all?

That sounds like science fiction but in fact is not that different from what already happens in some examples of nature conservation. What can we learn, for example, from endangered Chilean palms that have been planted in rows and rows in the image of a banana plantation, rather than being allowed to regenerate naturally, according to species-specific interactions and terrain-specific contingencies? (see:, . As we struggle to understand what the ecology of urban spaces consists of, we face a similar problem of nature expressed according to garden imagery ( and subordinated to the plans of human needs and logistics. Only where it escapes our knowledge and our control, as bats living in crevices, hedgehogs navigating back alleys, and spontaneous vegetation in abandoned lots, does an urban ecology that we can learn about and from emerge.

--Meredith Root-Bernstein, 1 Nov 2019

Thumbnail for Animals live in cities too…

Animals live in cities too…

Are cities a human phenomenon?

Thumbnail for A bit on art, meaning and gardens

A bit on art, meaning and gardens

Contemporary art has a lot to say about…

Thumbnail for What is biodiversity?

What is biodiversity?

BIOVEINS looks at where biodiversity is…

Thumbnail for What is green and blue infrastructure?

What is green and blue infrastructure?

BIOVEINS focuses on the role of green…

Thumbnail for Cities: what are they?

Cities: what are they?

We take cities for granted. But where…

Thumbnail for BIOVEINS Blog!


Welcome to the BIOVEINS blog!

Thumbnail for A morning with the Agence de Biodiversité of Ile-de-France

A morning with the Agence de Biodiversité of Ile-de-France

On a beautiful May morning, I meet…

Thumbnail for Biodiversity in European cities: a historical perspective

Biodiversity in European cities: a historical perspective

Is biodiversity in cities a futuristic…

Thumbnail for Call for papers

Call for papers

"Governing Urban Natures:…

Thumbnail for Cities, restoration & art

Cities, restoration & art

I met the artist Carmen Bouyer over the…

Thumbnail for Data collection for BIOVEINS: Bacteria and birds!

Data collection for BIOVEINS: Bacteria and birds!

Collecting data on leaf bacteria and…

Thumbnail for Data collection for BIOVEINS: Bees, bats, flowers and insects

Data collection for BIOVEINS: Bees, bats, flowers and insects

Recently I caught up with the BIOVEINS…

Thumbnail for Dispatch from Lesotho

Dispatch from Lesotho

I write to you from Lesotho, where I am…

Thumbnail for Diversity of our roots

Diversity of our roots

The new Exorigins project brings a…

Thumbnail for Field notes from Chile

Field notes from Chile

I am doing fieldwork in a village in…

Thumbnail for Flies are cool

Flies are cool

Flies are diverse, amazing, urbane, and…

Thumbnail for Flora of Urban Wastelands: An interdisciplinary project

Flora of Urban Wastelands: An interdisciplinary project

Interview with Audrey Muratet, one of…

Thumbnail for Fragmentation


One of the ecological concepts that…

Thumbnail for Urban amphibians

Urban amphibians

Are there frogs in cities? Where are…

Thumbnail for Experimenting with cities: news

Experimenting with cities: news

Matera, European City of Culture 2019;…

Thumbnail for Heat islands

Heat islands

Its hot and I’m in a heat island.

Thumbnail for Identifying urban plants

Identifying urban plants

Recently I found a guidebook to the…

Thumbnail for Imagination and Creativity for Cities and Nature I

Imagination and Creativity for Cities and Nature I

The Paris Forum on NBS [Nature-Based…

Thumbnail for Imagination and Creativity for Cities and Nature II

Imagination and Creativity for Cities and Nature II

Philosophies of action for resilient…

Thumbnail for Imagining the future of nature

Imagining the future of nature

On a recent visit to Milan I got to see…

Thumbnail for Integrating ecology in landscape design

Integrating ecology in landscape design

I recently went to meet Yves…

Thumbnail for IPBES Global Assessment Summary

IPBES Global Assessment Summary

The IPBES has today released a summary…

Thumbnail for Living with wildlife in cities

Living with wildlife in cities

Coexisting with wildlife in cities can…

Thumbnail for Living with wildlife in cities II

Living with wildlife in cities II

Reflections on mice and work.

Thumbnail for May Day- nature does no work

May Day- nature does no work

May Day is both a celebration of spring…

Thumbnail for Milton St., Northampton, MA, USA

Milton St., Northampton, MA, USA

I was taking a walk in Northampton,…

Thumbnail for Nonhuman gardens?

Nonhuman gardens?

If nonhuman animals made gardens, or if…

Thumbnail for Participate in our public participatory park mapping survey!

Participate in our public participatory park mapping survey!

Bioveins is starting a new survey to…

Thumbnail for Art posters for you

Art posters for you

We made posters to give away.

Thumbnail for Remarkable Trees

Remarkable Trees

Are there remarkable trees in cities?

Thumbnail for Ruins!


Ruins are poignant because they point…

Thumbnail for Methods in ecology: counting things and sampling

Methods in ecology: counting things and sampling

You do ecology by counting things. But…

Thumbnail for Some facts about the urban habitat

Some facts about the urban habitat

What kind human habitat is a city?

Thumbnail for strawbAIRies!


Strawberries and bacon-and-eggs:…

Thumbnail for Structural heterogeneity

Structural heterogeneity

What makes me enjoy certain gardens…

Thumbnail for Succession vs. land-use change

Succession vs. land-use change

In past posts we have talked about how…

Thumbnail for Surveillance and urban nature

Surveillance and urban nature

What does surveillance, data, and…

Thumbnail for Think like a large disturbance-making omnivore

Think like a large disturbance-making omnivore

A recent paper in Science summarizes…

Thumbnail for Urban pastoralism

Urban pastoralism

Recently I saw some eco-sheep in Paris.

Thumbnail for Urban transformative capacity

Urban transformative capacity

How to create an ideal city (?)

Thumbnail for Urban transformative capacity 2

Urban transformative capacity 2

Biopolis Summer School in Biology and…

Thumbnail for We love trees

We love trees

A real and an imaginary art exhibition…

Thumbnail for What are gardens?

What are gardens?

I consulted a couple of books about…

Thumbnail for What does it feel like to be an urban animal?

What does it feel like to be an urban animal?

Analogies from a city in chaos

Thumbnail for Its World Bee Day!

Its World Bee Day!

Every day that the world keeps on…

Thumbnail for World Cities Day

World Cities Day

Today is World Cities Day, promoted by…