Notes / Outcomes


part-time Terschelling inhabitant and urban planner Jan de Graaf | landscape artist Jeroen van Westen | nature conservationist and researcher Sander Turnhout | Cultural Anthropology student Michelle Geraerts | researcher and designer Sjef van Gaalen | artist and interaction designer Paul Seidler | artist and educator Tivon Rice | primary school pupil Jackson Rice | artist and researcher Theun Karelse

A glossary of terms
DATES: Arrive 19th March, workdays 20/21/22 ends 23rd.


The fieldwork trajectory

The team started from a birds-eye perspective of the island (Jan and Jeroen), and via the nature of environmental data (Sander), moved towards the character of artificial agents and machines as co-authors in creative practice (Paul, Tivon) and finished by modelling some systems in the local context of the island (Sjef) in an attempt to find patterns and characteristics that may be generalised beyond that local context.


An overview of main topics

An AI cannot think outside the box, because it is the box -
An AI has goals, whereas nature is just process. How can machine learning deal with environmental dynamics and complexity? When models are the only things that can be recognised by the system, what will it end up looking at? Perhaps this means moving away from the idea of a monolithic singular AI towards a tapestry of distributed artificial actors which become active/passive under changing conditions. (Algorithmic speciation) A system that evolves habits instead of laws and is geared towards recalibration. Does recalibration demand validation? And does validation demand a critical historic perspective?

Data and models -
Any known-unknowns in the field become unknown-unknowns in the models. Taxonomy is a debate rather than a fixed grid. Biodiversity observations are based on human interpretation, practice, morals, habits, ethics. That data is not a representation of reality, but rather translations (Latour) or creations (Gallon) and so are the models and trend-indications that are based on the data. A human expert may be able to deal with that trough field experience, but what constitutes field experience in an AI? How do you translate that level of environmental complexity into training data? Does it make more sense to train machines or humans? If it’s range of understanding is limited, perhaps the role of an AI is more like a watchman, which flags-up issues and then alerts / consults human experts? (Artificial-artificial intelligence) When does it become subversive? Can it use triangulation (Oracles) as a means of bypassing biases in data?

Governance -
If an AI operates publicly then PR becomes a relevant factor and to be successful the AI must win the harts and minds of the public. Will an AI construct a range of typologies of human observers to do that? To an AI different interest groups may look like style-transfers. Does the AI analyse only at landscape level or also at the level of environmental policy and procedure? If it operates legally it would start from where ever it has mandate in the landscape and try to expand from that. (building consensus) Otherwise deploy mind-control..

Value -
Can the value generated by the system fund exploration? Generate new knowledge and expertise? (Civic science) Can smart contracts evolve into something adaptable, governable? How to deal with intrinsic value of landscapes: attempts to integrate that kind of value tend to destroy our current economic models. What if the computational theory behind AI is not based on market philosophy but on ecology?


note: Random Forests are a type of analysis in machine learning in which a large number of simpler operations called 'Decision Trees' are combined. This 'Random Forest analysis' only uses portions of the selectors (at random) on the data, so individual trees can vary more, which increases accuracy.

This fieldwork session is organised in collaboration with the Wadgasten programme by Jan de Graaf and Jeroen van Westen.