Click the star to add/remove an item to/from your individual schedule.
You need to be logged in to avail of this functionality, and to see the links to virtual rooms.

Accepted Paper:

Reframing the View from Nowhere as a View from Somewhere: Machine Learning, Remote Sensing and Environmental Science  
Erik Ljungberg (KTH Royal Insitute of Technology)

Paper short abstract:

In this paper I aim to show how machine learning applied within the domain of remote sensing serves to strengthen the hegemony of the techno-scientific world-making project. I further propose that it is necessary to reverse-engineer the God-trick and to ground the view from nowhere.

Paper long abstract:

In this paper I argue that environmental science is implicated in a particular world-making project whose basic logic is to flatten the multiple co-existing worldspaces of the pluriverse and to produce a world-image that brackets embodied ways of enacting the world in favor of the world as viewed from the view from nowhere. I further highlight this dynamic as it appears in the application of machine learning within the domain of remote sensing. Remote sensing is fast becoming one of the domains of environmental science in which machine learning is becoming a methodological mainstay. In order to contest the denigration of the more-than-human world and the ontological hegemony of the techno-scientific world-making project, I propose that it is necessary to reverse-engineer the God-trick. Thus, in this paper I aim to bring into view the intricate capillary networks of digital knowledge infrastructures from which the combination of remote sensing and machine learning emerges, and to reframe the view from nowhere as a phenomena that is specific to particular practices of handling data. I focus specifically on the technique of feature extraction and show how one of the novel effects brought forth with machine learning techniques is the ability to combine features within digital hyperdimensional planes. The ability to relate features with an n-dimensional hyperplane sets the stage for an almost endless succession of combining and recombining features, a phenomena for which I propose the term hyper-combinatorialism. This may be viewed as a considerable amplification of the decontextualizing capacity of environmental science.

Panel P12
LIE-DARs: Grounding remote sensing and environmental AI in perspectives of algorithmic injustice and colonial legacies
  Session 1 Monday 6 June, 2022, -