This paper presents an infrastructural take on deep neural networks like Google’s DeepMind and examines the imaginary potential of the entanglement of research and engineering in contemporary Artificial intelligence.
Paper long abstract:
Demis Hassabis' brainchild DeepMind is considered a leap so big in the universalization of information processing models for Artificial intelligence that Google has recently acquired these algorithms for its photo search engine and for employ in Youtube user recommendations. DeepMind enacts processes that supposedly govern how episodic memory occurs at the crossroads of remembering and imagining "deep" in the human brain.
"Cherry-pick[ing] the key principles behind how we think the mind works," Hassabis and his colleagues have been bypassing the reconstruction of the brain's physical architecture, focusing instead on the flat algorithmic infrastructure of deep neural networks for machine learning and data mining. "Learning from nature, but not too much," such approaches entangle distinct images of mindful infrastructures and layered processes.
This paper will examine the often conflicting discourses of memory, physiology, and machinery that produce the metaphorical and material entanglements that make up contemporary Artificial intelligence in order to excavate imaginary infrastructures and alternative futures buried in the seemingly pragmatic production of this contemporary technology.