A genealogy of the ambiguous use of digital memory in machine learning and artificial intelligence
pdf

How to Cite

Numerico, T. (2020). A genealogy of the ambiguous use of digital memory in machine learning and artificial intelligence. Mechane, (2), 25-40. Retrieved from https://mimesisjournals.com/ojs/index.php/mechane/article/view/1608

Abstract

This paper aims at a genealogy of the digital memory. The use of the same word for an electronic repository and for human skill produced a misunderstanding that caused the equivocal merge of two divergent concepts. Electronic data kept in a registry of information was interpreted like remembrances, recorded experiences, plus the capability of sorting out the right information when the quantity of stored data progressively available is too much. The more data was available the more machine learning techniques were compared to human cognitive processes necessary in understanding things and acting appropriately according to the acquired knowledge. Digital data retains a presumed ‘purity’ as the representing technical tool adopted to store information within a digital repository, whose name is always echoing a human faculty. The digital storage is, in fact, a ‘digital memory’. The use of such a world involves a pretence of identity between the human memory and the memory of the artificial device and is a key factor in disguising the mediation layer, necessary to supplement the digital storage.
The paper argues that these prejudgments relate to the genealogy of the idea of digital data, as preserved in computer memory, or in a computer network memory. According to Norbert Wiener – the father of cybernetics – human memory can be reproduced in digital machines, provided there is enough space within it. According to Wiener the idea of memory is relative to the quantity of accessible data that can be stored in it. In his work he continues the project also suggested by other scientists – such as Alan Turing, Warren McCulloch and Walter Pitts, among others – that human memory’s activity could be achieved perfectly well in the digital device, provided enough space is allowed for storing the relevant information.
This paper argues, instead, that data can only exist in accordance with a chosen interpretation and can be accessed through an implicit agreement on its inherent meaning, that strongly depends on quantification, measurement or on any other technique used to capture and organize it. Those interpretation choices are reticent, often unconscious, and always blurred inside the system.
The hypothesis that data gathering automation leads to a surplus of meaning, free of subjective judgment reinforces the autonomy of the system on a technical and symbolic level. It will be shown that this approach is not supported by valid epistemological arguments and that it involves a loss of control over the infrastructure of meaning, that could have retroactive effects also on human beings who should conform and obey to rules that they could not regulate or determine.

pdf