Misrecollection of forgetting’ -The absent minded programmers guide to machine learning

“Programming is forgetting” teaches Allison Parrish, an NY-based artist, and programmer (Parrish, 2016). What she means is that by translating functions, actions and material into the digital world by programming we make compromises and forget the complexity of the world. Such accommodation is essential; otherwise, it would be very hard to create any digital programs. For instance, translating sound into digital form means choosing what components of the sound are vital and what are not. Same goes for digital photographs. We have to choose what is essential in the sounds or images and then translate that into repeatable procedures and collectable data. As such, digital data always represent the choices of the one programming the procedures of collecting that data. Parrish has a simple example of collecting personal information. First and last name and gender (male/female) are common data to inquire. However, as Parrish, points out, such data presents the perspectives of a specific part of western culture. For instance, name is not that relevant in many other cultures. Moreover, asking for a binary gender is already problematic (if not plain wrong). As such, digitalising information is profoundly political, ideological and subjective. 

Our physical life is rarely binary, black or white as marked for instance by Rushkoff (Rushkoff, 2010) but in digitalising information, we translate that complex world into binary ones and zeroes. Thus, digital can only present a small fracture of our whole lifeworld. As our lifeworld gets filled up with digital practices (Or it already is)  then forgetting all the things that are left out of the digital procedures becomes a problem; If we treat our digital, and data filled, world as a whole, we settle for an impoverished world, that is created from limited perspectives.

These challenges become even more apparent, and complex, in the current climate of artificial intelligence systems, namely systems that employ machine learning and deep neural networks. These systems treat the digital data as their (only) world of information. Based on that data these systems are capable of producing solutions that human intelligence could perhaps not be able to do. However, we have to understand that these solutions are based on a  limited perspective of the world. Therefore they will always carry, and multiply, the biases and (mis)objectives of the collected data. Furthermore, the power in these systems is in their efficiency to produce solutions from a vast amount of complex data. What happens is mathematical recursion, and the solution is the most mathematically optimal one based on that data. Furthermore, the intelligence in AI systems is deeply rooted within a limited perspective of intelligence. As noted by Hayles, the intelligence refers to a limited view of intelligence as a “formal manipulation of symbols, rather than enaction in the human lifeworld” (Hayles, 2008 p. Xi.  ).

If programming is forgetting, then using machine learning and deep neural networks is forgetting that we forgot

If programming is forgetting, then using machine learning and deep neural networks is forgetting that we forgot; We treat data and learning models as the whole world when we are in fact dealing with a small portion of the perceived lifeworld. In the post-digital age of ubiquitous digitality, we should start to remember again that we are forgetting and to start thinking the process all the way through again.


Hayles, N. K. (2008). How We Became Posthuman: Virtual Bodies in Cybernetics, Literature and Informatics. Chicago: University of Chicago Press. Original work from 1999.

Parrish, A. (2016). Programming is forgetting. Presented at the Open Hardware Summit, Portland. Online version: http://opentranscripts.org/transcript/programming-forgetting-new-hacker-ethic/

Rushkoff, D. (2010). Program Or Be Programmed. Ten commandments for the digital age. New York: OR Books.