Algorithmic Culture – Striphas – 2015 – longer notes here
This article seeks to outline the development of data-driven algorithmic culture, using Raymond William’s (1983) sociohistorical approach to keywords to help define the artefact. The quantification of culture is examined here through language, with an inventory of language allowing for a questioning of the seeming objectivity and naturalness of today’s algorithmic culture. Three keywords are explored here as a part of this inventory of language: information, crowd, and algorithm.
The etymology of language is discussed first. Peters’ etymology (1988), states the origin of the term ‘information’ is ‘a history full of inversions and compromises. Gleich (2011) discusses the religious origin of ‘information’, meaning ‘giving form or essential character to something’. Modern understandings of ‘information’ relate to communicating something about a fact, subject, or event. This is a move towards object-oriented thought and communication. Information is seen as separate from the individual, as being a raw material to be transmitted between people. Wiener (1954) discussed ‘information’ in this manner, as being open and fractured knowledge, directed at anyone who will receive it e.g. automatic doors sliding open when something passes. This means information is not just of people but also of machines. Cultural life can thus be seen as being conducted by both humans and machines today but this does not guarantee it is improving.
The etymology of crowd is discussed second. The verb ‘to crowd’ comes from Dutch, German, and Frisian in the 17th century meaning to pressure or to push. The noun ‘crowd’ tends to be used negatively, to indicate large groups of unaccountable and unwieldy people. Mackay (1841) examined crowds in this manner, seeing them as having a ‘popular mind’, making people passive rather than generative. Le Bon (1895) suggested we were entering into the era of the crowd, with crowds having an ‘unconscious genius’, likened here to Smith’s (1776) invisible hand of the market. Hayek (144) and Olson (1971) expand on this, suggesting crowds do have an underlying logic in collective action. Williams (1958) rejects the idea of the masses, rather saying, ‘there are only ways of seeing people as masses’.
Finally, the etymology of algorithm is discussed, given that they have a central role in culture from the mid-to-late 20th century onwards. The term algorithm is essentially a corruption of the mathematician al-Khwārizmī’s name, which is latinised as Algorithmi. This led to the term algorism initially. This is tied to the origin of the terms algebra and cypher as well, essentially stating that the term algorithm started as a mathematical procedure to bring about some form of truth. This is juxtaposed to algorism, which can hide as well as reveal truth. Algorithm gained popularity over algorism due to Hartley’s (1928) Transmission of Information paper and Shannon’s (1948) Mathematical Theory of Communication paper. These basically suggest that we need to sort signal from noise to achieve a mathematical communicative order. This requires a set of procedures to govern successful communication, leading to the use of ‘algorithm’.
There is then a discussion of keywords connecting to culture. Arnold (1869) suggests culture is of the elite to uphold normative views of what society should be. Today, algorithms act as a principle of authority, producing culture through its information processing and reassembling the social. Algorithmically driven culture is produced privately, in black-boxed programmes. The end-results of algorithms are presented as authentic to reality, rather than decisions made by code, hidden from the user. ‘All this harkens back to the oldest sense of information – where some mysterious entity is responsible for imbuing people and objects with shape, quality or character’.
Digital failure: Unbecoming the “good” data subject through entropic, fugitive, and queer data – Bridges – 2021 – longer notes here
The concept of digital failure is introduced here, discussing when ML or AI technologies reify existing oppressions within society. Pasquale (2019) identifies two waves of algorithmic research around this: first, research focused on diversifying data inputs; and second, concerns around structural forces and power. This article deals with the second concern, asking how we can resist seemingly invisible platform governance? Many existing systems are deemed as ‘flawed algorithmic systems’ here, meaning error is central to how they work. Digital failure, suggested here, is about users failing to work in service of dataveillance. This refuses the idea of universal truth to be found in data, asking whose data is it and to what end is it being used? Entropic, fugitive, and queer data are examined here. Entropic data is the unaccountable data, subject to identity resolution by marketers; fugitive data is seen through misidentified data in imagine recognition software; queer data is seen through the posts of a sex worker and their posts which speak at moderators and platform owners. It is recommended to look at the article, as much of this is shown through images.
Firstly, however, data itself is examined. Data is said to be called into existence: not neutral but shaped (and shaping) sociocultural politics. Data is when the many becomes the one and the one represents the many (Galloway, 2015). Data is positioned as rhetorical here (as opposed to ontological or epistemological). It is neither inherently true nor untrue, rather it is a representation of ways of cultural seeing (Spiller, 1987). Digital failure is about unbecoming the digital subject, resisting the digital classification process to escape the conqueror’s hold and data’s claim to universal truth (Gauchet, 1997). This is a ‘low theory’, seeking possibilities through a refusal to conform to established hierarchies of knowing.
Big Data was supposed to make users entirely known to marketers but much of the data gathered has turned out to be indescribable or useless. Bridges describes this as entropic data, as it takes energy to gather, which then dissipates without a return. Data brokers seek to get around this through selling two types of identity resolution: deterministic and probabilistic. The latter uses entropic data to make guesses at a user’s identity. Despite being less accurate than deterministic data matching, probabilistic data matching has value through creating lookalike audiences, used for reach and scale (Frohlich, 2018). This is useful for marketers because Big Data needs to be ‘activated’ (meaning it can be sorted and cross-referenced across datasets). This allows ‘failing’ citizens to be classified and tracked, either by marketers or government bodies. This serves to solve the ‘problem’ of entropic data by making it knowable and valuable by creating the user as the ‘transparent I’.
Fugitive data has its origins in the ungendering of blackness in the antebellum period to allow slaves to move through spaces safely (Snorton, 2017). It also draws on the miscognition of patterns in data, called apophenia (Steyerl, 2016). Fugitive data has agency and refuses to be ‘cleansed’ by analysts. Crawford and Paglen’s (2019) ImageNet Roulette is given as an example of fugitive data in action. This allowed people to see how their photos were classified by the software ImageNet. It basically showed that ImageNet’s classifications were often predicted on stereotypes coded into the software. ImageNet went on to change millions of classified images after this. Pointing out inaccuracies here is an action for rather than of freedom: they are needed to show systemic assumptions and oppressions, which are only changed after the problem is highlighted by those subject to it.
Queer data is seen here primarily through its use of sousveillance, flipping panoptic power and watching from below. Britzmann (1995) suggests three forms of queer data reading: the study of limits; the sudy of ignorance; and the study of reading practices. The example used here is of Charlieshe (a sex worker and micro-influencer). She speaks parodically to Instagram moderators and its owner, Mark Zuckerberg, in order to highlight inequities in content moderation. The ironic humour used here is deemed a mode of disidentification, seeking to negotiate power and counter discourses (Muñoz, 1999), while highlighting the censoring of non-normative bodies/bodies which are deemed beyond comprehension by the platform.
The cases given here show the difference between a flawed system (one which is sexist, racist, classist, etc) and digital failure (in which people resist being understood by these systems). This intends to show that these technologies are not objective and cannot make claims to truth, as well as moments in which people are able to speak back to these systems.