Algorithm — Tarleton Gillespie, Cornell University

A brief search through any popular new outlet of the term in question will tell you that algorithms can do a huge variety of things. Algorithms could replace GP’s. Algorithms actively increase the likelihood of extremists in America. Algorithms can predict actors’ peak years. And even: algorithms literally run the world.* In popular media, algorithms take on a messy existence, often ontologically divorced from their creators. They are agents who make decisions in their own particular way: logical, analytical, and seemingly value neutral. They are a tech-superpower’s midas touch. A horde of youtube channels offer prospective marketers updates and tricks to game THE Facebook algorithm. The same came be said for Youtube and Google. In being referred to as such, algorithms take on an obscure and consequently enchanted reality in people’s consciousnesses. This discourse conjures up an image of an ethereal, treasured, creature. The algorithm creature is of course kept in the most secure part of these tech-superpower’s super-offices. I imagine this creature is similar to No-Face from Spirited Away. Treated treated like a god who has approached earth, fed whatever it desires, and in return it spits out hard, frigid pieces of gold.

Gillespie’s demystifies this image, picking apart this mesh of ideas that are attributed to algorithms by first presenting how a software engineer refers to an algorithm and then comparing this to how the term is used by reporters. For software engineers, an algorithm is simply a set of mathematical steps for quickly and efficiently reaching a set goal. This set goal is a response to a particular problem that the algorithm is intended to solve or improve. Necessary for this is some notion of success. This is determined by software engineers and of course, can be value-rich. A quick analysis of the now infamous example of racist face-detection models helps to explain this. Should focus on notion of success - success from this chosen dataset - dataset problems  I find Gilespie’s notion of ‘algorithm as synecdoche’ useful to conceptualise this. An algorithm is one part of a whole system of decision making.

After this, Gillespie considers how the connotations of the term (logical, analytical, value-neutral) can be utilised in the information industries. The connotations and mystifications of the term allows companies to allude criticism, by ‘passing the buck’ from their management decisions to an algorithm.

The piece finishes with a close examination of what can be considered ‘algorithmic’, concluding that the ‘algorithmic’ is to be committed to automatic and mathematical procedure. Gillespie’s focus is generally on the role of the algorithmic as a notion in epistemic practices. What does it mean for knowledge to have procedure so strictly aligned with what is perceived as logical or analytical. I agree with Gillespie that a demystification is necessary to reveal the human practices at the heart of the algorithmic trend. Otherwise algorithmically generated information takes an undeserved superiority to other epistemological methods.

One aspect of the algorithmic that I think is sorely missed in this analysis is the materiality of the algorithmic. Algorithmic models are not ethereal entities, existing on the ‘cloud’. They exist in outsourced data centers which comprise of huge racks of computer memory storage. These centers have military style protection and omit approximately 2% of the world's carbon emissions. This is comparable to the entire aviation industry. Algorithmic procedures such as deep learning are incredibly energy intensive. (maschatesus paper)