IP 3: Algorithms

In Neyland’s 2019)  treatise on the everyday life of an algorithm, he looks at how algorithms “participate in the everyday (13), how they “compose the everyday” (13), and how the algorithms become the everyday. Using this framework, I will examine how the ways in which algorithms participate, compose, and become our everyday have reshaped the meanings and bounds of the following concepts. 

Informed consent

Algorithms have enabled the mass extraction of personal data. The sheer amount of data extraction that algorithms can process has reconfigured informed consent to be something that is not necessary to acquire when a company or the government uses a person’s personal data. It is as though the process of becoming part of a mass amount of data depersonalizes personal data. That data is no longer seen as specific to a person and their history. Instead, it is viewed through the context of training data that can be used to support and refine machine learning instead. This depersonalization of data likely figures into why companies and institutions do not ask for consent, much less informed consent, to use people’s information which is retrieved via the web. Crawford points out how “machine learning systems are trained on images like these every day–images that were taken from the internet or from state institutions without context and without consent” (p. 94). Those of us who upload photos to Facebook, Instagram, Twitter, and other social media sites are not informed that such images may be used to train facial recognition systems, much less asked for our consent. Even researchers at universities (such as Duke, the University of Colorado, and Stanford) who should understand the importance of informed consent extracted and used people’s images without their consent to train facial recognition systems (Crawford, 2021)

Fair use

Leval (1990) explored and defined the factors that determined if the use or reproduction of copyrighted material was considered “fair use” so as not to stifle progress in the arts and sciences. These four factors were “the purpose and character of the use, the nature of the copyrighted work, the quantity and importance of the material used, and the effect of the use upon the potential market or value of the copyrighted work” Leval, 1990, p. 1110). The latter two factors are self-explanatory, but in relation to the first two factors, that meant texts which substantially transformed the original work were better able to assert fair use (Leval, 1990). As well, transforming or using works that were considered private instead of public publications made a stronger case for fair use (Leval, 1990). 

AI algorithms complicate this notion of fair use since they are trained by extracting and using the works of actual human creators. Would the writing produced by ChatGPT that aims to replicate human writing based on a collection of works by human authors then be considered a violation of copyright? According to the original principles of fair use, it would likely not be considered a violation of copyright. What if the prompt asks the generative AI model to write a summary of paywalled and copyrighted material? Would the summary then constitute fair use or a violation of copyright? This is currently an ongoing battle between the NYTimes and OpenAI, in which the NYTimes alleged that ChatGPT would produce work that was verbatim paragraphs upon paragraphs of their content (CBC News, 2023). These questions and complications suggest that the delineations of what is and what is not considered fair use may need to evolve to take into consideration how AI models are trained.

Discrimination and net neutrality

According to the CRTC (n.d.), net neutrality ensures “that all traffic on the Internet should be given equal treatment by Internet providers with little to no manipulation, interference, prioritization, discrimination or preference given”. The opposite of net neutrality is if certain content on the internet is given priority by internet service providers so that the consumer can access that type of content at a faster rate compared to other types of content. In theory, net neutrality should ensure that all types of content load at the same speed. However, as Slavin (2012) covers in his TED Talk, in reality, there can be a difference of microseconds in speed depending on one’s distance from where the internet infrastructure is located. Microseconds may not necessarily matter for us when we are loading a webpage, but these microseconds matter for Wall Street algorithms buying and selling shares (Slavin, 2012). The amount of data being exchanged and the speed at which it is exchanged is too great for us to register much less comprehend, but within the world of algorithms, these microseconds matter. Due to this, Slavin (2012) notes how buildings are being vacated and land is being shaped by the location of fiber-optic cables and internet infrastructure. 

Personalization

Originally, the notion of personalization related to being able to find content that would be most relevant to us based on location and past browsing and clicking behaviour. However, over recent years, it has become more and more clear that Google search algorithms are more interested in personalizing the advertising aimed at us. After searching for information about paddleboards, I started getting targeted ads on Facebook and Instagram for paddleboards. When I look up what the best beginner paddleboards, I am met with multiple sponsored ads from different companies pushing different types of paddleboards which are not necessarily the best beginner boards (see Figure 1). I have to scroll past 4 sponsored webpages before I get to the blogpost from a paddleboard blog that actually speaks to the information I am looking for. 

In Noble’s (2020) interview with Kandis, a black female hairstylist, Kandis shared how Yelp’s algorithms pushed her business down several pages later because she did not purchase their advertisement services. Kandis noted how she was the only black hairstylist in her area, and yet, when she put in targeted specific terms to find hairstylists in her area that specialized in working with Black hair, she still was not able to find her salon in the results. This means that if a Black woman were looking for a hairstylist who specialized in working with Black hair, they would not be able to find Kandis’s salon despite it being the only one catered towards Black women (Noble, 2020). This demonstrates how the algorithms are personalized with advertisers’ interests in mind, not the web users’. Personalization now relates more to how companies can use user data to personalize targeted web ads to manipulate users into buying more and consuming more. 

Friends 

Algorithms are shaped by user behaviour and while also shaping user behaviour. An apt example of this is how Instagram algorithms treat “friends”, those who you follow and who follow you, on your Instagram feed. In the past, on Instagram, friends’ pictures and videos would show up chronologically. Currently, Instagram’s various algorithms will take into consideration how often you like and interact with your social media friends’ posts (Mosseri, 2023). Using that information, its algorithms will ensure you see friends’ posts more often if you have historically interacted with their posts more. Or, its algorithms might push the posts of friends who don’t post as often or friends whom you have not historically interacted with much (in terms of liking and commenting on their posts) further down your Instagram feed. By doing so, they make it less likely for you to see the posts of those friends. As O’Neil (2017) astutely notes, “algorithms don’t make things fair. They repeat our past practices, our patterns, they automate the status quo” (6:01). In this case, how the algorithm interprets your relationship with your online friends will dictate how often you see their posts on your feed and thereby how often you can interact with their content regularly. We can see here how Instagram algorithms are using the past to configure the present and future. This is especially a problem when the algorithms uncritically take in systemic biases and reproduce those, such as when algorithms checking job applications are biased towards white-sounded names and biased against black-sounding names. Or, when, algorithms uncritically reproduce racist and misogynistic stereotypes around Black women within the search results that are first shown, as Noble (2020) demonstrated when she typed in “Black girls” into a google search that resulted in dehumanizing pornographic results. 

References

CBC News. (2023, August 14). New York Times explores legal action against OpenAI over copyright concerns. CBC News. https://www.cbc.ca/news/business/new-york-times-openai-lawsuit-copyright-1.7069701

Crawford, K. (2021). The atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.

CRTC. (n.d.). Different types of Internet services. CRTC. https://crtc.gc.ca/eng/internet/diff.htm

Leval, P. N. (1990). Toward a fair use standard. Harvard law review, 103(5), 1105-1136.

Mosseri, A. (2023, June 30). Shedding more light on how Instagram works. About Instagram. https://about.instagram.com/blog/announcements/shedding-more-light-on-how-instagram-works

Neyland, D. (2019). The everyday life of an algorithm (p. 151). Springer Nature.

Noble, S. U. (2020). Algorithms of oppression. In New York University Press eBooks. https://doi.org/10.18574/nyu/9781479833641.001.0001

O’Neill, C. (2017, September). The era of blind faith in big data must end [Video]. Youtube. https://www.youtube.com/watch?v=_2u_eHHzRto&t=510s

Slavin, K. (2012, November.) How algorithms shape our world [Video]. Youtube. https://www.youtube.com/watch?v=ENWVRcMGDoU