some rando thoughts on algorithms

I have personal and professional investments in physical space. I depend on the nature of its borders and its porosity, its opacity and transparency. But that being said, I’m interesting in the way the authors we read discuss digital space as physical phenomena. Bucher says, in the forth chapter of her book, “Facebook’s newsfeed algorithm is explicitly framed as an architectural form that arranges things (relationships between users and objects) in a ‘right disposition.’” (38) (She mentions this concept earlier, quoting Foucault, claiming, “government is about ‘the right disposition of things.’”) Of course, her use of architecture and arrangement evokes space—the kind made up of feet and inches and oxygen and mid-century modern furniture. And, while I have absolutely no desire to have all space be digital space, I’m curious how seeing digital space as space might help us better understand its “multiplicity.”

Finn mentions the cybernetic ambition to “erase the embodied nature of information through abstraction” and reminds us that information is dependent upon a medium to exist. (30) This “medium” bringing to mind this same physicality but maybe on the level of hardware. I want to think about the “embodied nature of information.” To be clear, not to claim that all embodiment is informational, but to posit information as physical.

Both authors deal with the rhizomatic nature of all space, both digital and physical, the way power exists in the human, but also in the machine. Noble as well is looking to describe an integrated and reciprocal relationship, one of omnipresent prejudice. As Finn states, “implementation runs both ways,” (49) this suggesting a linearity that I don’t believe he meant to imply. He renders this implementation as a gap, another nod to physical space, and claims this gap lies between “computational and cultural constructions of reality,” created by the cultural machine that is in reality us. And I can’t tell if I’m misreading or reading into his claims, but as he continues, his argument feels lost in the computer, or at least in information, triangulating a “a kind of magic” between “the computationalist quest to continually expand the boundary of the effective procedure […] and the human desire for universal knowledge.” (52)

I wonder if we can think of information, and the digital, as spatial or occupying space, of its implementation as spatially constructed, might we develop a different relationship to its “liveness.” As Facebook arranges, it (we might say literally) places things at a distance from us, and of course places others within reach. Our “dispositions” and their arrangements are dependent on our physical positions (attentions and attitudes). When the algorithm abstracts, what happens to the matter faced with abstraction, and is this abstraction actually a throwing away, or even a death. Finn makes the point that “the boundaries of implementation seem endless because they are the boundaries of the material universe.” (48) My issue is that knowledge and cultural construction are (unfortunately?) not my entire material universe. And I wonder if it would be more fruitful to place the computer in our material reality, rather than our material reality in the computer.

Core Post 3: Cultural Algorithms

This week’s readings with their explicit positioning in STS or Critical Information Studies definitely hit home for me, and actually address many of the fundamental queries that guide my current study on the convergence of music streaming services and mobile dating platforms. The fictions of mechanical neutrality and “culture-less” science have come to occupy a central position in my research, and unearthing the cultural, social, economic, and political embedded in technological infrastructures that tout rationality and objectivity constitutes much of my current work on Spotify algorithms and their capacity to regulate user behavior and self-concept. But to focus specifically on the four readings for this week, all of the authors, in one way or another, bring to our attention the crystallization and proliferation of hegemonic discourses through the performance of algorithmic objectivity. Considering, here, Gillepsie’s discussion of the “recursive loop” (p. 17) in which he points out that the scope of the algorithm is essentially limited to the archive to which it is applied, I wonder if the algorithmic biases that Noble lays bare are, to a certain extent, not entirely algorithmic biases, but rather somewhat reflective of actually existing available information or “cultural algorithms” (p. 25) that are products of larger regimes of structural oppression. To elaborate, I am wondering to what extent is the reproduction of racist and sexist discourses online the product of existing problematic cultural frameworks being reflected in the algorithmic architecture compared to the actual negligence of corporations in reproducing them. To what extent is it inevitable vs. ‘designful’? Bleakly put, can algorithms ever exist outside the agendas of White masculine hegemony, neoliberalism, and capitalism, all of which are also tightly bound together themselves? I’m thinking here of Gillepsie’s example:

… a search for the phrase, “she invented,” would return the query, “did you mean ‘he invented’?” While unsettling in its gender politics, Google’s response was completely “correct,” explained by the sorry fact that over the entire corpus of the web, the word “invented” is preceded by “he” much more often than “she.” Google recognized this – and mistakenly presumed it meant the search query “she invented” was merely a typographical error. Google, here, proves much less sexist than we are. (p. 25)

Considering that Noble points out that Google has the capacity to “tweak” or “fix” problematic search returns when they are exposed (p. 82), perhaps what is desirable, then, is not the mechanical neutrality that algorithmic applications tout, but rather, explicitly ‘partial’ ethical algorithms that mobilize their totalizing capacity to actively counter the oppressive social and cultural discourses they currently perpetuate…

Core Post: Algorithmic Identity and Objectivity

One thing that I really liked about this week’s readings are how they managed to define algorithm in conversation with the social practices around it. I felt that the authors recognized the advantage of dissecting the digital in a more sociopolitical way. Bucher puts it perfectly in her introduction, by writing that we should “appreciate the question of how algorithms intersect with power and politics on a more philosophical and sociological scale” (20). After all, the digital space being analyzed wouldn’t be what it is if it weren’t for the technology-human interaction. This cycle of technology affecting us and us affecting technology is important to defining technology.

I particularly liked how Gillespie broke down the definition of algorithm in a way that displays our anxieties related to algorithm (and its objectivity) and the role it plays in our lives as well as how it constructs our algorithmic selves.

First, Gillespie builds on the idea of how digital trace and shared information between websites and companies work together to create a “digital dossier” or “algorithmic identity.” He talks about it more from the invisible side, where there is this database of binary information that constructs our identity based on our past and present digital ventures and entries. I this also transcends to the visible level – mostly with online content creators, who are constantly making an effort to make themselves “algorithmically recognizable” (18) by the platform in which they want to be featured at the top. These two sides of the algorithmic identity seem to complete a digital alter-ego, where there’s the combination of how the digital reads a person and how a person chooses to provide information about themselves to the database, choosing how their algorithmic identity will be shaped.

Going back to objectivity, I think Gillespie’s discussion on searches and how certain items become more relevant to be featured at the top of lists is a very important one. He writes about how each website has its different secret criteria for judging relevance and how it’s complicated to define this relevance, especially with the competition between websites and the coding that goes behind them. However, I do think that we deserve to have a better understanding on what exactly is relevance for each platform. How different is the “sort by relevance” on my Amazon account from yours? Or how different is that sort between different stores? Is something more relevant based on my preferences or on other people?

These platforms are designed to look and feel objective, but we should keep questioning the options we are given. I thought it was ironic when Gillespie brought up that “No provider has been more adamant about the neutrality of its algorithm than Google” given the many issues that YouTube has had with algorithm. Recently, several LGBTQIA+ YouTubers filed a lawsuit against YouTube for restricting their videos, while featuring similar content on the main trending page (hi James Charles). This made many creators question YouTube’s objectivity as it looked like they were prioritizing “A-list” creators and discriminating others. It’s an interesting case to look at. If you’re interested, here’s a link to a Forbes article on it:  https://www.forbes.com/sites/rachelsandler/2019/08/14/lgbtq-creators-sue-youtube-for-alleged-discrimination/#74407c80788e

Core Post 4: The Ease and Unease of Algorithms

It was interesting to read Noble’s points about whiteness and maleness as the dominant hegemony that configures algorithms, but my recent experience with digital technologies was actually quite different. Instead, I found myself being bothered by the utter lack of generality, and the hyper-customization of information services, which approximated my identity to a “shadow body,” as described by Gilespie (Gillespie 8).

Just yesterday, the undergraduate students in my section for Introduction to Television expressed all kinds of shock and horror at the fact that Netflix ‘targets’ its posters for content according to the racial identities of the viewers. See the relevant Guardian article here: https://www.theguardian.com/media/2018/oct/20/netflix-film-black-viewers-personalised-marketing-target.

One of the students, who identifies as a gay black male, relayed his experience about clicking on a show because the poster centrally featured someone who looked like him — but even after three seasons, the character was barely developed. Other students expressed similar distress and unease at the niche, stereotyped marketing tactic driven by customized algorithms and user data. They were outraged to be reduced to a category, to be rendered as “shadow bodies” in place of the whole of their identities. I asked students to send in screenshots of their Netflix pages so that we can explore this phenomenon further.

Reading this week’s articles made me think about the ease and unease associated with algorithms, and where that border lies. In a neoliberal world, when we use certain devices and services, we inevitably become laborers: as Gillespie notes, “it is seductive… for information providers to both track and commodify that activity in a variety of ways” (Gillespie 7). Our labor allows information systems to generate “user-caricatures” that categorize and typify us, like Netflix. But our labor is voluntary — even if we are uneasy about the system, the ease that it provides for us usually outweighs our unease.

My students are probably going to continue being bothered by the racialized posters, but also continue using Netflix. I continue to use my Echo dot to turn on my lights, even though I am aware and wary that it is always listening (Does it speak Korean?!). There certainly exists a paradox here… We know, and we fear, but we still labor for these systems. Are the systems inescapable? At what point does our unease become sufficient to resist the ease of the algorithms?

And I am thinking about all of this and writing this post in a Google Chrome Incognito window…

Core Post: Long Tails

In Algorithms of Oppression, Safiya Noble argues that Google’s search engine algorithms are not neutral; quite the opposite, they prioritize corporate interests, automate cultural biases, and push damaging stereotypes to the top of the list. The resulting representations have deleterious effects on groups represented as well as those forming opinions about different races, genders, religions, and other marginalized groups.  In the context of PageRank functionality, it is also important to note that Google Search is unwaveringly the most visited and linked to site, ensuring it’s own supremacy as an epistemic model: "[algorithm’s] logics are self-affirming” (Gillespie). In 2016, however, Google officially turned off its display of PageRank data. The public was no longer invited to see the live rankings (or the unconditional supremacy of Google itself). Interestingly, around the same time, PageRank was upgraded with PageBrain, which employs machine learning to deliver query results. With this shift, Google claims that it can locate users’ true intent, reinvoking the same issues of relevance and credibility at the heart of Noble’s critique. PageBrain also specifically addresses the question of how to respond to long tail queries that have not been entered previously. Jumping ahead in the text, I was particularly interested in Noble’s examination of Long tails: Noble articulates the increasingly reductive nature of query keywords. Users generally enter the simplest possible word combinations, casting a wide net and relying on the search engine to return relevant, popular, and credible results. Consequently, advertisers are incentivized to frame content under the most generic terms. As paid content rises to the top, critical perspectives are forced further down the list. Locating non-commercial content now often requires more descriptive and specific search terms; these ‘long tails,’ are often the only tools for directing algorithms away from commercial and pornographic content (Noble, 88). Even if PageBrain is more responsive to long tail queries, questions of simplification, bias, and causality have even higher stakes in the context of machine learning, as Taina Bucher lays out in If...Then:Algorithmic Power and Politics. These algorithms operate as cultural logics with productive power, shaping subjects and events. In The Relevance of Algorithms, Tarleton Gillespie asserts that “it is important that we conceive of this entanglement not as a one-directional influence but a recursive loop between the calculations of the algorithm and the ‘calculations’ of people.” Although machine learning algorithms are are fundamentally human, seeded by human values engaged in training processes, the recursive loop creates so much complexity that predictive algorithms in particular are increasingly beyond their engineers’ comprehension. Algorithms and effective computability may represent a new model of knowledge, but it is one we can’t quite wrap our heads around.

"Everything becomes buffering"

I recently watched a video by UK novelist Tom McCarthy, in which he imagines an extremely technological future where everything has already been written, all information recorded. What was most interesting to me was how he discussed thinking and memory in comparison to algorithm and buffering. First, he describes buffering as “hordes of bits and bytes and megabytes all beavering away to get the requisite data to [us].” Then he talks about the experience of watching a video online. Specifically "when the red part of the line under the video catches up with the gray part and buffering sets in." McCarthy compares this to memory: “This is how time and memory work: we need experience to stay ahead if only by a nose of our consciousness of experience so that the latter can interpret, narrate, the former. When the narrating cursor catches right up with the rendering one, we find ourselves jammed, stuck in limbo. Everything becomes buffering and buffering becomes everything.” I thought this was a really interesting and cool analogy when thinking about how the digital processes memory and ways of thinking. I recommend checking out the video:

core post: my abstraction of their abstractions

To borrow Finn’s titling of his sub-section, one “thread” throughout the readings is algorithm as abstraction. For each author there are varying implications, or consequences, of these abstractions reflected in their approaches and descriptions of algorithms. Whether the algorithms abstracted “effective computability,” “desired output,” or power and oppression, there’s a mutual recognition in some kind of currency (linguistic, political, racial) imbued within any algorithm. The notion of the non-cultural, apolitical algorithm, due to the  genealogy of algorithm through mathesis universalis, the Turing machine, information theory, etc. is held under scrutiny—to varying intensities. From these different intensities, I found Bucher’s and Noble’s selections to be more compelling, and frankly, easier to read because of their immediacy and urgency towards the consequences of, and around, algorithms. 

Both Bucher and Noble immediately foreground the urgency of their work, which guided me through their argument and its stakes. Referencing Bucher’s inclusion of the “if-then statement”: if there are future cases similar to the discriminatory incidents with Google’s photo app, its search engine, or Flickr’s racist labeling, then what does that say about the conditions/parameters/desired outputs that Google and Flicker implemented for that algorithm or model? Bucher and Noble hold someone(s) (Google, Flickr, Facebook as corporate giants comprised of design engineers)accountable instead of something (the algorithm’s ontology or epistemological claims), seeing past the algorithm’s computational “nature” or essence (Noble 22). 

One distinction between the Bucher’s and Noble’s selections are the scopes/scales framing their arguments:

Bucher’s concerns lie within the broad and varied scope in which algorithmic power can emerge from: "society and human interaction" and “states of domination”, generally referring to governments/governmentality (Bucher 33). However, algorithmic power varies according to the types of algorithms that she lists: one is more deterministic because it produces the same outputs by having an established set of rules/conditions, and the other is “predictive” or “machine-learning”. Finn similarly categorized these distinctions as the engineer’s algorithm, geared towards defining solutions for problems, and the computationist’s algorithm that hinges on the algorithm’s “desire” to learn what is “effectively computable” (Finn 23). 

Noble points to the technological racialization within a clearly defined landscape of neoliberal America that has historically developed social and economic policies written with the promises of individualism, “personal creativity,” and participation, bridging the (digital, racial, class) divide, remaining to be delivered. I also want to share my awe towards her critique of American neoliberalism/neoliberal capitalism through the allegorizing the “democratic landscape” of the internet through Google and the Alphabet (Noble 84); it was such an effective, compelling, and convincing move. 

Thus, with respect to their scopes, I found Noble’s selection to be more generative, and exemplary for future considerations of similar projects, insofar as the relations of power,—racial, economical, etc.—,were specific: the historical continuity of racial and gender formations, along with racial and gender normativity, around White-American-(male)ness reconfigured through forms of algorithmic oppression in contemporary, neoliberal, capitalist America. Lastly, I admire Noble’s unrelenting and bold tone/rhetoric in her critique of engineering curriculums across the biggest research universities as well as characterizing Google and the alphabet as agents of cultural imperialism, colonizing the web.

On a lighter note, one way that I might imagine myself exerting some kind of “power” over algorithms and their corporate overlords is not listening to navigational apps or their suggested routes. Small victories. For example, Google Map lists three metro-transportation routes that I can take back home. Google Maps navigational algorithms are based on the logic of Djikstra's algorithm and A*, both attempt to "solve" the "problem" of computing the shortest route possible. For these algorithms, shortest means some combination of distance travelled and ETA, presuming that users desire this kind of efficiency. However, along the most efficient route to home is an inescapable walkway filled with bird crap because it’s underneath a bridge in which a bunch of a pigeons nest and coop along the ceiling construction. Google maps insists that it’s the “fastest” or best route, but I would like to see any Google maps software /app engineer walking through hills of bird poop, roadkill, or dead animals.