for the fifth class of Hannah Davis' Generative Music class, we were asked to find an example of bias in machine learning to write about and discuss next class.

i am not sure if it is possible to create a concrete application of statistics without some / many aspects of bias. a year ago, i saw allison parrish give a talk where she said "data is a form of forgetting". hearing that almost felt like a relief. it succinctly explained all of the conflicted feelings i had towards the adoption and efficacy about this current era of machine learning hype. it'd be hard to get anything done if one saw the atoms for the forest, but closing your eyes should disqualify any claims of objectivity or absolution of guilt.

many egregious examples of accelerated prejudice via machine learning have been covered throughly and thoughtfully in journalistic and academic outlets, and many more examples have yet to be given their time on the newsfeed. nobody has the capital to invest in "data-driven" decision-making and incentive to proactively preserve white supremacy like the neoliberal state and market. Cathy ONeil defines a "weapon of math destruction" as an opaque computational application deployed at scale in situations where the potential for societal consequences for the (usually unwitting) "user" is extreme. there are few things as widely scaled yet elusive as the american world-police corporate partnership, and few things as oppressive as the manic obsession of western imperialism to protect its racist, sexist, and capitalistic hegemony at any societal and environmental cost. one might need a big data application just to count up the number of lines of code written to tell the police to only hang out in black neighborhoods, categorically withhold opportunities to generational wealth to those who not already have it, and remotely bomb those who happen to be in the way of white businessmen.

in the presence of these fascist practices, bringing up the topic of the ad tech industry might sound myopic, but i think there is potential value in acknowledging that many of these data-centric methods were not only created using technologies and tools originally funded by a small handful surveillance-capitalistic organizations, the majority, if not all, of their popularity is the result of the financial success of facebook, google, and amazon. the "startup approach" is credited with "saving" healthcare.gov. more and more government agencies are creating "incubator" programs every month. the attraction is mutual. many consumer technology companies, confident in their colonization of citizen wealth are now looking to capture the capital in public contracting by embedding themselves in the public sphere as a form of "disruption". google looks to tap into all wireless internet traffic with "free" wifi. the founder of oculus is selling "call of duty for border control". uber and lyft look to tell you to take the subway instead, because they don't foresee it as being their "competition" for much longer. with the near-hourly mentions of facebook and google in daily conversation, it becomes simply more convenient to believe the narrative that these are simply companies successful as the result of their collective ingenuity and long-term investment in paying as many steve-jobs-fans as possible. it also becomes harder to remember that that digital advertising is fundamentally based on "Direct Marketing" which is an elevated term for referring to spam mail. i once went to a talk presenting a metaanalysis of reports on the efficacy of digital ad targeting. the study found that the average claim for the rate of return on this method is to misreport by more than 1000% (one thousand). why do people pay for online advertising then? in practice, one reason is because there are far more resellers of facebook and google ad space than facebook and google. another reason is that doesn't look "good for brands" to not insist that they're not old by dumping time and money into online advertising like everyone else.

of course facebook, google, and amazon are not the people putting others in jail (although they frequently provide the apis and hourly-billed server time). these companies are making tons of money because they tried selling advertising "credits" and "microdollars" and the world believed them. what's more american than tom sawyer's story?

the adtech industry's real product is the fantasy that these are magical methods that lead to objectivity, speed, or that they simply achieve anything different at all. this fantasy is what inspires the signing of contracts to buy the "money laundering for responsibility" that seemed worked so well for silicon valley so far.