The Sad Choice between Corporate Jazz and Algorithmic Racism
Navigating social cues vs. tackling black boxes
When SADE scraped together the last 5 cents to record their first 2-track demo, which consisted of the now well known Smooth Operator and Your Love is King, the record labels hated it. ‘Too long, too jazzy.’ SADE didn’t give up and instead, they tried to hype themselves up around London using lead singer Adu Sade’s very own fashion connections, including, among others, Jean Paul Gaultier.
They got on the cover of a trendy magazine THE FACE and performed in a small sought-after nightclub HEAVEN with all of London’s music press attending. Over night, the record labels made a u-turn, and suddenly everyone wanted to sign SADE.
Few other bands have become so effortlessly famous, but SADE’s success was also understandable. They played to the sensibilities of their age, the yuppie 1980s. Despite being dirt poor, their music embodied moneyed elegance and upper class refinement, sometimes vulgarly labeled ‘corporate jazz’ by the press. This may have not diminished the value of SADE’s music, but it certainly spoke to their ability to rhyme with the age.
Seems to me that every single new thing that is in any way above par has to suffer through this initial ‘no.’ that requires reinvention and confidence that absolute majority of people never find in themselves.
Worse, it demands this horrible talent for navigating social cues, winning favors and being comfortable to stay in personal debt to others for them. This many can’t do, or they can train themselves to do it, but only by killing part of themselves, stupefying their nature with an artificial visage of a salesman.
This then, you could say, leads to sad outcomes because things succeed based on social structures and coincidence at least as much as their quality? To do anything, from music and art to business, science, … anything, you have to belong, be a part of a human group, be liked and like, exchange favors and depend on them.
Platform.
Except no! you don’t because we now live in an era of platforms, when you can sell your work directly to your audience, independent of publisher’s judgment, unaided and free, subject only to the platform’s algorithm. Oh, the algorithm.
Replacing the human system of coincidental favor exchange by a cold machine code may be a progress (look not everyone likes people), but only as far as it takes us to a lesser form of inequality. The bias and coincidence is replaced by a profit function.
A platform will only give exposure to your work if it makes the platform money. That is to say, engaging things keep paying subscribers and/or bring ad money.
But, you know, over time this could lead to algorithmic racism of sorts because 1. users with most followers get all the spotlight, 2. selection is left to popular vote, so if you strive for less utilitarian (guides, advice, reviews), god forbid culturally valuable content, you may be forgotten, but hey who cares, 3. it is quantity over quality, and 4. the platform itself becomes a self-propelling vicious system, where the most engaging content consists of people giving advice about how to create content on the platform to their followers who want to get on top of the pyramid.
Black Box.
See this is how the mighty algorithm works: I’ve spent good 60% of the text on a pseudo-intellectual tirade, and so now, scared of the all-seeing algo, I feel the need to add something informative (or at least marginally useful) to the text.
But it is actually interesting to ask questions about what the algorithm really is. Thanks to Elon’s sloppy takeover, one that gets all the attention is Twitter, of course. There are many guides explaining Twitter algorithm, mostly because people want to game it. But what is it really under the hood? How does it look and what is its architecture?
Every time you open your twitter feed, the algorithm gives every single tweet of the people you follow a ranking based on:
the tweet’s own quality (recency, if it includes media such as video, etc.);
who tweeted it (how much you interacted with the author’s tweets before);
what you found interesting in past.
So, to vastly oversimplify, every time you open Twitter, a function is called that multiplies every tweet with a factor y = x (the tweet’s quality) + y (the tweet’s author) + z (you). Of course, how Twitter determines x, y, z (and maybe some additional variables) is a black box and changes every week. It is also rapidly evolving because Twitter uses machine learning, which is itself a fast changing and improving field.
Simply put, machine learning means writing a function that is supposed to estimate an outcome. For example a function that can recognize that a photo of a dog is a dog. A photo is only a group of pixels, which are only 0s and 1s. The function knows that for the picture to be a dog, the pixels (0s and 1s) in certain parts of the photo (pixel field) must have a certain value, i.e. either 0 or 1.
So, again for the sake of sanity imagine some a + b + c + d + …, weights that each unit of a pixel field is given (in order to figure out if the picture is a dog), then feeding it with inputs and observing results. Each time the function correctly guesses that it is a dog, you alter the a, b, c, d, … so as to strengthen it in that direction, and every time it misses, you change them to move it in the other direction. It is, of course, not a linear function, and I am vastly oversimplifying to make a point here.
Twitter’s machine learning models deal with vastly more complex data than pictures. Every single interaction (from tweeting to just scrolling and clicking) of every single user is a data point in an immense web of interconnections, which can be transformed into such functions as described above in various ways, and some of those methods are still only evolving (e.g. graph machine learning).
For less idiotic reading on this topic, you will find Twitter’s own researchers blogging interesting here and here. For more on machine learning, refer to the excellent 3Blue1Brown.