Paper presented in Urbino (Italy) RC51 Conference, 2019 (background research) Your daughter is the devil.

You might be wondering what this is all about, and I’ll give you some background information in order to set a couple of coordinates with regards to this artistic research endeavour.

This is a preliminary set of observations which is going to be developed into an artwork at the end of this year, and to which the performance of yesterday is directly connected. I’ve been moving between two different layers for this research: the user perspective on the Streaming Apps, and the background metadata editor perspective behind some of the many black boxes of the metadata (which doesn’t imply that I will disclose sensitive information gathered during my employment years, but just my artistic perspective and reaction to it).

The background metadata editor perspective implies first-hand insights on: - organisation of content - algorithms - recommendation and curation tools - capitalising on data collection

User perspective implies observations on: - taste change - memory replacement - nostalgia triggers - market-driven choices and obsessions

The professional point of view is looking critically at the streaming platform that looks at me as a user and consumer.

Rewind of some years and you can find me in my room, in my hometown in Italy, after school, throwing my backpack on the floor and reaching for the cassettes and CDs on my desk, or switching the TV on MTV to catch the latest music videos. What follows is a kid singing and screaming in her room on the notes of her favourite . Once my mother came back home and knocked on my door, to report something that had just happened on her way back from work: the son of my neighbours met her on the staircase - he was going downstairs while she was walking upstairs - and, with an exhausted tone and face he just told her: “Your daughter doesn’t stop singing. I can’t study for my exams. Your daughter is the devil”. I felt a bit guilty for this - also because my mother scolded me for annoying the neighbour’s son, but the lure and the seduction of the MTV-powered pop music was stronger than any comment or complaint. I just went on singing. Inexhaustibly. We can fast-forward again of a few years, and I’m sitting at my office desk, almost 2 years ago, and a colleague just rolls on her chair towards my desk, to tell me about this new personalised playlist released by Spotify: “Hey! Check this Time Capsule thing out, it’s amazing! IT IS SOOOO MEEE! It’s really crazy how they do this, please check it out, let me know what you got in there”. I was very busy with a thorny artist to classify at that moment and I didn’t check it right away. Some time later another colleague invites me to a dinner party, and as soon as I sit on her couch she screams of joy “OMG I am so happy, I have the perfect playlist for this dinner, it’s my Spotify Time Capsule, haven’t you checked it yet?”. Of course my answer was no, I always had problems with herding practices, but I eventually, with a remarkable delay (2 months in internet time), did check it out, and I started listening.

The Spotify’s Time Capsule is supposed to detect your past taste in music according to your current (“adult”) listening habits, collecting songs that were supposedly your obsessions or your favourite hits during your teenage years. The first listen to my own Time Capsule generated an immediate reaction of wonder, because indeed I could recognise some songs that were totally part of my teenage years. Hence, I did have a A-ha moment, in which I did recognise songs and tunes that were part of the sonic media sphere I used to inhabit, so I felt very content with it. Content. Happy. Satisfied. I felt seen. How do you know, Spotify? When I find something catchy, or something I can back up with a narrative - this is it, most of the times, I need a narrative - I obsess. So I started listening again, in an obsessive way. The more I listened the more I became suspicious. I realised that what I was listening to was a mixture of my actual taste, mixed with the taste artificially created by my use of Spotify at work (as a music specialist editor for the Italian market) together with other algorithmic assumptions.

Somebody wrote in an article that Spotify is one of the few “harmless” algorithms we shouldn’t really worry about. I read that article the first time with a sigh of relief. Then I thought: wait a minute, is it really harmless? But more on that later.

Back to the Time Capsule as an object of consumption: I did an effort to actually try and compare actual memories with the baits that were inhabiting that playlist. I felt like navigating in the artificial puberty and adolescence that Spotify had algorithmically crafted for me, and I figured that I had to start thinking about this more intensely, or at least to do something and react somehow to this devilish concoction prepared especially for the devil.

My first exercise was to subject myself to a process of earworming. For something like two months and a half I listened ONLY to this specific playlist. I listened to it while biking, while taking showers, while buying groceries, I listened to it on the plane (I have travel narcolepsy, and I did fly to South Africa and back, a total of 24 hours, sleeping with my earphones on, the Time Capsule feeding my brain Nostalgic Algorithmic Muzak on REM phase). What happened in these two months and a half was a complete detachment from the actual memories I had of some of the music in the playlist. I couldn’t remember clearly anymore the track lists of I DID touch and consume on physical supports like cassettes and CDs, whose booklets I read avidly and repeatedly, trying to grasp the meaning of Spice Girls’ (if you ask me, I did start English classes with a private teacher only because I wanted to understand what Wannabe was all about - you couldn’t learn it in school). All of a sudden the only track list that existed was the Time Capsule playlist, no booklets, no rewinding cassette tapes to your favourite track..I was left only with touchscreen skip or shuffle selections.

I hope you enjoyed this little trip in my own past. I won’t linger on this anymore and I’ll get to the hard core of all of it: WHY did I do all of this? WHY am I still doing this?

The reflection on the streaming platforms is a research which is underlying two different projects. One is a forthcoming project titled exactly Your Daughter Is The Devil, the other one is a public art project that started in April and will end in August, on which I am working as Noiserr, and it is titled Schulterplayy.

The two projects are about algorithmic listening and listening to algorithms, I am observing the platforms I use, and that my peers use, and which are able to shape taste, trends but most importantly they shape our notion of “self” and how this is affected by how we let the algorithm influence us and convince us of the truthfulness of its assumptions. This is my main concern and the challenge I want to take on while doing field research, observations, and theoretical research for my art projects.

The algorithms provide a new way of looking and thinking about human beings. It is something that Eran Fisher and Yoav Mehozay two scholars from Israel (from the Open University of Israel and the University of Haifa) discuss in a recent paper titled “How algorithms see their audience: media epistemes and the changing conception of the individual”. They highlight how “digital media are particularly suited to be a hub for the rise of an algorithmic episteme, for at least three reasons. First, being interactive, digital media are sites where big data, particularly data representing users’ behaviour, is gathered and compiled on a continuous basis. Digital media platforms are specifically designed to be able to monitor such large quantities and diverse types of data (Zuboff, 2015). Second, the big promise of digital media is personalisation, or mass- customisation of content. Digital media sites are therefore heavily invested in creating knowledge about users from their data in order to deliver on the promise of personalisation as effectively as possible (Bruns, 2008). And third, as digital media are free for users, their business model is based predominantly on their ability to commodify data by turning it into knowledge and selling it to content producers and advertisers (Fisher, 2018).”

Fisher and Mezohay analyse the notion of the self within the framework of the digital episteme by comparing it to the notion of the self that emerged from the analysis of audiences in the past, first assuming universal persons (basically white western upper and middle-class males), then using scientific methods (social theories, empirical research, and samples) to ascriptively assign individuals to categories that could be sociologically and culturally characterised. The latter is the era of the mass media, where “the audience is comprised of different but internally homogeneous, social categories. It then tailored content for each category, based either on members’ presumed interests or on the demands of advertisers.” The audience was somehow constructed from the vantage point and interest of institutions and marketers. With the rise of the digital media, the episteme that emerges is not a continuation or an improvement of the previous one, but it’s something entirely new.

We are immersed in a complex apparatus designed to translate user-generated data (a lot of data, big data) into knowledge about users, and in which the conceptualisation of the human being becomes a performative one, because the machines detect behaviours and they assume what a human being is, according to those very behaviours. This means that the knowledge about users can be hyper individualised because assumed from the registered behaviours of a specific user. If we think of how the users move through the digital, which is mainly through two types of actions/activity: Search or Discovery (the Search being the user actively and actually looking for something - the Discovery being the serendipity of finding something on the platform that one is using) we can say that media platforms want to use the knowledge about their users for targeted advertising which makes use of serendipity. This brings us closer to some of the main features that made Spotify the colossus it is now: the serendipitous quality of periodical playlists such as Discover Weekly or New Music Friday, the personalised products such as the Time Capsule, as well as Daily Mixes organised specifically for the user according to the listening habits and grouped by genre, and Release Radar keeping you up to date about latest releases according to your preferences.

How does Spotify work?

Spotify can reach its serendipitous and personalised goals mostly by combining three different approaches (disclaimer: in all companies there is much more going on than what is publicly disclosed, and I am sure these three approaches are not the only ones implemented by Spotify to power its products):

1) natural language processing 2) audio analysis 3) collaborative filtering

NLP (Natural Language Processing) is a branch of artificial intelligence where a computer is trained to understand “human” language, and in the case of Spotify it searches the web for text content related to artists, and determines similarities and differences between songs and artists, which means that the similar artists and songs will most probably be recommended together or appear in the same playlist.

The Audio Analysis is the analysis of audio files through convolutional neural networks (a kind of deep learning applied to recommender systems) which allows music recognition and comparison, according to filters which identify tempo, use of music keys, mood, loudness and so on.

Collaborative filtering is the algorithm that interests me the most, because it triggered a lot of metaphoric thinking in my mind and I was imagining transpositions and translations of it in the real world (one of these is exactly my project Schulterplahyy). In Collaborative Filtering Spotify recommends music from listeners with similar tastes, drawing information from its database where each listener and the related music habits are stored. Collaborative Filtering doesn’t analyse at all how the music sounds like, or which kind of news/comments or lyrics are related to it. It assumes that the cross-reference of the preferences of listeners with similar taste will lead to the selection of music that would not be too far from the preferred genres of the listener to whom the recommendation is directed to.

This database grid where the Collaborative Filtering algorithm operates, together with a bit of Natural Language Processing, a bit of Audio Analysis - and who knows what other kind of data that Spotify is able to collect or outsource, make the magic happen…together with of course human curation for specific Spotify products.

But we’re not looking at the human agency here, we want to be dystopian, and I assure you that the goal of the industry is the FULL automation in certain specific sectors, so I’m not guessing anything too far from reality here.

What does all of this imply?

I am very fascinated by Spotify, but I want to go back to the critical gaze towards the two different points of view that look at the algorithm that looks at me.

I said previously that some journalist declared how harmless the Spotify algorithms are and how we shouldn’t really worry and panic about what Spotify is doing in the music industry. Shut up and stream, basically. There are a lot of implications in the simple act of streaming and using the Spotify platform. A flattening, I would say, but something not entirely uninteresting as well.

The core of my presentation here is the notion of the self that emerges from the use of Spotify, but the use of the platform ramifies further, affecting the market logics of production, distribution, and remuneration of artists, which I will mention in a bit, together with the implications of the streaming logic itself in political and social agency.

As a user in the data grid that Spotify uses for collaborative filtering, we appear as quantified Selves. To know an audience nowadays means knowing its digitally registered behaviour - meaning the objective digital footprint of a user. This is a performative behaviour, which can be conscious or unconscious; the role of the machines is to collect and group together actions into patterns of behaviour, in order to predict. The predictive aspect is the main goal of nowadays’ algorithms. They don’t analyse, they don’t do qualitative work, they predict and anticipate, they try to detect whatever consumption you might be interested in, in the future. There is a big chunk of processes that cannot be datafied, and that are excluded from the way the machines look at us, and the notion of the self that emerges from this algorithmic “gaze”. “Reflexive and narrativistic models of knowing need to be overcome in order to get to the building blocks of the self: a heavy load of data points representing discrete digital events.” All the digitalised events we create while using certain platforms do not create an entire image of a self, but rather they generate blocks of a self. They represent parts of what we do while we are connected. Machines cannot interpret psychologically or sociologically what we do on the internet, and what we do cannot put us into specific a priori categories responding to specific variables because our behaviour on digital platforms allow our image to be constantly changing when our patterns change, and these changes are based on very mundane aspects of “humanness”. All of this though doesn’t really offer an image of what a human being is, but rather helps the construction of the notion of the human being itself within the parameters of the algorithmic episteme. Most of the times we don’t know how algorithms work, but we know we are watched, and the way we respond to this is to develop an “algorithmic imagination”, considering the content we are offered as an indication of how we are seen, a reflection of ourselves.

How can we respond to this flat image reflected into the black mirror of our Spotify accounts? How can we intervene creatively and develop processes of consciousness and an epistemology of the self through the digital and through the reflection of something as flat as a digital user?

I say creatively because that’s the position I occupy, but it is also because this kind of research could and maybe should gain a speculative momentum, since the content we might have to look into and investigate is proprietary content, it’s money, it’s profit, it’s capital, and will never be shared, not even with scholars scientifically interested in these processes and phenomena. I want to touch now a few points that render, according to my own observation, the complexity which is behind a very simple action such as streaming music on Spotify.

In a very interesting article published on The Baffler in December 2017, Liz Pelly discusses the automation of simple tasks like paying at the cashier, usually accompanied by a person hired to teach customers how to use a technological device that will eventually render her presence useless. When we all know that automating that kind of task is not really central to our lives. The author then goes on drawing a parallel with Spotify, a platform which for her is replacing/creating something in the music industry which is not answering any specific and urgent need of the industry itself.

We can start from the playlist logic, which is destroying the music as a craft and a product, and is changing the way artists look at their own artistic process - I am thinking here of Drake and his More Life, which was advertised as a playlist rather than an album. Playlists also induced a new kind of trend, a kind of algorithmic version of Muzak, which is the lean back listening, the inattentive listening, the emotional wallpaper music listening. We don’t listen to music. We listen to music while we do things. In times in which we think we don’t have enough time to do all the things we think we are supposed to do, we are overworked, we are anxious, hence we delegate to Spotify the task of choosing the music for us, which we don’t end up listening to in dedicated times, but rather while we are busy with something else. We are always busy.

Another issue Pelly analyses is in relation to branded playlists, which don’t grant the few indie artists that end up there to be paid fairly, meaning: a bit more than the royalties Spotify actually pay to artists. Pelly defined this the “automation of selling out”, minus the part where the artist gets actually paid. Indie artists, even when they do know what is at stake in using Spotify as a distribution platform, upload their music on it anyway, because that’s most of the times the only way of “making it”, the only way of existing in the world and being noticed (just think of the fact that Spotify just reached 100 million paying users, and has a total of 217 million monthly users, including the free subscribers; and it aims at increasing of 10 millions the number of paying subscribers and of 9 to 11 millions the number of free subscribers within a few months. Apple Music in the meantime is left behind with a total of 50 millions paying subscribers.) All labels are rethinking their workflows according to the streaming logics of Spotify. Most of the playlists are populated by artists produced by major labels (or sub labels of these majors), which are a priority for Spotify, while indie labels hope for their artists to be inserted in some trendy playlist at some point. Moreover, the royalty payment method that Spotify implements is a “pro-rata” one, meaning that “artists are paid a percentage of the total pool of royalties relative to how their stream count stacks up in the entire pool of streams, meaning the tiniest of payouts for most independent ”. Here's the math: Spotify pays about $0.006 to $0.0084 per stream to the holder of music rights. And the "holder" can be split among the record label, producers, artists, and songwriters. In short, streaming is a volume game. Despite the very low royalty rates, Spotify also started inserting into extremely popular playlists (with no less than 1 million followers) fake artists that are created in-house (by paying flat fees to producers to compose those very tracks), allowing to save a huge amount of money to be paid for songs that racked up millions of streams, and whose master copyright is controlled by the company itself. Spotify has started also producing original content to be released on its own platform, like the “I’m with the banned” series, and it won’t take long until the partnership with Vinyl Me Please will lead to the release of Spotify Singles on 7” records. In the meantime Spotify is working on a better model to produce and distribute podcasts on its platform, whose first updates were launched recently. While trying to survive and cope with the presence of Spotify, artists and labels might be working towards their own obsolescence and replacement in the near future, Pelly claims. I won’t say here that I agree entirely with the forecast drafted by Pelly, but my reflection aims at figuring out what kind of agency we can carve out of this capitalist monolith, without claiming that all was good and better in the past.

My interest here is nots only in relation to the music industry, but I want to link back to the notion of the self and its political and social implications. What I mentioned above has serious implications of taste change and preferences through addiction, within the frameworks of the idea of “renting” a humongous amount of music instead of buying the music we really like, flattening the notion we have of ourselves while also flattening our agency and capacity to choose. We learn to live without “owning” any music but also without owning the autonomy of choice, favouring the algorithmic and privatised serendipitous instead of empowering our search skills: this would imply critical judgement, non-algorithmically individualised taste and preferences, a community around us which can give us advice, support us, and favour a different kind of serendipity, instead of a community of strangers inhabiting with their stored clicks a database grid for collaborative filtering.

I want to use some references in the closing of my presentation: Kittler on the one hand, and the notion of the military-entertainment complex, and Foucault on the other, with his idea of power as something which is everywhere, it is dispersed and pervasive. In an analysis of Kittler’s military-entertainment complex, Steve Goodman highlighted how the difference between a technology designed for military purposes, and its conversion into civilian use, is the lack of what can be defined as “talk-back capability”. This means that a device initially designed for a two-way use and interaction, ends up being commercially distributed for one-way use only. This process transforms it, in my eyes, into a tool for hegemony and coercion. If we apply the Foucauldian thought to a one-way technological device, supposedly one on which we can run a Spotify app in this case, we can see how power, as a source of social discipline and conformity, could reinforce itself through the learned behaviour of trusting the machine to work for our benefits, and believing the image it gives back to us after recognising our digital footprint patterns. All that I just mentioned could be tied then as well to what Attali discusses in “Noise - The Political Economy of Music”, describing the cycles of production and distribution of music in society as mirrors of power structures, and noise and changes in music as heralds of future systemic reactions and adaptations, returns to order, hence modes of exercising power. Attali makes a systemic overview of the structures that supported the music industry in its historical development, and the hegemonic scheme is described in a top down fashion to then trickle down to an atomised agent of resistance which he recognised ultimately in the , as the only possibility to escape capitalist logics of production (I am simplifying here). I want to actually mention Attali as a reference, because what I did was to start from the user, the individual, and granularly infiltrate into some of the systemic implications of just one act of consumption, emanating from a flattened image of the self with a reduced political agency or will to act, reinforced by learned behaviours such as delegating the act of choosing what music to listen to, or which songs should be representative of your teenage years.

Looking at the algorithms that looked at me, while I was looking at them from behind one of the black boxes, while also looking back at myself as an average user, resulted in a critical reflection. In this reflection I figured how Spotify’s algorithms are not really harmless. Steering the preference of a user towards a specific genre, persuading the masses into obsessing over viral tracks, pushing the listening habits to the consumption of emotional wallpaper music compiled in playlists in which half of the artists might even be fake (and we tend not to realise it autonomously - because of the way we are conditioned to use the platform), and ultimately believing that the algorithm is so accurate that it can not only guess what new music you might like but also deduce what you were listening as a teenager, all are part, in my opinion, of a big hegemonic scheme, with big implications on awareness and self reflection and consciousness, consequences in the market and labor conditions, and so on and so forth.

What is being triggered by Spotify, with the cooptation of playlists by big labels and market interests as we previously described, is not only a progressive flattening of the self, but a flattening of taste, where eventually the more the users listen to the viral tracks, the privatised playlists, and promoted content as well as fake content, the more the algorithm will source from pools of very similar data in the collaborative filtering process.

I want to mention here briefly a text that Henrique Iwao, a Brazilian and philosopher from Belo Horizonte, wrote on his label’s blog. I hope I can render properly into the translation from Portuguese what he wrote, but overall the text is a comment on the latest anime series by Shinichiro Watanabe: Carol & Tuesday. The anime is set on a terraformed Mars, where basically 99% of music hits are made by artificial intelligence. Henrique describes what happens in the anime: “with a sufficiently large database, careful sampling and intelligent numerical analysis, it is possible to compose the success that will thrill the population. Even the expected commotion is parametrisable by simply adjusting its factors.” Carole & Tuesday apparently are part of that 1% of population capable of producing music out of real empathic encounters, becoming a social media case when they break into a theatre to check, out of pure curiosity, what a grand piano sounds like. The music Carole & Tuesday play though is something very conventional, and this is what Iwao underlines critically, it is music that might have as well been composed by an artificial intelligence. Hence his conclusion is that the 1% C&T represent has an antagonism with a 99%; on the background of both there is the ideology of the artificially prefabricated. Quoting Iwao “Even if Carole & Tuesdays compose, as characters, hits by themselves, their imagination is only a mirror of what the cultural industry and its artificial intelligences are constantly producing. As humans, they only fill a marginal space in a mode of production that would already be entirely, 100% taken by forms of retroactive manipulation. They can make up what they want, what they feel to be important and meaningful. But the reverb, the production values in the scene in the theatre indicate: what they want is what was demanded and what they feel is what they were induced to feel.”

This is why my current obsession over this specific product is being translated into different projects that highlight different points of my ongoing reflection.

A performance restituting the multiplication of a digital image of the self in a vicious circle of narcissistic self-reference flattened along the notes of one single playlist; a future artwork, a contradictory and absurd product impossible to commercialise, containing the performance of the adolescent Spotify thinks I was; and a project in which I am trying to replicate in an area of Hamburg an analog version of the collaborative-filtering algorithm, while condensing the interviewed population into a virtual digital user, reflecting the varied taste of an area which lost its identity due to gentrification processes, and which might find itself again in the digital.

I don’t have any specific conclusion here, and I thank you for listening.