A new iconoclasm or post-modern pastiche: is AI hydrating an internet that's f*cking boring?

This article is the second in a series in which I'll try to articulate a lingering undercurrent of unease I experience in the digital industry for quite a while now: despite (or because of?) the almost endless variety and zero-entry availability of tools, assets, frameworks, methods, insights, qualified people and in particular generative capabilities, why does so much produced feels so similar? Is it my own insular mindset? Simply the cynical by-effect of ageing? Or are we actually tilling a digital monoculture, void of serendipity and authenticity? Void of, you know, fun?

Episode 2: the audience enters the stage

Taking a piss at art

In the spring of 1917, for the inaugural exhibition of the Society of Independent Artists in New York, Marcel Duchamp placed a porcelain urinal on a pedestal he allegedly purchased in a plumbing fixtures outlet, signed it "R. Mutt" and titled his entry Fountain. Its reception? Outrage, confusion, and a prompt rejection by the Society's board, despite Duchamp himself being a founding member and the Society's claim that they would accept any piece of art. But the urinal, was it art, or merely a prank? Could a prank be art? The piece, in its blunt utility, did more than poke fun at convention - it punctured the very idea of what constitutes art, challenging the authority of institutional gatekeeping through a denial of definitions, upending association and classification.

His admission's rejection made Duchamp resign from the board in protest, and in that rupture, conceptual art found its genesis. And until today, art historians continue to debate the validity of Fountain and attempt to classify Duchamp as an early modernist or proto-postmodernist (or, more interestingly, as another case of male artists getting attribution for female creative brilliance). Regardless, Fountain was voted the most influential piece of art of the 20th century in 2004, a fitting title for a piece that continues to unsettle as much as it inspires.

Art is what you can get away with

While Duchamp's radicalism challenged the essence of art, Andy Warhol took aim at the very role of the artist itself. In 1962, as pop art melded with the undercurrents of postmodernism, he seized upon the freshly departed image of Hollywood's most famous star, Marilyn Monroe, and elevated her status into that of a cultural icon by turning her image into a reproducible, mass-market commodity. Using silk-screen printing - a process chosen for its efficiency rather than its artistry - Warhol pressed Monroe’s face across canvas after canvas, flattening her identity into a bright, lurid assembly line of celebrity, fitting for a can of soup. The image’s vivid colours jarred against the stark reality of her tragic end, a haunting indictment of fame’s commodification. Here too, the boundaries of fine art were tested: what once was institutional and exclusive now shamelessly engaged with pop culture, tainting itself to become cheapened, repetitive, a consumable.

Art critique's response was, predictably, polarised: some celebrated his elevation of pop culture into art and the aesthetic innovation applied, others scoffed his commercially exploited cynicism and 'anti-art'. Warhol himself carefully remained as oblique as superficial on his intent, even purposefully distancing himself from his works' production:

Why don’t you ask my assistant Gerry Malanga some questions? He did a lot of my paintings.”

In Warhol's view, fame, not form or craft determined relevance. Anyone could become an artist - or even pretend to be one - giving rise to the power of the audience and their attention as the currency at trade. The relentless rhythm of repetition and variance in Monroe’s visage echoed the diversity of the masses' preferences; there should be, after all, a version of her to anyone's liking.

Giving spirit to his fascination with commoditisation and consumerism, with the unique and the mundane, but above all else with fame and mass production, Warhol catapulted himself to artists' tier one status. A rise dripping with irony, as his work mirrored the very culture it critiqued, the artist becoming art himself: Shot Sage Blue Marilyn - most probably produced by Gerry Malanga - became one of the most expensive works of 20th-century art, reaching $195mn at an auction at Christie's in New York in the spring of 2022.

Art (f)or the people

This deliberate distancing of articulated intent - at least on the surface - and leaving it to the viewer was taken up a notch a few decades later. In 1994, two US-based conceptual artists Vitaly Komar and Alexander Melamid - both Russian émigrés - took the rather atypical step of commissioning a public research polling firm for their next work. Its brief: to find out what Americans desire most in a work of art. Over a thousand US citizens were asked about their favourite colours, brush strokes, compositions, scenery and meaning in a painting, all intended to populate a dataset of preferences and dislikes.

Promoting this dataset to a fictitious client that's commissioning a piece of work, Komar and Melamid went out to paint a pair of canvases, labeled The Most Wanted and The Least Wanted; each painting including the aggregate of what respondents said they wanted or did not want in a painting. They repeated this approach throughout a patchwork of countries across the globe, expecting to capture each country's unique collaboration and culture in distinct visual expressions. The outcome? In practically every country all what people really wanted was a figurative composition that depicted a landscape with a few people and some animals in the foreground against a blue sky. Hardly any cultural or geographical distinction surfaced. Thus, after finishing the series, Komar noted:

"We have been travelling to different countries, engaging in dull negotiations with representatives of polling companies, raising money for further polls, receiving more or less the same results, and painting more or less the same blue landscapes. Looking for freedom, we found slavery."

Of course the art Komar & Melamid were after were not the paintings itself; their creation concerned the process, and their own comments on its outcome, producing non-art and art at the same time. Looking for self-agency and individualism, they laid bare conventionalism and cliché. Lace that with Warhol's prophetic vision of a hyper-saturated consumer space and the socio-technical challenge of curation at scale is slowly taking shape.

How metrics became our muses

Andy Warhol never got to experience the internet. To comment on the instant cultural transmission of icons through technology, built for duplication and distribution, paid for with our attention. The algorithmic reach to millions at near-zero cost. People as brands, their instant yet transient fame. David Bowie did witness this alien life form he named the internet land on earth, and in a 1999 BBC interview augured that rock 'n roll's torch of carrying subversion was passed on to it. A communal medium, a conveyor of rebellion, giving context to content as a construction for the interplay between artist and audience, between user and provider; anything but a space for (corporate) monopolies and redaction, with creation and curation very much a human exchange of value and feedback. The manifestation of Duchamp's seeds and Warhol's roots that "a piece of art is not finished until the audience comes to it" (Bowie, 1999).

Bowie's envisioned cultural shift to digital interaction materialised as we all know now, as the internet created a new space where people interacted, exchanged opinions and built virtual communities. Digital spaces became topical, and in their organisation started to reflect a reliance on peer networks for decision-making, even identity. Sociology and behavioural studies grounded the cultural understanding that people organise their social arrangements by others' behaviours and preferences, captured in the concept of networked individualism. Commercial trade entered the digital space through e-commerce, and Amazon enabled people to connect over their shared purchasing interests; laying the foundations of becoming the first hyper-scaler, they birthed an omen of things to come.

As general wealth and disposable incomes increased, consumer culture began to emphasise choice and personalisation. Products and services tailored to individual preferences became a symbol of technological advancement. Customer centricity became a primary strategy for customer satisfaction, marketed through the acknowledgment that people are different. Yet the way these differences manifest themselves we've learned follow predictable patterns, patterns that - at least at first glance - are surprisingly straightforward to grasp in predictive models, granted their computational feasibility made economic sense. And thus, Collaborative Filtering (CF), the practice of serving automatic predictions on a users' interests by leveraging preferences or taste information collected from others to power recommender systems, emerged at the intersection of technical innovation, the cultural practices of trust and influence, and the evolving needs of a digital society.

A global RMSE race

An important catalyst to the improvement of these CF algorithms concerned the infamous Netflix prize. An open competition issued by Netflix in 2006 (back when they were still shipping DVDs!), it challenged competitors to beat Netflix's own recommendation engine Cinematch by at least 10% on RMSE, granting the winner a 1.000.000 dollar prize; with commercial data science still in its infancy, a hefty reward. The community's interest was wild; less than a year into the competition, 20.000 teams had registered themselves, submitting over 13.000 prediction sets on the anonymised training dataset provided. To keep appetite alive, Netflix granted yearly progress prices, until three years in two teams had managed to beat Cinematch by 10,10% in a last call round, one team trumping the other by submitting their winning algorithm a mere 20 minutes earlier.

The competition's impact was profound. It not only ignited the fields of data science and machine learning, with breakthrough progress on the refinement of matrix factorisation - vastly improving the scalability and accuracy of recommender systems, especially with sparse data available - but it also promoted open innovation, distributed collaboration and transparency within data science and engineering. Equally impactful, it underscored the business value of personalised recommendations, a tactic extremely valuable for companies that rely on user engagement and retention for their growth.

From there on, recommenders quickly became the de facto agents of guidance in our digital decision-making, gaining increasing sophistication and complexity, through the addition of methods like content-based and contextual filtering, in an effort to capture or synthesise as much environmental context as possibly available. Within entertainment & media, e-commerce & retail, travel & hospitality, online education, journalism, social media & networking, all businesses built their digital platforms with these recommender capabilities at their core, curating content to hydrate our scarce screen estate and time with.

How moneyball has changed the game

The thing is though, algorithmic recommendations, and the ways we interact with them, have compressed the intricate art of taste—once a nuanced, ever-evolving judgment on aesthetic and artistic matters—into a tidy sequence of quantifiable data points. No matter how sophisticated these systems may be, capable of parsing the richness of context, they inevitably flatten complexity into something that determines a position on a screen, awaiting a yes/no verdict. Because the sheer abundance of this content is diametrically positioned against the surgical precision that is required to capture one's attention in context, our interactions with these recommendations enter feedback loops that favour the lowest common denominator forms of content, skewing distribution into a small set of winners and endless graveyards of losers. This zero-sum bias is creating a kind of self-reinforcing blandness and cultural homogeneity that is getting increasingly difficult to circumvent, as it influences the way culture is distributed and consumed.

Yet perhaps more consequential than the mechanics of these recommenders is their mere existence, which has cemented their role as the ultimate gatekeepers of culture, arbiters of what we see, hear, and consume. Much like the quantitative revolution has impacted sports, this data-driven optimisation is shaping the contours of a gaming mechanic that makes it increasingly difficult for originality to make it past these gatekeepers, as they force the creators of movies, books, games, music, fashion, anything to adapt to the logic and pressures of the algorithmic system. Appeal to as many people as possible, quickly.

One could argue that ironically, we, the audience, have called this blandness upon ourselves, by favouring more of what we know or understand. Detaching our eyeballs from our screens for a moment and gazing into the distance reveals a bigger picture perhaps, one in which we are witnessing a tectonic shift from 200.000 years of human creation and curation (remember van Heemskerck?) through a mere two decades of human creation and algorithmic curation towards an algorithmic machine of both curation and creation, pushing us back into the role of a passive audience.

David Bowie did not foresee - despite its best architectural intentions - the asymmetric nature the internet would develop into, nor its omnipresence. Neither him or Warhol could have fathomed we'd consume on average 208 ads an hour now, served by a handful of ad networks, hydrated by a similar small number of global advertising agencies, capitalising on our dormant desire for rapid - if not instant - gratification. It is exactly this asymmetry that I believe lies at the core of these challenges, which I'll further explore in the next episode. Also, interestingly enough, if we look at Komar and Melamid's The Least Wanted, they carry much more diversity and culturally attached preferences; perhaps we're asking the wrong questions to be fed into these systems?