Social media algorithms were the first unaligned AIs. They were designed to paperclip for attention, regardless of where that attention was directed. They were deployed at scale without safeguards. Their results are all around us. We do not need to imagine what an unaligned AI scenario might look like, we are already living in a post-apocalyptic culture. The “real world” ended in 2012, as the Mayans predicted.
AI gets blamed for what the algorithm did: flooding feeds with mass produced slop. Yet before AI flooded Amazon with ChatGPT books, Spotify with ghost artists, and YouTube with automated AI videos, there were human equivalents driven by platform incentives. Kindle Unlimited paid per page read, Spotify paid per stream, and YouTube paid for watchtime.
While these incentives might not seem harmful—shouldn’t artists be paid based on how many see their work?—quality requires time. A six-minute animation created over months pays less than an hour-long talking-head podcast streamed in an afternoon. When YouTube switched from rewarding views to watchtime in 2012 it killed animation on the platform. Hundreds of choices like this, which have nothing to do with the taste of audiences, have shaped our culture.
What is odd about this cultural apocalypse is that the prior system was functional and superior. Before streaming, the internet democratized direct transactions. You could get any song for $0.99, any movie for a $3.99 rental, and any book for a $9.99 download. Artists could sell their work on those platforms. Rather than an algorithm, money and time were the quality filters.
Streaming claimed to offer a better deal. Rather than paying for a single work of art, consumers could pay a monthly fee for a full catalog. This is a better deal the way that a buffet is “better” than fine dining. While streaming might offer more product, the incentive shifted from each offering requiring quality, to having a few “must-watch” shows that bring subscribers in the door, padded with slop, since overall quality does not increase revenue.
Algorithmic platforms worsened these incentives. Now, the consumer is not even selecting content, but having it fed to them. These platforms monetize through advertisers rather than subscription. The incentive is to keep users on the platform for as long as possible, regardless of the experience they are having. The most common experience of these platforms is described as “doomscrolling,” where one loses track of time and self. Doom is what one experiences during an apocalypse.
There has always been a prisoner’s dilemma between artists and audiences. Artists create great art, which requires an investment of time, skill, and labor. Audiences reward that art with attention, money, and status. Both sides must cooperate to create culture. Artists can spend years developing their craft if they know audiences will pay for it. Audiences will pay if they know that their undivided attention will be rewarded with quality. If either side defects, they might win one round and experience a temporary boon where they get quality art for free or profit from lower quality work, but the relationship degrades and the other party will remember that defection the next round.
For the past ten years, there has been escalating defection between artists and audiences. Audiences stopped paying for art and instead paid streaming platforms who paid pennies on the dollar. Artists responded to incentives by producing larger quantities of lower quality work. “Clickbait” was not a trick, but a platform mandate. Audiences defected further, by paying zero on free platforms. “AI slop” is the latest escalation, allowing artists to invest zero from their side. Now, neither side is investing anything. Neither side realized they were defecting, because they were not in relationship with each other, but with platforms and their algorithms.
Relationships have undergone a sophisticated man-in-the-middle attack. Rather than direct relationship, communication is mediated by algorithmic platforms that decide who will receive what content. The earliest social media platforms had a chronological timeline where all content was delivered in the order it was posted to those who opted-in to receiving it. These platforms allowed artists to have a direct relationship with their audiences, but they also allowed you to passively consume relationship, receiving updates about those you met in real life.
Real relationships are boring. They require presence even when the other person isn’t maximizing engagement. Because of the time required, most only have a few. Early Facebook engineers found that users who added at least ten friends in their first two weeks were more likely to stay. To hit that quota, their growth team built “People You May Know” to add weak connections. This made the feed overflow with content users didn’t care about.
There was an inflection point here. The platform could have limited follows, segmented relationships into different feeds, or given users the filtering tools, but all of these solutions might have limited growth. Instead, they chose what any cancer cell would. They introduced the algorithmic timeline. The algorithm was a solution to a real problem, but a problem they created, and their solution meant that Facebook was no longer delivering content but curating it. All future platforms followed the same.
Attention became so valuable a commodity, platforms could charge for it. Every platform followed the same process: build audience with free tools, insert itself as a man-in-the-middle, and charge for access to the audience humans built. Facebook invited business to build pages, throttled reach, and sold it back as advertising. Google went from zero ads to paid placements comprising a third of all clicks. YouTube shifted from glorified video hosting to the majority of views coming from recommendations rather than searches. X required a premium subscription for algorithmic boosting. In each case, platforms began as a bridge and became a toll booth. By the time TikTok launched, the pattern was so established, it didn’t even pretend to be social. The algorithm was the product.
Since humans were now competing for attention, they began optimizing themselves for it. The influencer was born. What separated influencer culture from the past hundred years of celebrity was their competition. Entertainers compete with other entertainers. Influencers compete with your friends. The people in our feeds were no longer people we knew, but people skilled at being known. Being known at scale, to strangers, through a screen, is performance. The influencer’s core talent is the simulation of intimacy without any of the reciprocity, commitment, or physical presence closeness requires. Platforms rewarded this because influencers created engagement. Real-life friends were still on the platform, but outperformed.
Yet the influencer was merely a waypoint on the route to human removal. Once a platform adopts an algorithmic timeline, they are no longer in the business of relationships, but content. Influencers outperform friends, but AIs outproduce both. Human influencers experience burnout. They require sleep and payment. AIs do not. Every so-called “social media” platform is looking to replace their human creators with AI-generated content. If they are not, users are actively doing it for them due to economic incentives.
Rather than providing social connection, platforms are plugging users into the matrix. They began by connecting humans to other humans, mediated those connections with an algorithm, and are slowly replacing those connections with AI-generated content. Users’ reaction has been clear: People don’t want this. When humans realize there are no other humans at the party, they go home. When there are no users, there is no one to serve ads to. The entire model collapses.
Platforms have started to realize their mistake and are actively combating bots. However, their entire business model incentivizes its own demise. Human creators cannot compete against AI on an algorithmic timeline. In the time it takes a human artist to produce one album, AI can produce hundreds. Even if the product is inferior, AI can repeatedly iterate and optimize at a speed humans cannot. Most human creators are already effectively AI-augmented due to their reliance on analytics and the way they optimize their work for algorithms.
AI is not an issue in direct transactions. If an artist produces art with AI and sell it to an informed consumer, each have made a human value judgement about the quality of the work. There is an economic and reputational incentive for the artist to produce quality and the consumer to ensure what they buy is good, whether or not AI is part of that process. The issue arises when value judgements about human creativity are made on behalf of humans by non-human intelligences.
Human consumers could avoid “AI slop” with one simple trick: pay for art. The reason AI rather than algorithms receives the blame is that the latter would shift blame to the consumer and acknowledge their cultural defection. The art you get for free will be the art that is free to produce, and no human’s time is free. Anything you consume from a feed is slop.
What consumers have done is outsourced curation to paid services. It’s not cheaper. The average streaming service is around $30 a month. For that price, one could rent eight individual movies. Most subscribers aren’t watching eight movies a month. Yet suggest a movie to someone and their first question will be what service it’s on. The revealed preference is that consumers do not want to make their own choices. They want AI to make choices for them. Making a purchase requires active choice. Streaming makes some choices for you. Scrolling makes all of them. Active participation requires friction. Algorithms ensure what is shoved in the consumer’s face is lubed.
Every other author who has critiqued algorithmic culture—Kyle Chayka’s Filterworld, Cory Doctorow’s Enshittification, Max Fisher’s The Chaos Machine, etc. lets the consumer off the hook by framing them as passive rather than active participants. They also propose active participation as the solution, not seeing the participation already occurring, and misdiagnose the solution as some form of willpower or regulation, which is willpower applied at a civilizational level.
Willpower has a 95% failure rate in dieting. What has worked is Ozempic. The solution to algorithms is not to “log off,” but giving the consumer an algorithm of their own. Rather than pitting a monkey-brain against optimized capital and fighting technology with “willpower,” consumers will win when they fight technology with technology.
The parallels are clear. Content platforms, like food manufacturers, have made their product more addictive. We have biological reasons to overeat and overstimulate. Despite a massive “wellness” industry dedicated to solving both problems, most Americans are still fat and glued to their phone.
Those who lose weight using GLPs are practicing the old-fashioned advice of “eat less.” Rather than applying willpower, they reduce craving, lowering the desire willpower acts against. An attention Ozempic might do the same. Rather than applying willpower against algorithms, they would reduce the willpower required for superior choices.
Food and art share other parallels. A healthy meal leaves you full. “Empty calories” leave one hungry. People can absentmindedly eat an entire bag of chips, but no one wants another meal after fine dining. Likewise, most don’t want another film immediately after sitting in the theater for two hours, but one can doomscroll the “empty calories” of short form video for the duration of multiple features and not feel satiated.
However, higher artforms demand of the audience. Fine dining and movie theaters require you to sit with others and give the experience your full attention. Slop requires nothing. Art is communal. The algorithm is isolating. Films, concerts, theater, galleries, fine dining, etc. are normally experienced with a group. The algorithm is experienced alone even if millions see the same content.
Loneliness exacerbates addiction. Bruce Alexander’s Rat Park experiment in 1978 demonstrated that rats in isolated cages compulsively consumed morphine, while rats in a social environment with space, play, and other rats ignored it. Screen addiction is both a symptom and cause of loneliness. When people are lonely, they scroll, but scrolling all day keeps people lonely. An attention Ozempic would create connection.
Ironically, loneliness is what social media was meant to solve. Facebook’s original mission was to connect the world. In a podcast, Mark Zuckerberg talked about how the average American has fewer than three close friends, and proposed Meta’s AI chatbots could fill the gap. Mark, your platform says I have over a thousand friends. What did you do with all my friends, Mark? Social media platforms have the data to connect us. Facebook could easily scan a billion people’s interests, personality, and location, and recommend real-world friendships. Instead, Zuckerberg is proposing friendship as a subscription service (FaaS). It’s a man-in-the-middle attack on the entire concept of human relationships.
Like slop, AI relationships will find a market. Chatbots require nothing. Friends do. Many are already choosing synthetic relationships. Character AI is one of the largest AI platforms. Human relationships might be superior but require more than the press of a button. The solution is not willpower. Ozempic doesn’t require willpower. Ozempic just requires the push of a needle. Why can’t technology create connection at the press of a button?
Here is how you can replace all algorithmic discovery with one-button AI right now—not in some hypothetical future: Give the AI of your choice a context file of art you enjoy. Then, describe the experience you’d like to have—emotions, aesthetics, genre, etc. You can even describe the spiritual themes you’d like to explore or what you are currently processing in your life and ask it to recommend a work of art that will help you explore those themes. These recommendations will be aligned to your prompt rather than the interests of middle-man algorithms.
AIs can only recommend what they know. However, as this method becomes more popular, artists will make their work legible to AIs, the way they did to algorithms. Unlike algorithms, AIs can understand artistic themes. Algorithmic platforms were a crude intermediary between the art and audience, no longer necessary once each side can be directly read by AI. While platforms might attempt to wall their gardens, as AI improves any platform that doesn’t let you bring your own algorithm will be offering an inferior product. Eventually, AI will dissolve the platform by pulling content from the internet rather than pushing it through algorithms.
The algorithm served the interests of the intermediary, not the audience, artist, or their relationship. In this scenario, customized AI serves the individualized interests of each audience member. Audiences will pay more for less friction. Marketing art in a pull environment requires making your work legible, rather than blasting content or ads. The relationship between artist and audience is aligned, because it is direct and reputational. AI can read the artist’s work directly and remember user response.
Artists have always had to make their work legible. Before AI, legibility meant that films had to be optimized for the trailer, logline, and poster, music optimized for genre, hook, and playlist fit, and books optimized for category, cover, and keyword. When AI can read the underlying work, these surface elements become less important. Like a connoisseur who knows your tastes, AI can recommend the gems you would not otherwise discover and ensure complex art finds its audience.
AI could create human connection using the same method. Rather than “logging off” and “putting yourself out there”—a push method of human connection—you could allow AI to search on your behalf and pull connections. If social media platforms allowed, I would use AI to search users by location, analyze profiles for compatibility, and direct message likely friends. AI could do this now, except that platforms lock down their data. Connection is the antidote to screen addiction, and therefore antithetical to their business model. However, as this method becomes known, humans might deliberately make their data more accessible for connection.
Voluntarily disclosing data is different than surveillance, because surveillance is performed by unaccountable external parties without our consent. What I am describing is intimacy—consciously revealing yourself and allowing yourself to be known. AI allows humans to move past the false binary of stumbling through relationships on their own or having their data harvested without consent for the purpose of manipulation, to use machines to better understand themselves and meet their relational needs. While there is a risk to intimacy, it is required for connection.
We already exchange data for connection. When you meet someone new and they ask, “so what do you do?” they are asking for data so they can better connect with you or evaluate the potential for connection. Social media platforms gathered as much data as they did by offering to connect us with everyone in the world. Now, they only offer “content.” The exchange has changed. Using AI, it is possible to reject this new offer in favor of the original.
Creators might still need platforms to deliver their content, but as discovery moves to AI while platform delivery is throttled by algorithms and flooded with slop, that offer is changing as well. If platforms offer a choice between slot machine odds or paid delivery, then creators who pay will drive audiences toward direct connection like paid offers or email lists and those who play the odds will become indistinguishable from AI content. The result will look like a choice between AI slop and human connection, even if humans and AI are involved in both. The original dream of social media was connection. Algorithms created a nightmare of isolation. AI is both a deathblow to algorithmic platforms and an opportunity to restore connection. Algorithms are intermediaries. AI allows audiences to choose their own intermediary, reducing the willpower required to cooperate in the prisoner’s dilemma of culture. Through AI, audiences can directly ask for art and connection based on themes and needs rather than surface elements. They could ask for slop. They could ask for high culture. They could also ask for an experience that does not yet have a name, undiscoverable by previous intermediaries, until now.