We blame "the algorithm" for everything now. We watch one clip from a Netflix documentary and suddenly every app believes we are emotionally available for celebrity scandals, cult analysis, true crime theories and a forty seven part TikTok series called What They Don't Want You To Know.
A teenager watches one gym video and suddenly his feed is full of crypto, testosterone supplements and a man shouting about "discipline" from inside a rented Lamborghini in Dubai. We watch one lasagna recipe and spend the next three weeks being followed around the internet by air fryers, parmesan brands and cheerful women reorganising refrigerators into matching glass containers.
The algorithm, apparently. At this point, "the algorithm" has become the modern version of "something in the air". We all talk about it, almost none of us can properly explain it. And honestly, the data says this is not just a feeling. AGCOM has found that roughly two thirds of Italians have low or no algorithmic literacy. In the UK, Ofcom found that around a third of adults say they do not understand how platforms use their data to decide what content they see online.
Which is extraordinary when we think about it, because algorithms now shape what we buy, what we watch, what we believe, who gets visibility, which news spreads, which jobs we get considered for and whether YouTube has quietly convinced us we should suddenly care deeply about Icelandic hiking equipment. We live inside algorithmic systems all day long, we understand them only vaguely, and the funny part is that algorithms are not even new—they are ancient.
An algorithm is simply a sequence of instructions designed to produce an outcome: a recipe is an algorithm, long division is an algorithm, a skincare routine involving twelve expensive serums and emotional optimism is also, structurally speaking, an algorithm.
For a very long time, algorithms stayed relatively simple, with step by step procedures, fixed rules, and predictable outputs. Then computers arrived and humanity essentially handed caffeine and steroids to algorithms because computers did not invent algorithms—they industrialised them.
We can follow one recipe, computers can follow billions simultaneously without getting tired, distracted, or suddenly deciding to "just eyeball it". And once instructions become computational, almost anything can become computational too—traffic systems, advertising, dating apps, shopping, recruitment, insurance pricing, music recommendations, news feeds, human attention itself. This is where things start becoming less theoretical and more "why does TikTok know I am having a mild existential crisis at 11:30pm?"
From Library to Prediction Machine
The early web behaved more like a library—we searched for something, the system retrieved information. The modern internet behaves differently, with today's systems increasingly predicting behaviour.
And this is where AI enters the picture, because AI is not the same thing as an algorithm, despite the fact we now use the two words interchangeably like exhausted relatives at Christmas trying to explain cryptocurrency to each other. Traditional algorithms follow explicit rules, while AI systems learn patterns from enormous amounts of data and continuously adjust predictions over time. Twenty years ago, if engineers wanted to create an email spam filter, they might manually write rules: IF email contains "WIN MONEY NOW!!!" and fourteen exclamation marks, move to spam folder. Modern AI systems are trained on millions of examples—nobody manually writes every rule anymore, the system learns patterns statistically.
Every click, every pause, every replay, every search, every purchase, every moment we linger slightly too long watching a woman refill matching oat containers in a beige kitchen while quietly wondering whether our own life has collapsed structurally—all of this becomes data, and data in this context simply means information generated by behaviour. AI systems analyse these behavioural patterns and ask: "What is this person statistically most likely to do next?" That is the modern web: prediction infrastructure.
Folk Theories of the Algorithm
Researchers studying YouTube users found that people develop what they literally called "folk theories" about algorithms. We experience recommendation systems constantly while understanding them only partially, so we start building little personal mythologies around them.
"The algorithm listens to my conversations." "TikTok knows when I'm depressed." "Instagram punishes you if you post links." "If I pause too long on a video it thinks I'm obsessed."
Some of these theories are partly true, partly false and partly misunderstood. Which is exactly why these systems feel mystical. The interesting thing is that the folklore is not completely wrong. We often read the effect correctly and explain the mechanism badly.
Most of the time, our phone is not literally listening to our conversations—the system does not need to hear us talk about hiking boots. It already has search history, location signals, browsing behaviour, purchase patterns, similar audiences and thousands of tiny clues from people who behave like us. The result can feel like listening, mechanically it is prediction.
TikTok does not know sadness the way a friend might, but it can read behaviour: late night scrolling, slower swiping, repeated viewing, certain searches, certain music, longer pauses. The system is not diagnosing emotion, it is detecting patterns that may correlate with emotional states. The feeling is personal, the mechanism is statistical.
Around 90% of what people watch on TikTok comes from algorithmic recommendation rather than accounts they intentionally chose to follow—one of the largest media systems in human history is organised around prediction.
The Commercial Layer
The old internet waited for us to search, the modern internet predicts first, ranks first, suggests first and increasingly decides first. Even Google now increasingly answers questions directly without sending us to websites—convenient for us, slightly catastrophic for half the internet economy. Underneath all this sits the final layer we rarely think about: money, because many of these systems are connected to advertising and commercial optimisation, and the longer we stay online, the more opportunities exist to monetise our attention.
Creators adapt to what performs, journalists adapt to what gets clicked, brands adapt to what spreads, teenagers adapt to what receives visibility. We all begin adjusting ourselves, slightly, toward what the platforms reward, which means algorithms are no longer just technical systems quietly sorting information in the background—increasingly, they shape the conditions under which modern life gets experienced.
And now AI accelerates this process again, because AI systems no longer simply rank or recommend information, they generate it, summarise it, rewrite it, filter it, interpret it and speak it back conversationally in warm emotionally supportive voices at 1am. The internet is shifting from information retrieval to behavioural mediation—once we understand the mechanics, the internet becomes slightly less mystical. The algorithm is not reading our soul, it is reading behavioural signals. We used to navigate the web, increasingly the web navigates us back.
Selected Sources
- AGCOM, literacy and algorithmic awareness research in Italy.
- Ofcom Adults' Media Use and Attitudes reports on understanding platform data and feeds.
- Eslami et al. (2015), Invisible Algorithm Awareness Study on Facebook News Feed.
- Rader & Gray (2015), studies on user beliefs about algorithmic curation.
- Folk Theories of Algorithmic Recommendations on YouTube (2020).
- Washington Post TikTok recommendation analysis and behavioural study.
