YouTube, Facebook, TikTok, Instagram, Google Search, these platforms don't show you the actual world. They show you a model of the world constructed from your behavioral data.
Two people open their laptops and do a search for the same news topic. They live in the same city. They are roughly the same age. They pay the same taxes, breathe the same air, send their kids to the same schools. What they see on their screens, however, has been individually curated by algorithms that have been studying their behavior for years, and what they see is not the same. By the time they close their laptops and go about their day, they will inhabit different versions of reality.
This is not a hypothetical. It is the designed and documented outcome of the recommendation engines that now mediate most of our information consumption, which is arguably the most consequential technological development of our lifetimes.
The mechanics are straightforward enough. YouTube, Facebook, TikTok, Instagram, Google Search, these platforms don't show you the actual world. They show you a model of the world constructed from your behavioral data: what you've clicked on, how long you’ve watched, what made you stop scrolling, what made you share. The goal of that model is not accuracy, it's engagement. And engagement, it turns out, is not driven by truth or balance or nuance. It's driven by outrage, fear, and the neurological warmth of tribal confirmation.
Guillaume Chaslot learned this from the inside. Chaslot joined Google in 2010 and worked on YouTube's recommendation algorithm. What he found disturbed him enough that, after he left Google, he built a nonprofit, AlgoTransparency, which is dedicated to exposing what the algorithm actually does. The system he helped build, he has said, was optimizing for a single metric: watch time. Not quality. Not accuracy. Not what was good for the person watching. Just time on screen. And the content that maximized watch time, he found, was conspiracy theories, outrage, and increasingly extreme ideological material because those things are, neurologically speaking, harder to look away from than the truth. When Chaslot proposed internal fixes such as algorithms designed to break filter bubbles and present dissenting viewpoints, he was told it wasn't a good idea to work on that. Shortly after, he was fired.
Facebook’s version of this story is even more starkly documented, thanks to Frances Haugen, a former product manager who, in 2021, walked out of the company with tens of thousands of pages of internal research. What those documents showed was that Facebook's own engineers knew the algorithm was amplifying hateful, divisive, and emotionally extreme content.
One internal document, as reported by The Wall Street Journal, noted that the platform's core mechanics—virality, recommendations, engagement optimization—were a significant driver of why misinformation and divisive speech flourished. Proposed fixes were considered and largely rejected. Not out of malice, Haugen has been careful to say, but because of the numbers and the bottom-line. Safer content meant less engagement, and less engagement meant less revenue. The algorithm stayed.
“Facebook makes more money when you consume more content,” Haugen told 60 Minutes. “And it turns out the content that keeps you consuming is the content that makes you angry.”
So here is what we have built, largely without intending to: a global system for the personalized curation of reality, optimized for emotional activation, running continuously, invisibly, and at a scale that dwarfs anything in the history of human communication.
The printing press transformed civilization. The broadcast television era transformed it again. But neither of those technologies could look at you individually, study your psychology in real time, and construct a version of the world specifically calibrated to keep you engaged.
The philosophical implications of this are hard to overstate. Humans have always lived inside information bubbles, but those bubbles had permeable membranes and it wasn’t uncommon to encounter people who disagreed with you. The friction was inconvenient but epistemically vital.
Algorithmic curation has removed that friction, replacing it with a frictionless stream of digital content that always confirms, always validates, always enrages in a satisfying direction. The philosopher’s term for the result is “epistemic closure.” The algorithm's term for it is a “highly engaged user.”
But the algorithm isn’t malicious. It’s just optimized for engagement, and that’s what makes it so different from previous information threats. Propaganda requires intent. The recommendation engine requires only a metric. Point it at human psychology and tell it to maximize engagement, and it will find, with the relentless efficiency of machine learning, every crack and crevice in our cognition where fear and outrage and tribal identity can be pried open. It doesn't even need to lie. It just needs to choose which truths to show you.
Neil Postman warned us 40 years ago that every medium of communication carries within it a hidden curriculum—a set of values and assumptions that shapes the consciousness of its users independent of any specific content. Television, Postman argued in Amusing Ourselves to Death, didn't just change what we watched; it changed how we thought. It rewired our tolerance for complexity, our appetite for nuance, and our capacity for sustained attention.
The printing press gave us the Reformation and the Wars of Religion—a century of catastrophic conflict seeded by the sudden, uncontrolled democratization of information. The recommendation engine has given us something structurally similar: a sudden, algorithmic fragmentation of shared reality, happening too fast and too invisibly for most people to even notice it’s occurring.
The two people doing a search for the same news topic but being served up different versions of reality can’t have a rational conversation about the world. They’ve been carefully, efficiently, and profitably sorted into their various information silos one scroll at a time by a machine that doesn’t care what they believe.
It only cares that they keep on scrolling.
