SWITCHED ON
The daily technology series nobody asked for but everyone needed
The Algorithm Made Me Do It: Who's Really Running Your Feed
The invisible force deciding what you read, watch, believe, and buy has one job — and your wellbeing is not it.
The algorithm does not know you, does not care about you, and has no opinion about whether what it shows you is true, good, healthy, or real. It has one job: keep you engaged. Everything else is a rounding error.
In Episode 6 we talked about the metaverse — the hundred-billion-dollar bet on a virtual future that humans, with characteristic stubbornness, declined to turn up for on the schedule that had been arranged for them. Today we're going somewhere we've been circling since Episode 1. We've talked about AI and jobs, social media and children, synthetic reality, data privacy, facial recognition, and the metaverse. And underneath almost every single one of those conversations, there has been a presence — invisible, omnipresent, endlessly optimised — that we haven't examined directly yet.
The algorithm. More specifically: the recommendation algorithm. The system that decides, in the fraction of a second between you opening an app and content appearing on your screen, what you see. In what order. With what emotional framing. At what moment. With what effect on your mood, your beliefs, your purchasing behaviour, your political views, and your understanding of what the world is actually like.
You did not choose most of what you saw online today. The algorithm chose it for you. And the algorithm has one job: keep you engaged. Everything else is a rounding error.
01 — What an Algorithm Actually Is
The word "algorithm" has become so ubiquitous and so vague that it's worth starting with what we actually mean. In its most basic sense, an algorithm is just a set of rules for solving a problem. Your recipe for making pasta is an algorithm. The word itself is morally neutral. The recommendation algorithms we're talking about are vastly more complex, but pursuing a similarly well-defined goal: given everything we know about this user, what piece of content should we show them next to maximise the probability that they keep scrolling?
To answer that question, the algorithm draws on an extraordinary volume of data. Not just what you've clicked on, watched, or liked — though it knows all of that, in granular detail, going back years. It knows how long you paused on a piece of content before scrolling past. It knows what you searched for at 2am that you'd never tell another human being. It knows, with reasonable accuracy, your political leanings, your relationship status, your anxieties, and your approximate net worth.
It knows, based on the behaviour of millions of users who look behaviourally like you, what you're likely to engage with before you know it yourself.
02 — The Engagement Trap and Why Outrage Travels Fastest
Here's the thing about optimising for engagement that the platforms understood very early and have been somewhat reluctant to discuss publicly: the emotions that drive the strongest engagement are not the pleasant ones. Joy drives engagement. Amusement drives engagement. But they drive it less reliably and less intensely than anger, fear, outrage, and anxiety. Content that makes you angry keeps you on the platform longer than content that makes you happy.
The algorithm doesn't understand any of this in the way a human would. It doesn't know what anger is. What it knows is that certain patterns of content produce certain patterns of behaviour, and it optimises relentlessly for the behaviours it's been told to maximise. The result is a system that functions as if it were deliberately designed to make you angry — even though no human being sat down and said "let's make people furious." The fury is an emergent property of optimising for engagement without any constraints about what kind of engagement is acceptable.
This is not a conspiracy theory. Frances Haugen's leaked internal documents from Meta showed the company's own researchers had concluded the algorithm was amplifying divisive and hateful content because it drove higher engagement. The company knew. It continued anyway.
03 — What the Algorithm Does to Your Politics
The original filter bubble thesis — that algorithms create hermetically sealed information environments where people only see content that confirms their existing beliefs — is probably too strong. The research is genuinely mixed. People do encounter opposing viewpoints online, more than pure filter bubble theory would predict.
What is robustly documented is something more subtle and more insidious. Algorithms don't just show you what you already believe. They show you the most emotionally engaging version of what you already believe — which tends to be the most extreme, the most outraged, the most convinced-of-its-own-righteousness version available. They don't create ideological uniformity. They create ideological intensification. They take whatever political and social views you start with and dial them up, because the dial-up version performs better.
The effect, across millions of users, is a population whose views on contested questions are simultaneously more extreme and more confident than they would be in the absence of algorithmic amplification. This is not good for democracy, which depends on at least some willingness to tolerate uncertainty and engage with complexity. It is very good for engagement metrics.
04 — TikTok's Algorithm Is Different and That's the Point
TikTok's For You Page operates on a fundamentally different principle to what came before. It doesn't care about your social graph. It doesn't particularly care about your stated preferences. It cares about your behaviour — specifically, what you watch, how long you watch it, and what you watch immediately after. The algorithm starts from almost zero knowledge about you and rapidly converges on a highly accurate model of your interests through direct behavioural observation. New users report that the app "figured them out" within twenty to thirty minutes of first use. That's not an accident. That's the system working exactly as designed.
The speed and accuracy of TikTok's personalisation is one reason it's so effective at capturing and holding attention. It's also one reason it's the subject of significant national security concern — the worry being that an algorithm controlled by a company subject to Chinese government influence could, in principle, be used to shape the information environment of hundreds of millions of people in ways that serve Chinese state interests. Whether that's happening, in what form, and to what extent is genuinely contested. That the theoretical capability exists is not.
05 — Can It Be Fixed? The Honest Answer
Several approaches to reforming algorithmic recommendation systems have been proposed, debated, and in some cases partially implemented. The honest assessment is that none of them fully solves the problem. Chronological feeds are available on Instagram as an opt-in alternative. They're used by a small minority of users. The algorithmic feed drives dramatically higher engagement, which means more time on the platform, which means more advertising revenue. Platforms have little commercial incentive to make the chronological option the default.
The fundamental problem is that the algorithmic recommendation system is not a bug in the social media business model. It is the business model. The attention it captures is the product being sold. Reforming it meaningfully requires either changing the business model — which the companies will resist with every resource available — or regulating it in ways that impose real costs on engagement-maximising behaviour. Which brings us back to the regulation question we've visited in several episodes of this series, and the same conclusion: too slow, too partial, actively contested.
Tomorrow we're going somewhere genuinely controversial — autonomous weapons. AI systems making lethal decisions without human intervention. The debate about whether machines should be permitted to decide who lives and who dies is happening right now, in defence ministries and international forums, and most people have absolutely no idea it's occurring. Tomorrow we make it unavoidable. See you then.
Switched On is a daily technology series covering AI, social media, data privacy, and the digital forces reshaping modern life — with no corporate spin, no false comfort, and absolutely no mercy for buzzwords.
