We are drowning in information while starving for clarity. More data is produced in a single day now than was generated in entire decades of the twentieth century. A person with a smartphone holds in their hand more accumulated knowledge than any library could have contained a generation ago.
And yet, for all this access, many of us have never been more certain that we are right, and more blind to the possibility that we are not.
This is the predictable outcome of a collision between an ancient quirk of human cognition and the most powerful technological systems ever built. Together, confirmation bias and algorithmic media are quietly constructing a version of reality for each of us: one that feels complete, feels objective, and is almost certainly wrong.
What Confirmation Bias Actually Is
Confirmation bias describes our tendency to search for, interpret, and remember information that confirms what we already believe, while dismissing evidence that challenges those beliefs.
This is not a flaw unique to certain people; it is a feature of human cognition itself. Our brains did not evolve to find abstract truth in isolation. They build mental models of the world and preferentially notice evidence that supports those models because doing so saves mental energy. Revising a worldview requires effort, discomfort, and sometimes a painful reckoning with identity.
Consider how this plays out. If you dislike a public figure, you will notice and remember every instance that confirms their dishonesty, while rationalising away actions that complicate that view. The opposite is true if you admire them. Over time, two people can watch the exact same set of events and walk away with two radically different accounts, both feeling entirely justified by the facts.
Because this operates below conscious awareness, it is extraordinarily difficult to counteract without deliberate effort.
How Algorithms Distort the Other Side
If confirmation bias is the kindling, algorithmic media is the accelerant.
Platforms like YouTube, TikTok, X, and Instagram are built on engagement-driven recommendation systems designed to keep you scrolling. The most reliable way to hold attention is to serve you content that triggers an emotional reaction.
For years, critics warned that this would trap us in isolated "filter bubbles" where we never see opposing views. Modern empirical research shows the reality is much worse. We are exposed to the other side, but recommendation systems preferentially show us the opposing perspective in its most extreme, provocative, and caricatured forms.
We do not encounter our opponents' best arguments. Instead, we encounter them through mocking commentary, outraged reaction videos, or out-of-context clips. This constant, hostile exposure drives an emotional aversion to the out-group. We are not trapped in silent bubbles; we are trapped in a noisy arena where we learn to despise the other side based on a funhouse-mirror depiction of their beliefs.
When different communities operate from fundamentally different portions of the facts, productive disagreement becomes nearly impossible. People are no longer arguing different conclusions from the same evidence; they are living in entirely different evidentiary universes.
The Authority Illusion
This dynamic is compounded by a second problem: the illusion of expertise created by the democratisation of media.
In previous decades, publishing a book or broadcasting to an audience required passing through institutional gatekeepers who imposed a baseline of accountability. Those gates have fallen. Anyone can now launch a podcast, build a YouTube channel, or cultivate a massive following. While this has allowed marginalised voices to flourish, it also means the markers of authority are easily mimicked.
A confident voice, slick video production, and a high follower count create an impression of credibility that has nothing to do with actual rigor.
This creates a structural disadvantage for real experts. True specialists are bound by methodology and the natural uncertainty of data. They must use tentative language, qualify their findings, and acknowledge what they do not know.
Self-styled internet experts face no such constraints. They can speak with absolute, unblinking certainty. Because algorithms reward bold, emotionally resonant declarations over nuanced caveats, the loudest voices easily crowd out the most qualified ones.
The ability to distribute content is not the same thing as the ability to analyze information accurately. A person with a microphone is not automatically a person with a method.
Developing an Information Practice
We cannot simply retreat to trusting old institutions, which have their own biases. Instead, individuals must develop a personal practice of due diligence when evaluating sources:
Expertise is non-transferable. Success or intelligence in one domain does not confer expertise in all domains. A compelling storyteller is not automatically a reliable analyst.
Follow the incentives. Who benefits from a particular narrative being believed? What does a creator gain financially or socially by stoking your anger or validation?
Seek out the strongest version of opposing arguments. If your only exposure to a contrary position comes from someone who is currently mocking it, you have not yet encountered that position.
Distinguish between primary sources and interpretation. A person reading a headline to you is selling an interpretation. Look for the underlying data, methodology, or original document.
Paths Forward
There is no single solution to a problem rooted in both human psychology and technological design, but changes at two levels can help.
At the individual level, we must practice information hygiene. This means auditing your information diet to see if your favourite sources merely validate you, and embracing the intellectual humility to say, "I do not know enough about this to have an opinion yet."
At the systemic level, we need pressure for algorithmic transparency and design interventions. Tech platforms must explore modifications to recommendation systems that introduce perspective diversity rather than optimising purely for raw engagement. Finally, media literacy education must focus on teaching people how to recognise cognitive biases and understand the basic mechanics of how curation algorithms target them.
Conclusion
Humans have always been susceptible to seeing what they expect to see. What is new is the scale and precision with which technology can now weaponise that susceptibility. Curation algorithms are not malicious, but they are highly effective at creating a comfortable illusion: that the world we see on our screens is the world as it is, rather than the world as it has been selected for us.
Access to information is not the same as access to understanding. Understanding requires something that no algorithm can provide on our behalf: the willingness to be wrong, the discipline to investigate, and the humility to recognise that the version of reality we carry in our minds is always, at best, a draft.
Citations and References
Bruns, A. (2019). Are Filter Bubbles Real? Polity Press.
Iyengar, S., Sood, G., & Lelkes, Y. (2012). Affect, not ideology: Social identity and partisan polarization. Public Opinion Quarterly, 76(3), 405-431.
Mercier, H., & Sperber, D. (2017). The Enigma of Reason. Harvard University Press.
Pariser, E. (2011). The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think. Penguin Books.
Sunstein, C. R. (2017). #Republic: Divided Democracy in the Age of Social Media. Princeton University Press.
Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129-140.
