The Gettier Problem and the Filter Bubble: Your Reality Might Be Personalized

Last updated: October 22, 2026

The Gettier problem is the discovery that a belief can be justified and true and still fall short of knowledge. Edmund Gettier set it out in 1963 in a three-page paper in the journal Analysis, “Is Justified True Belief Knowledge?”, using two short cases about a man named Smith, a job and a car. His target was the idea, which he found in different wordings in A. J. Ayer and Roderick Chisholm, that to know something you have to believe it, have good grounds for it and be right. Gettier showed that all three conditions can be met by luck. Epistemologists have been hunting for the missing ingredient ever since, and the hunt matters more now that much of what we believe arrives through feeds sorted for each person.

The meme at the top of this page jokes about Gettier’s other claim to fame. A man in a dark red shirt stands in front of a green chalkboard covered in handwriting and squints hard at a scrap of paper the size of a stamp, pinched between two fingers; the caption reads “Reading the complete works of Edmund Gettier”. The memorial page of the UMass Amherst philosophy department, where he taught from 1967 until he retired in 2001, lists his published output as two papers and one review, all from the 1960s. He died in 2021, aged 93.

Ten Coins, a Ford and Brown in Barcelona

Gettier’s two cases show a person who reasons well from good evidence, lands on a true belief, and is right only by accident.

In the first case, Smith and Jones have applied for the same job. Smith has strong evidence that Jones will get it and has counted ten coins in Jones’s pocket, so he concludes that the man who gets the job has ten coins in his pocket. Smith himself gets the job, and he happens to have ten coins too. His conclusion is true and justified, yet it rests entirely on facts about Jones. In the second case, Smith has good evidence that Jones owns a Ford and infers that either Jones owns a Ford or his friend Brown is in Barcelona, a city picked at random. Jones in fact drives a rented car, and Brown happens to be in Barcelona. The disjunction is true and Smith is justified, yet he has no idea where Brown is.

Gettier traced the definition back to antiquity, carefully: in a footnote to the paper, Plato “seems to be considering some such definition at Theaetetus 201”, and “perhaps accepting one at Meno 98”. Our post on Socrates and the theory of knowledge follows that thread through the Theaetetus. What Gettier added in 1963 was a recipe: take a justified false belief, infer something true from it, and you get a justified true belief that is plainly not knowledge.

The Stopped Clock Belongs to Russell

The stopped clock that happens to show the right time, a favourite illustration of the Gettier problem, comes from Bertrand Russell, fifteen years before Gettier wrote.

Russell gives the case in Human Knowledge: Its Scope and Limits, first published in 1948. You look at a clock you believe is working; it has stopped, but you glance at it at the one moment it is right, and, as the Stanford Encyclopedia of Philosophy (SEP) quotes him, “you will acquire a true belief as to the time of day”, yet you “cannot be correctly said to have knowledge”.

Stephen Hetherington, who edited a Cambridge University Press volume on the Gettier problem, calls it a “similar pre-1963 puzzle”. Some popular explainers credit the clock to Gettier, probably because it is tidier than his coin and car cases, whose disjunctions tend to empty a lecture hall.

Dharmottara’s version is older by more than a millennium. The same SEP entry cites the Indian philosopher Dharmottara, around 770 CE: a traveller on a hot day thinks he sees water, which is in fact a mirage, but when he reaches the spot there happens to be water under a rock. Russell used the clock to show that true belief is not enough, and Dharmottara’s case belongs to a debate about perception; Gettier aimed his cases at the full three-part definition, which is one reason the problem carries his name.

Sixty Years of Patches, and a County Full of Fake Barns

According to the SEP, no analysis of knowledge has been widely accepted, and one influential argument says that adding conditions to justified true belief cannot work.

The obvious repair: Smith inferred his true belief from a false one, so require that knowledge is not inferred from any falsehood. The SEP notes that this “no false lemmas” proposal is not widely accepted, partly because of a case first given by Carl Ginet and published by Alvin Goldman in 1976. Henry drives through a county full of barn facades that look like barns from the road. He has no reason for suspicion, points at the one real barn in the county, and believes it is a barn. No false premise is involved, his belief is true and justified, and it still looks lucky, because almost anything else he could have pointed at would have fooled him.

Later proposals looked at the environment too. “Safety” accounts, following Ernest Sosa, ask whether the belief could easily have been false; reliabilist accounts ask whether it came from a process that tends to produce truths in that environment. Linda Zagzebski argued in 1994 that any such addition to justified true belief that does real work will leave room for new Gettier cases, and Timothy Williamson proposed in Knowledge and Its Limits (2000) to stop analysing knowledge and treat it as basic, “knowledge first” in his phrase, as the SEP entry summarises the debate.

Fake Barn County Now Has a Recommendation Engine

A personalised information environment turns the barn county question into an everyday one: the reliability of what you see depends on a setting you did not choose and usually cannot inspect.

Maps are a clean example. In February 2020 the Washington Post reported, as summarised by Digit, that Google Maps showed Kashmir as part of India to users in India, while users elsewhere saw a dotted line marking a disputed territory. A Google Maps product director told the paper that in countries with a local version of the app, “we follow local legislation when displaying names and borders”. Two people on either side of a border can form justified beliefs from the same app and believe different things, without either of them seeing the other’s version.

Strictly, the map example is about disagreement; Gettier luck shows up somewhere subtler. If a ranked feed shows you a true report because it predicted you would click, your belief is true and your grounds look fine, but whether it reached you had little to do with its truth. On a safety reading, many of your accurate beliefs could easily have been inaccurate ones served by the same process. That reading is contested: a reliabilist would reply that a search engine that mostly returns accurate results gives you knowledge, whatever its commercial motives.

What Eli Pariser Warned About in 2011

A filter bubble is the personal information environment that algorithms build around each user by predicting what that user wants to see, a term Eli Pariser made famous with his 2011 book The Filter Bubble; Merriam-Webster dates its first known use to 2010.

The book’s publisher, Penguin Random House, dates the turning point precisely: “In December 2009, Google began customizing its search results for each user.” Instead of the most broadly popular results, each person got a prediction of what they would click, which in the publisher’s words risks “creating individual universes of information for each of us”. The description stresses invisibility: you did not choose the filter and cannot see what it leaves out, which separates the worry from the older one about people who read a single newspaper.

The philosopher C. Thi Nguyen sharpened the concept in a 2020 paper in Episteme, “Echo chambers and epistemic bubbles”. Members of an epistemic bubble, he writes, “merely lack exposure to relevant information and arguments”; members of an echo chamber have been taught to distrust every outside source. “Mere exposure to evidence can shatter an epistemic bubble”, while the same exposure may harden an echo chamber. On Nguyen’s terms, Pariser’s worry describes an epistemic bubble.

The Bubble Is Mostly Self-Inflated, the Studies Say

Three large studies suggest that people’s own choices and networks narrow their news at least as much as ranking algorithms do.

In 2015, Eytan Bakshy, Solomon Messing and Lada Adamic, working with data from 10.1 million US Facebook users who had reported their political ideology, reported in Science that the News Feed ranking reduced cross-cutting hard news by 8 percent for liberals and 5 percent for conservatives, while users’ own clicks reduced it by 6 and 17 percent. Their summary on Facebook’s research blog: “The composition of our social networks is the most important factor”. The authors worked at Facebook, whose ranking they were assessing, but independent work points the same way. Ronald Robertson and colleagues compared what Google Search showed people around the 2018 and 2020 US elections with what they clicked, and concluded in Nature in 2023 that exposure to partisan and unreliable news was “driven not primarily by algorithmic curation but by users’ own choices”.

Like-minded content is still common. A 2023 study led by Brendan Nyhan found that like-minded sources make up the majority of what US adult Facebook users see; in a field experiment with 23,377 users during the 2020 election, cutting that exposure by about a third “had no measurable effects on eight preregistered attitudinal measures”, including polarisation and belief in false claims, according to the NYU Center for Social Media and Politics. Our post on the Plato’s cave meme and social media adds a literature review and a feed experiment that point the same way. The limits matter too: these studies are all American, mostly about news, and run over weeks or months, so they say little about long-term effects or about short-video feeds.

What Arendt Feared, and What a Feed Can Add

One deeper worry in this debate has a classic statement in Hannah Arendt: people losing the ability to tell fact from fiction, a problem that filters can feed and that is older than any of them.

In The Origins of Totalitarianism, Arendt wrote that the ideal subject of totalitarian rule, more than any convinced Nazi or Communist, is “people for whom the distinction between fact and fiction”, along with the distinction between true and false, no longer exist. She was writing about isolated people under terror and propaganda, so applying her to newsfeeds is a stretch. What carries over is narrower: a public that shares no common set of facts finds it hard to disagree productively, because disagreement needs a shared object.

Gettier helps put the personalisation scare in proportion. His cases showed that being right and knowing came apart long before any algorithm, and that knowledge can hinge on conditions outside the believer’s head. A personalised feed is one more such condition, and so far the evidence suggests it is no stronger than our own choices about whom to follow and what to click. Much of the practical work stays with us, starting with the plain question of how a piece of information reached us.

If the search for a final justification keeps going in circles, read our post on the Münchhausen trilemma and why proving everything is impossible.