Occam's Razor

011 min

Occam's Razor at a glance

• What it is: A tie-breaking rule for choosing between explanations that already fit the evidence equally well, not a rule that the simplest story is automatically the true one.
• Origin: William of Ockham (c. 1287-1347) used the principle of economy repeatedly in his own writing; the now-famous Latin phrasing was formulated by the Irish philosopher John Punch in 1639, nearly three centuries after Ockham died.
• Confused with: Hickam's dictum, the medical counter-principle that a patient can have more than one disease at once.
• Watch for: it only decides between explanations that predict the same observations. When two explanations predict different things, evidence decides, not simplicity.

021 min

The problem Occam's Razor names

The mistake Occam's razor names is adding an explanation's assumptions faster than its evidence. A theory can always be rescued from an awkward observation by tacking on one more special case: one more hidden variable, one more exception, one more unobserved cause. Each addition can be made to fit the data, so fit alone never stops the process.

Pre-Copernican astronomy is the standard case. A model with the Earth at the centre could still predict planetary positions, but only after generations of astronomers added further corrective circles on top of the original ones to match new observations. Every fix worked. The list of fixes kept growing anyway. Occam's razor names the discomfort with that pattern: a theory that needs a growing patchwork of unconnected fixes has stopped explaining and started accommodating.

032 min

What Occam's Razor actually says

William of Ockham's own principle of economy, in the plainest translation of his Latin, is "it is futile to do with more things that which can be done with fewer." Notice what the sentence does not say. It does not say the explanation with fewer things is true. It says that doing something with more things, when fewer would do the same job, is pointless.

That restriction is the whole mechanism. Occam's razor is a comparative principle: it only applies to two or more explanations that already predict the same observations equally well. Among those, it prefers the one that introduces fewer new assumptions, entities, or causes to reach that prediction. If one explanation predicts something the other does not, the razor has nothing to say. Evidence decides, the way it always does.

Isaac Newton stated a version of the same rule as a rule of reasoning for natural philosophy: "We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances. Therefore, to the same natural effects we must, as far as possible, assign the same causes." Newton's clause "as far as possible" carries the same restriction Ockham's does — the rule binds only once the effects are actually the same.

The reason the razor works as a guide, rather than as proof, is statistical rather than metaphysical. Every extra assumption a theory carries is one more thing that can turn out to be wrong. An assumption that adds no predictive power only adds a place for the theory to fail. This is close to the reasoning behind model selection in statistics and machine learning: a model with unnecessary parameters can fit the data used to build it while doing worse on new data, a problem statisticians call the bias-variance tradeoff. Wikipedia's summary of the mechanism states it directly: excessively complex models are affected by statistical noise, a problem also known as the bias-variance tradeoff, whereas simpler models may capture the underlying structure better and may thus have better predictive performance.

041 min

Where Occam's Razor comes from

The attribution to William of Ockham is older than the phrase "Occam's razor" itself, and the two are not the same claim. Ockham, an English Franciscan friar, used versions of the economy principle across his writing rather than stating one canonical sentence. In his Summa Totius Logicae, he wrote "Frustra fit per plura quod potest fieri per pauciora" — "it is futile to do with more things that which can be done with fewer."

The Latin sentence most people quote today, "Entia non sunt multiplicanda praeter necessitatem" and its close variants, does not appear in Ockham's surviving works. The phrasing usually repeated, "Entities are not to be multiplied without necessity," was formulated by the Irish Franciscan philosopher John Punch in his 1639 commentary on the works of Duns Scotus, nearly three hundred years after Ockham's death in 1347. Ockham inherited the idea, too: Duns Scotus had already written "Pluralitas non est ponenda sine necessitate," "plurality is not to be posited without necessity," before Ockham was born. What made the principle stick to Ockham's name was how often and how forcefully he used it, not that he invented it.

052 min

Occam's Razor in medicine and in machine learning

The clearest worked example is medical diagnosis, because it turns the razor into a daily professional habit with a name. Theodore Woodward, who taught at the University of Maryland School of Medicine for nearly fifty years, is credited with the version doctors still repeat. According to the medical historian John Sotos's research, Woodward told his students beginning in the 1940s not to look for zebras on Greene Street, the road outside the medical school, meaning: when you hear hoofbeats, expect a horse, not an exotic animal. A patient with a cough, fever, and chest pain almost always has a common respiratory infection. A doctor who first reaches for a rare autoimmune disease to explain the same three symptoms has added an unnecessary entity to the explanation, and Occam's razor says to prefer the common diagnosis until the common one stops fitting the evidence.

The same habit shows up in current machine learning research, in a more technical form. A 2023 paper by Chris Mingard, Henry Rees, Guillermo Valle-Pérez, and Ard A. Louis studied why heavily overparameterized deep neural networks generalise to new data at all, instead of simply memorising their training set. Their answer, from the paper's own abstract, is that these networks carry "an intrinsic Occam's razor-like inductive bias towards (Kolmogorov) simple functions" strong enough to counteract the sheer number of complex functions the network could otherwise learn. The network is not reasoning about simplicity. Its training process is structured so that, among the many functions that fit the training data equally well, it lands on a simple one far more often than chance would predict.

062 min

How Occam's Razor shows up in product, design, and engineering work

For a PM triaging a bug report, the razor is a question to ask before escalating: does this really need a new, unverified cause, or does an existing, already-known failure mode explain it? A support ticket describing "random" data loss is more often explained by a known race condition already in the backlog than by a new, unobserved one. Reaching for the second before ruling out the first adds an entity the evidence does not yet require.

For an engineer, the same discipline shapes refactoring: when two designs pass the same tests and meet the same requirements, the one with fewer moving parts is the one to ship, because every additional class, flag, or special case is a place the next bug can hide. This is not a claim that the simpler design is always faster or better — only that, once two designs are equally correct, the simpler one has fewer places left to go wrong.

For a founder reading a growth dashboard, the razor cuts the other way from how it is usually invoked. A number that jumped after three changes shipped in the same week does not, by itself, tell you which change caused it. Naming one change as "the" cause because it is the simplest story is not applying Occam's razor — it is confirmation bias borrowing the razor's reputation. The razor only compares explanations that already fit the same evidence; a single unverified guess is not automatically the simpler explanation just because it is the first one that comes to mind.

071 min

When Occam's Razor is misapplied

The most common misreading, addressed directly above, is treating "simpler" as a synonym for "true." The razor breaks ties between explanations that already explain the same observations. It has no authority once two explanations diverge in what they predict; the more complex one can simply be correct.

Francis Crick, who co-discovered the structure of DNA, stated the field-specific limit plainly: "While Ockham's razor is a useful tool in the physical sciences, it can be a very dangerous implement in biology. It is thus very rash to use simplicity and elegance as a guide in biological research." His reasoning was that biological systems are the product of evolution by natural selection, which has no obligation to produce the most elegant solution to a problem, only one that worked well enough to survive. A cell's machinery being needlessly redundant is not evidence against a hypothesis; redundancy is exactly what an evolved system produces.

Medicine has its own named counter-principle for the same reason, covered next. A patient's symptoms are not guaranteed to trace back to one cause just because one cause would be a tidier chart note.

081 min

Occam's Razor vs. Hickam's Dictum

The concept this audience confuses Occam's razor with most often is Hickam's dictum, a direct counter-principle from medicine. Where Occam's razor says to prefer the explanation with the fewest causes, Hickam's dictum states, in its usual phrasing, "a man can have as many diseases as he damn well pleases." Wikipedia records that the principle is attributed to an apocryphal physician named Hickam, possibly John Bamber Hickam, a doctor who later chaired the medicine department at Indiana University.

The two are not actually in conflict, because they answer different questions. Occam's razor says not to invent a second cause before the evidence requires one. Hickam's dictum says not to force multiple real symptoms into a single diagnosis just because a single diagnosis is more elegant. A patient who has both diabetes and a broken wrist has two ordinary, independent explanations for two different sets of symptoms — reaching for one exotic condition that would explain both is exactly the mistake Occam's razor itself warns against, and reaching for one instead of two diagnoses just to keep the chart simple is exactly what Hickam's dictum warns against. The deciding fact is whether the observations in front of you actually come from one process or from more than one; neither principle can answer that by itself.

091 min

How to apply Occam's Razor

1. 1

   Check that the explanations actually predict the same thing first.

   If they do not, Occam's razor is the wrong tool — go find the observation that tells them apart instead.

2. 2

   Count entities, not words.

   A short explanation that quietly assumes an unobserved cause is not simpler than a longer one that only uses causes already on the table.

3. 3

   Ask what the extra assumption buys you.

   If removing it changes no prediction, it is decoration. If removing it changes a prediction, it was never optional, and the razor does not apply to it.

4. 4

   Do not use it to dismiss real complexity.

   Correlation and causation get confused here too: a clean, single-cause story is often just the first story someone reached for, not the one with the fewest actual assumptions once it is checked against the evidence.

5. 5

   Revisit the simpler explanation when new evidence arrives.

   The razor is a working preference among current explanations, not a verdict that closes the question.

102 min

Frequently asked questions about Occam's Razor

?6 questions

Questions people ask

§8 sources

Sources