PhilosophyCounterintuitive resultExplained★ Famous
Occam's Razor
Also known as Ockham's Razor, Law of Parsimony, Principle of Parsimony
When two explanations fit the evidence equally, which should you believe? The one that assumes less.
Counterintuitive result. Nothing is wrong with the reasoning. The result is simply correct and human intuition is simply unreliable here. Not a contradiction at all.
Discover
Two doctors look at an X-ray. The patient has a persistent fever, night sweats, and weight loss. The X-ray shows a suspicious shadow on the lung.
One doctor says: "This is most likely tuberculosis. It explains all three symptoms, it's consistent with the X-ray, and we see it regularly in this region."
The other says: "It could be lymphoma, coexisting with an undiagnosed fungal infection, with the weight loss explained by stress. Or it could be—"
The first doctor is applying Occam's Razor. Tuberculosis is one entity that explains all the observations. The second doctor is multiplying entities unnecessarily — invoking several separate causes where one will do.
The razor says: prefer the first explanation until the evidence requires something else.
Understand
Occam's Razor is not a claim about what is true. It is a claim about what to believe, provisionally, when multiple explanations are consistent with the same evidence.
The principle is usually stated as "the simplest explanation is best," but this is imprecise. "Simple" is ambiguous — simple to understand? Short to state? The correct formulation is about assumptions: the explanation that introduces fewer unneeded assumptions.
An assumption is "unneeded" if removing it still allows the explanation to account for the observed evidence. If you need to posit three separate entities where one would do, the three-entity explanation is not preferred — not because entities don't exist, but because you haven't shown they're necessary.
What the razor cuts
Redundant causes. If two symptoms can be explained by one disease, do not posit two diseases until you have evidence for the second.
Unobservable machinery. If an explanation works equally well without positing a hidden mechanism, the mechanism is unneeded until evidence for it appears.
Ad hoc patches. If a theory requires special explanations for every exception, and a competing theory handles the exceptions naturally, the patched theory has more assumptions, not fewer.
What the razor does not cut
Complexity that is required. If the world is complex, Occam's Razor does not simplify it. It says prefer the less complex explanation, not that the less complex explanation is true. Einstein is supposed to have said: "Everything should be made as simple as possible, but no simpler." The razor is the first part; "no simpler" is the constraint.
Established entities. The razor says do not multiply entities beyond necessity. Entities that are already established — by evidence, by prior argument — are not being multiplied.
Examples
| Choice | Occam's Razor preferred | Why |
|---|---|---|
| Motion of planets: epicycles vs. ellipses | Ellipses (Kepler) | Fewer adjustable parameters; simpler model |
| Life on Earth: special creation vs. evolution | Evolution | One mechanism explains the diversity; special creation requires a separate explanation for each species |
| Coin flips 10 heads: fair coin vs. biased | Fair coin (usually) | No extra parameter needed; 10 heads in a row is unlikely but not impossible |
| Ghost explains creak: ghost vs. thermal expansion | Thermal expansion | Posits no new entity; thermal expansion is already known |
Explore
Occam's Razor and Bayesian reasoning. In Bayesian probability, Occam's Razor has a formal justification. A model with more…
Occam's Razor and Bayesian reasoning. In Bayesian probability, Occam's Razor has a formal justification. A model with more parameters can fit more data — but it requires assigning probability mass over more possible values, so its prior probability for any specific outcome is lower. A simpler model concentrates its predictions and makes higher-probability specific predictions. When evidence is consistent with both models, the simpler model has a higher posterior probability. The razor is Bayesian.
Overfitting in machine learning. The razor has a precise technical instantiation in machine learning: regularisation. A model that perfectly fits the training data but performs poorly on new data has overfit — it has learned the noise, not the pattern. Regularisation penalises complexity, forcing the model to find the simplest pattern consistent with the data. L1 regularisation (which drives coefficients to zero) is Occam's Razor implemented as a penalty term.
Where the razor misleads. The world is sometimes genuinely complex. The simple explanation is often wrong. Tuberculosis sometimes coexists with another disease. The patient sometimes has the rare thing. Occam's Razor is a tiebreaker for competing hypotheses that fit the evidence equally well; it is not a substitute for looking for more evidence.
Multiverse theories. Physicists who propose theories that require many unobservable universes to explain our one observable universe are challenged by Occam's Razor: the theory multiplies unobservable entities enormously. The defenders respond that the number of fundamental laws is smaller in the multiverse theory — it is a disagreement about what "simplicity" means when some entities are observable and some are not.
Historical context
c. 320 BCE. Aristotle states something like the principle: "We may assume the superiority, other things being equal, of the…
c. 320 BCE. Aristotle states something like the principle: "We may assume the superiority, other things being equal, of the demonstration which derives from fewer postulates."
c. 1320. William of Ockham develops the principle most systematically. He did not apply the metaphor of a razor; he wrote in technical Latin about the conditions under which postulating additional entities is justified.
14th–17th centuries. The principle is widely used in medieval and early modern philosophy. Johannes Kepler uses parsimony in choosing ellipses over epicycles. Isaac Newton explicitly cites parsimony as one of his four "Rules of Reasoning in Philosophy" in Principia Mathematica (1687).
19th century. The name "Occam's Razor" appears in print. William Hamilton coins the razor metaphor around 1852.
20th century. Karl Popper, in developing falsificationism, treats parsimony as a desideratum of scientific theories. Bayesian statisticians formalise the razor as a consequence of probability theory.
Why it matters
Scientific reasoning. Parsimony is one of the core principles of scientific explanation. It prevents the multiplication of…
Scientific reasoning. Parsimony is one of the core principles of scientific explanation. It prevents the multiplication of theoretical machinery beyond what evidence requires, keeping theories falsifiable and comparisons clean.
Medicine. Diagnostic reasoning under Occam's Razor — look for one cause before postulating two — prevents over-investigation and over-treatment. The competing principle, "Hickam's dictum," notes that patients can have multiple diagnoses, and the rare thing does sometimes happen. Both are needed.
Epistemology. When we cannot determine truth directly, we need a principle for choosing among equally-fitting explanations. Occam's Razor is one of the most durable such principles — not because simplicity guarantees truth, but because complexity, without evidential support, is a form of claiming more than you know.
Modern relevance
Machine learning regularisation. Every machine learning framework that uses L1, L2, or elastic net regularisation is implementing…
Machine learning regularisation. Every machine learning framework that uses L1, L2, or elastic net regularisation is implementing a version of Occam's Razor: penalise unnecessary complexity in the model. Without regularisation, models overfit. Regularisation is the razor made computational.
Sycophancy in language models. Models that learn to agree with users, add unnecessary caveats, or generate elaborate but unsupported explanations are violating Occam's Razor: they are adding explanatory machinery beyond what the evidence (the training data and the prompt) requires. The cleanest model of the user's intent is usually also the most honest one.
Conspiracy theories. Conspiracy theories systematically violate Occam's Razor: they require many additional entities (the conspirators, their coordination mechanisms, their motivations, their silence) where a simpler explanation (the ordinary, uncoordinated explanation) fits the evidence equally well. The razor is one of the clearest tools for articulating why conspiracy theories should have low prior probability.
Sources & further reading
Encyclopedia
(2019)William of OckhamStanford Encyclopedia of Philosophy.
Comprehensive overview of Ockham's logic and the parsimony principle, with careful attention to what Ockham actually wrote.
How this connects
- Related to
- Goodhart's LawGoodhart's Law concerns what happens when you optimise a proxy beyond its range. Occam's Razor is a principle about not adding assumptions beyond what is needed — which is the prior that makes Goodhart's Law meaningful.
- Hanlon's RazorHanlon's Razor is a specific application of Occam's Razor to the question of intent. The malicious explanation posits an additional entity (evil intent) that the incompetence explanation does not.