Occam’s Razor

“Entities should not be multiplied beyond necessity.” — William of Ockham (c. 1287–1347), paraphrased

Often rendered as: among competing explanations, prefer the one with the fewest assumptions — or more colloquially, the simplest explanation is usually correct.

The Claim

When multiple hypotheses fit the evidence equally well, prefer the one requiring fewer independent assumptions. Each additional assumption is a load-bearing claim that could fail; stacking them multiplies the odds of being wrong somewhere.

The razor does not say simple explanations are true. It says: under uncertainty, simpler hypotheses have a higher prior probability of being correct, and are easier to falsify if they are wrong.

Why It Holds

In Bayesian terms, Occam’s Razor is a prior over hypothesis space. Each additional free parameter expands the space of possible worlds the hypothesis could describe. A hypothesis with three free parameters “fits” many more observations than one with zero — which means, for any specific observation, the three-parameter hypothesis has less predictive mass per possible-world. Simpler hypotheses make sharper predictions and are cheaper to test.

This is formalised in Solomonoff induction and the Minimum Description Length principle in machine learning: prefer the hypothesis that compresses the data most efficiently. Modern Bayesian model selection (BIC, AIC) is Occam’s Razor with actual math attached.

In This Wiki

• Munger’s latticework. Munger explicitly names Occam’s Razor as a foundational mental model: when the facts are confusing, the simplest consistent story is usually right. Paired with Hanlon’s Razor (don’t attribute to malice what carelessness suffices to explain) it clears most organisational mysteries.
• Companion to first-principles-thinking. First-principles reasoning tells you what’s actually true at the bottom; Occam’s Razor tells you which of several consistent explanations to prefer when you don’t have enough data to pin down the truth.
• Companion to inversion. Inverting Occam: “how would I guarantee being wrong?” Answer: multiply assumptions, add exceptions, prefer complicated stories that protect a prior belief.
• Antidote to confirmation-bias. Confirmation bias favours complex stories that preserve existing beliefs; Occam favours simple stories that may overturn them. The simple story is also harder to defend with selective evidence — Occam is a Popperian “make it easy to falsify” posture.
• Connects to KISS and DRY in software design. Both are Occam’s Razor applied to code: prefer the simplest design consistent with the requirements; prefer the representation with the fewest redundant assumptions.
• Connects to SOLID Principles. The “S” (Single Responsibility Principle) is Occam at the class level: each unit should have one reason to change. The fewer reasons, the simpler.
• Respects Gall’s Law. Gall says complex working systems evolved from simple working systems. Occam’s Razor applied over time: start simple; add complexity only when the simple version proves insufficient.

Software Applications

• Debugging. When a system misbehaves, the cheapest hypotheses (off-by-one, null pointer, race condition, typo) are more likely than exotic ones (cosmic ray bit flip, compiler bug, hardware fault). Check the simple ones first.
• Architecture. Prefer the simplest architecture that meets known requirements. Every additional service, queue, cache, or database is a failure point and an operational burden. YAGNI is Occam applied prospectively: don’t add what you don’t need.
• Incident analysis. When the root cause could be “a rare combination of three unlikely events” or “one person forgot to click a button,” bet on the button.

Limits

1. “Simplest” is under-specified. Simpler in what sense — fewer entities, fewer parameters, shorter description? The razor is most rigorous when paired with a formal measure of complexity (Kolmogorov complexity, MDL).
2. Simple can mean wrong. Occam is a prior, not a likelihood. If evidence decisively supports a more complex hypothesis, Occam loses. Einstein’s general relativity is more complex than Newtonian gravity and also more correct.
3. Doesn’t apply uniformly to biology or social systems. Biological systems accrue complexity through evolution; the simple explanation is often wrong because evolution is not parsimonious. “Why is this gene here?” rarely has a clean answer.

The Einstein Corollary

Attributed (probably apocryphally) to Einstein: “Everything should be made as simple as possible, but no simpler.” Occam’s Razor with an explicit floor — simplicity is bounded below by adequacy. This is the engineering reading.

Sources

• source—laws-of-software-engineering — in the Decisions cluster.
• William of Ockham, Summa Logicae (c. 1323) — the original formulations.
• Ray Solomonoff (1964), Jorma Rissanen (1978) — formalisations as Solomonoff induction and Minimum Description Length.
• source—poor-charlies-almanack — Munger on Occam’s Razor as a foundational mental model.