HOW INTELLIGENCE WORKS · EXPLAINING AWAY · eduKateSG
How Evidence for One Cause Changes the Probability of Another Cause Competing to Explain the Same Effect
Explaining away is the intelligence process in which two possible causes compete to explain the same observed effect. Once strong evidence supports one cause, the other cause may become less necessary and therefore less probable, even though the two causes were independent before the effect was observed.
Possible cause A + possible cause B → shared effect observed → both become plausible → evidence for A arrives → A explains more of the effect → support for B falls.
This article belongs to the How Intelligence Works series. Causal Reasoning owns separating causation from correlation. Diagnosticity Reasoning owns asking whether evidence favours one explanation over rivals. Explaining away owns the specific intercausal update that occurs when competing causes meet at one observed consequence.
Two Independent Causes Can Become Dependent After We Observe Their Common Effect
Suppose either a faulty battery or an alternator problem can cause a car not to start. Before seeing the failure, the two causes may be largely independent. Once the car fails to start, both become more plausible. If a battery test then shows the battery is dead, the alternator problem becomes less necessary as an explanation.
The causes compete because the effect already has an explanation.
An explanation does not merely fit the evidence. It can also change how much explanatory work remains for its rivals.
1. The Structure Begins With a Common Effect
Explaining away requires at least two candidate causes capable of producing the same observed effect. The effect acts like a meeting point.
Before the effect is observed, learning about one cause may tell us little about the other. After the effect is observed, evidence for one cause can change our belief about the other because the explanatory burden is shared.
2. The Update Depends on How Completely One Cause Explains the Effect
| Evidence state | Effect on competing cause |
|---|---|
| Effect observed, no cause evidence | Several causes may become more plausible |
| Strong evidence for cause A | Cause B may lose probability because less unexplained effect remains |
| Cause A only partly explains the effect | Cause B may remain relevant |
| Causes can co-occur frequently | Explaining-away effect may be weaker |
| Evidence for A is unreliable | Support for B should not fall much |
| Effect is much larger than A can account for | B or another cause may still be needed |
3. Explaining Away Is Not the Same as Proving the Rival Impossible
Evidence for one cause usually reduces the need for another explanation; it does not necessarily drive the rival probability to zero.
Multiple causes may operate together. Strong reasoning therefore asks whether the supported cause is sufficient, partial, or compatible with co-causation.
4. Common Effects Create a Special Kind of Dependency
When two independent causes feed into one effect, observing the effect couples our beliefs about the causes. This structure is sometimes represented as a collider in causal models.
The key reasoning lesson is practical: once we condition on a shared consequence, relationships among possible causes can change.
Evidence can create dependence in our inference even when the causes were independent in the world.
5. Explaining Away in Mathematics and Formal Models
The mechanism appears most clearly in probabilistic graphical models. Competing causes point toward a shared observation, and new evidence about one cause changes the posterior probability of the other.
For learners, the deeper lesson is not the notation. It is that probability updates depend on structure: what causes what, which variables were observed, and which paths become informative after conditioning.
6. Explaining Away in Learning Diagnosis
A student may fail a question because of weak vocabulary, an unstable mathematical concept, attention loss or arithmetic error. The wrong answer makes several causes plausible.
If a follow-up item removes the language load and the learner still fails in the same structural way, vocabulary explains less. If the learner succeeds immediately, the language explanation becomes stronger and some competing causes become less necessary.
Diagnostic teaching therefore uses evidence not only to support one cause, but to reduce the plausibility of alternatives.
7. Competing Causes Should Not Be Treated as Mutually Exclusive by Default
Real outcomes often have several contributing causes. A project’s delay may involve poor estimation and supplier failure. A student’s weak performance may involve both knowledge gaps and time pressure.
The intelligent update asks how much explanatory work one cause performs, not whether discovering it gives permission to stop thinking.
8. Explaining Away Can Also Mislead
If one cause is accepted too quickly, the mind may discount a second genuine cause prematurely. This is especially dangerous when the first explanation is familiar, salient or socially convenient.
A good safeguard is to ask whether the observed magnitude, pattern and timing are fully accounted for by the favoured cause.
9. Explaining-Away Failure Atlas
| Failure | What happens | Repair |
|---|---|---|
| First-cause capture | The first credible cause crowds out all rivals | Ask whether it fully explains the effect |
| Mutual-exclusion error | Possible co-causes are treated as impossible together | Model joint causation explicitly |
| Weak-evidence overupdate | Unreliable support for A sharply suppresses B | Weight source reliability |
| Magnitude neglect | A explains some but not enough of the observed effect | Compare expected and actual magnitude |
| Structure blindness | Probability is updated without considering causal graph structure | Map the common effect |
| Convenient explanation bias | A socially easy cause explains away a harder one | Protect rival hypotheses until tested |
| Residual neglect | Unexplained variance is ignored after one cause is found | Track what remains unexplained |
10. Explaining Away and Diagnosticity Reasoning Are Different
Diagnosticity reasoning asks whether evidence is more expected under one explanation than another. Explaining away describes a specific change in relative belief when competing causes share an observed effect and one cause gains independent support.
Diagnosticity is the comparison principle. Explaining away is one causal pattern that makes the comparison change.
11. Teams Need to Record What Remains Unexplained
After a team identifies one credible cause, it is tempting to close the investigation. A stronger review asks what observations still sit outside the favoured explanation.
A cause earns closure only to the extent that it actually closes the explanatory gap.
12. Institutions Need Multi-Cause Incident Models
Complex failures rarely belong to one actor or one mechanism. Incident reviews should allow several causes to coexist and distinguish triggering events from enabling conditions, latent weaknesses and recovery failures.
This prevents a visible proximate cause from explaining away the deeper system conditions that made the event possible.
13. Artificial Intelligence and Competing Causal Explanations
AI systems can generate several plausible causes for one observation. The difficult step is updating those causes when new evidence supports one of them.
A reliable agent should not simply delete rivals after one supportive clue. It should ask whether the favoured cause is sufficient, whether causes can co-occur, and what residual observations remain unexplained.
Maintaining a structured causal hypothesis set is safer than allowing one fluent explanation to monopolise the answer.
14. The Explaining-Away Audit
- Effect: What observation needs explanation?
- Causes: Which independent or partially independent causes could produce it?
- Prior: How plausible was each cause before observing the effect?
- Common effect: Does observing the effect make several causes more likely?
- New evidence: What independent support has arrived for one cause?
- Sufficiency: How much of the effect can that cause explain?
- Co-causation: Can the causes operate together?
- Residual: What remains unexplained after accepting the favoured cause?
- Reliability: How trustworthy is the supporting evidence?
- Next test: What observation would distinguish residual competing causes?
15. CivDJ Reading: One Master Can Reduce the Need for Another Without Erasing It
In the CivDJ frame, several Masters may explain the same World Return. When a new signal strongly supports one Master, competing Masters should lose weight only in proportion to how much of the return has actually been explained.
If unexplained structure remains, another Master may still be active.
Do not mute the rival merely because one channel got louder. First ask whether the mix is fully accounted for.
16. Return to the Competing Cause
Explaining away shows that causal reasoning is relational. Evidence for one explanation changes the role of others because they compete to account for the same consequence.
The intelligent update is neither “one cause explains everything” nor “every cause remains equally plausible.” It asks how much explanatory work has been completed and how much remains.
The mature mind accepts a good explanation without allowing it to erase genuine co-causes or residual evidence.