The question
If a conditioned response declines when the relationship that produced it changes, has the original learning been erased — or has something further been learned?
Definition
Extinction occurs when previously learned responding declines after the conditions supporting that response change, such as when a conditioned cue repeatedly occurs without its expected outcome. The resulting change involves additional learning that alters when and how earlier associations are expressed, rather than deletion of the original learning; aspects of that earlier learning can remain behaviourally relevant.
When a learned response becomes weaker, the explanation can seem obvious.
The organism learned something.
Conditions changed.
The response declined.
So perhaps the original learning was simply removed.
This interpretation is especially tempting when the response becomes so small that it appears to have disappeared.
But behaviour changing and learning being erased are different claims.
A response tells us something about what the organism is doing under particular conditions.
It does not necessarily tell us everything that remains from its learning history.
Extinction makes this distinction unusually clear.
response reduction ≠ learning erased
To understand why, we first need to look at what actually happens during extinction.
Extinction Begins With a Changed Relationship
As established in D5.3 — Classical Conditioning, events can acquire significance through their relationships with other events.
Suppose a cue repeatedly predicts an outcome.
Through experience, that relationship can influence subsequent responding.
Now change the relationship.
The cue still occurs.
But the outcome that previously followed it no longer does.
In classical conditioning, repeatedly presenting a conditioned cue without the outcome it previously predicted is the canonical extinction procedure. Across this changed experience, conditioned responding typically declines.
At the behavioural level, that decline is clear.
But notice what the organism is actually encountering.
Not nothing.
A changed relationship.
The cue that previously predicted an outcome is now repeatedly occurring without it.
That new experience can itself support learning.
This is why extinction should not be imagined as learning simply fading because nothing is happening.
Something important is happening:
the contingency has changed
As established more broadly in D5.2 — Associative Learning, experience with relationships among events can alter subsequent responding.
Extinction extends that principle.
The organism learned from the original relationship.
It can also learn when that relationship changes.
extinction ≠ forgetting
Extinction is not simply the passage of time making earlier learning less accessible. It involves new experience under a changed contingency.
Behaviour Does Not Reveal the Whole Learning State
As extinction continues, conditioned responding may become substantially weaker.
That is evidence of real behavioural change.
But what exactly does the behavioural observation tell us?
Suppose a conditioned response falls almost to zero.
We observed:
very little conditioned responding under the current conditions
We did not directly observe:
the original association being deleted
The second statement is an inference about what produced the first.
This distinction follows from D5.1 — What Is Learning?
Learning and current behavioural performance are related, but they are not identical.
observed behaviour ≠ complete measure of learning
This is also a measurement problem.
As established in D2.2 — Measurement and Operational Definition, conclusions depend partly on what was actually measured.
If we measure conditioned responding, we have measured behaviour.
We have not directly measured the complete associative history responsible for that behaviour.
This caution works in both directions.
We should not see a response decline and simply declare:
"The earlier learning is secretly still there completely unchanged."
That would infer more than the behaviour tells us too.
Instead, we need additional evidence.
And extinction research gives us exactly that.
If Extinction Erased the Old Learning, Why Can Responding Return?
After conditioned responding has declined substantially during extinction, it can under some later conditions become evident again.
It can return after time has passed.
It can return when contextual conditions change.
It can return following later exposure to the original outcome.
And when the original cue–outcome relationship is restored, conditioned responding can sometimes be reacquired relatively rapidly.
These are different phenomena produced under different conditions.
Together, they create a serious problem for simple complete-erasure accounts.
return of responding challenges simple erasure
More precisely, reliable return of previously reduced responding is difficult to reconcile with extinction as complete deletion of all behaviourally relevant prior learning.
But even this conclusion requires care.
The return of a response does not allow us to look directly inside the learning system and declare:
"The original association was perfectly preserved and never changed."
Return provides evidence that prior learning history remains relevant to later behaviour.
It does not tell us that extinction left every aspect of that earlier learning untouched.
The evidence therefore supports a more precise conclusion:
Aspects of earlier learning can remain available after extinction even when conditioned responding has been substantially reduced.
That is enough to change how extinction must be understood.
Time and Context Can Change What Is Expressed
Two return phenomena make this especially clear.
The first is spontaneous recovery.
After conditioned responding has declined during extinction, some responding can return following a delay.
The word spontaneous does not mean the response appeared without causes or that every extinguished response eventually returns.
It refers to the reappearance of responding after time has passed.
The important lesson is:
expression can change across time after extinction
This is one reason D1.5 — Change Is a Process, Not a Moment matters here.
What behaviour looks like immediately after extinction does not necessarily tell us what it will look like later.
The second phenomenon is renewal.
A conditioned response reduced during extinction can sometimes return when testing occurs under contextual conditions different from those supporting extinction.
Again, the conclusion should remain conditional.
A context change does not automatically produce renewal.
Nor does renewal prove that extinction learning exists only in one particular place.
It provides evidence that contextual conditions can influence which previously learned patterns of responding become expressed.
Context here does not simply mean a physical room. Contextual cues and conditions can alter how prior learning influences current responding.
This also connects to D1.3 — State, Trait and Recurring Pattern.
A response appearing again under changed conditions does not automatically establish that the earlier change never happened.
Behaviour can vary because the conditions influencing its expression have changed.
Later Events Can Change Responding Again
Time and context are not the only things that matter after extinction.
What happens later can matter too.
One example is reinstatement.
After extinction, exposure to the original outcome can, under some conditions, contribute to the return of conditioned responding when the cue is later encountered.
This does not mean that simply presenting the outcome automatically restores the response.
Reinstatement depends on the conditions under which these events occur.
Its larger lesson is:
Later experience can change how previously reduced responding is expressed.
Another phenomenon is reacquisition.
Suppose the original cue–outcome relationship is restored after extinction.
Conditioned responding can sometimes return relatively rapidly.
That finding is difficult to reconcile with the idea that extinction necessarily deleted the organism's entire earlier learning history and forced acquisition to begin from nothing.
But rapid reacquisition still does not prove that the original association remained completely unchanged.
Reacquisition is itself another learning episode, occurring in an organism with a history of both acquisition and extinction.
The defensible conclusion is narrower:
prior learning history can remain relevant to subsequent learning
Spontaneous recovery, renewal, reinstatement and reacquisition therefore point in the same broad direction.
But they are not the same phenomenon.
Time changes in spontaneous recovery.
Context changes in renewal.
Later outcome exposure matters in reinstatement.
The original contingency returns in reacquisition.
Their shared lesson is not that one hidden mechanism restores the old response.
It is that:
behavioural reduction during extinction does not require complete deletion of learning history
shared challenge to erasure ≠ shared mechanism
Extinction Is New Learning—But Not One Universal Mechanism
We can now return to the title.
Why say:
Extinction Is New Learning
Because extinction exposes the organism to new information about a relationship that previously mattered.
During acquisition, the cue predicts an outcome.
During extinction, the cue repeatedly occurs without that outcome.
The organism can learn from this changed contingency.
That additional learning changes subsequent responding.
But the phrase new learning should not be mistaken for one settled mechanistic theory.
Some influential accounts emphasize inhibitory or competing learning.
On such accounts, extinction involves learning that reduces expression of previously acquired responding without requiring deletion of the earlier relationship.
Other accounts emphasize changes in prediction or expectation.
Repeated experience of the cue without the previous outcome can alter learned predictions about what follows the cue.
This does not require that those predictions be consciously represented or verbally reportable.
Contextual-control accounts emphasize how circumstances influence behavioural expression after extinction.
And other models allow extinction to modify aspects of prior learning itself.
These possibilities need not all be mutually exclusive.
The important hierarchy is:
simple complete erasure is inadequate
additional learning is central
the exact mechanism remains evidence-sensitive
Rejecting erasure does not require replacing it with one universal inhibitory mechanism.
And it does not require treating the original learning as permanently frozen.
persistence ≠ immutability
Extinction experience adds to and can modify the organism's learning history.
Earlier and Newer Learning Can Both Remain Relevant
A useful way to think about extinction is that learning history can contain more than one behaviourally relevant relationship.
Earlier experience established one relationship.
Extinction provides additional experience under changed conditions.
Both can remain relevant to what happens later.
This does not require imagining two fixed memory files fighting for control.
The underlying mechanisms may be more complex.
The important point is functional:
Multiple learned relationships can remain behaviourally relevant and influence responding differently under different conditions.
This helps explain why responding can remain low in one situation and become more evident in another.
It also helps explain why stable behavioural change does not necessarily require deleting everything learned before it.
As D1.7 — Stability Is Actively Produced makes clear, persistence itself requires explanation.
In the present case, stable reduced responding can remain dependent on conditions that support its expression.
That does not mean extinction learning requires constant active effort.
It means that behavioural stability under one set of conditions does not establish that earlier learning has disappeared.
Nor does the possibility of later recurrence make the changed responding unreal.
Extinction experience adds to the organism's learning history.
What gets expressed afterward can remain conditional.
A Returning Response Does Not Mean Nothing Changed
This point has consequences for how recurrence should be interpreted.
Suppose a conditioned response declines substantially during extinction.
Later, it returns.
One interpretation is:
"The extinction failed."
But that conclusion moves too quickly.
The return of responding provides evidence that prior learning history remains behaviourally relevant under the current conditions.
It does not demonstrate that no learning occurred during extinction.
recurrence ≠ failed learning
Likewise:
recurrence ≠ absence of change
Extinction learning and later return of responding can coexist.
The organism can genuinely learn from changed conditions while earlier learning remains capable of influencing later behaviour.
This does not mean recurrence is inevitable.
It does not mean extinction learning is permanent.
And it does not mean every recurring real-world behaviour is an example of Pavlovian extinction.
Laboratory extinction provides a narrower but powerful principle:
Behaviour can change without learning history being erased.
That principle is what later explanations of recurrence and relapse can build upon.
The Grounded Insight
When a learned response becomes weaker, something important has changed.
But the behavioural change does not tell us, by itself, that the earlier learning was deleted.
In classical extinction, the organism encounters a changed relationship.
A cue that previously predicted an outcome repeatedly occurs without it.
That experience can produce additional learning.
Conditioned responding typically declines.
And under some later conditions, previously reduced responding can become evident again.
Time can matter.
Context can matter.
Later exposure to an outcome can matter.
Restoring the original contingency can matter.
These findings are difficult to reconcile with a model in which extinction simply wipes the learning history clean.
But the alternative is not:
"Nothing happened to the earlier learning."
Extinction and subsequent experience can modify the broader learning history while aspects of earlier learning remain available.
So the useful distinction is:
extinction ≠ erasure
and:
extinction ≠ forgetting
The more complete principle is:
Extinction is additional learning under changed contingencies that alters behavioural expression without requiring complete deletion of previously acquired learning.
That is why an extinguished response can remain reduced under some conditions and return under others.
And it is why the return of an earlier response does not automatically mean that no meaningful learning occurred in between.
Change can be real without requiring the organism to lose its learning history.