Extinction in reinforcement learning is the gradual decline of a behavior when the reinforcement that previously maintained it is no longer provided. If a behavior used to produce a reward or a desired outcome, and that outcome stops happening, the behavior will eventually fade. The key idea is that extinction doesn’t erase the original learning. Instead, the brain lays down new learning that competes with the old, essentially telling the person or animal “that behavior doesn’t work anymore in this situation.”
How Extinction Works
Every learned behavior is sustained by some form of reinforcement, whether that’s food, attention, money, or relief from discomfort. Extinction begins the moment that reinforcement stops being delivered. A rat that learned to press a lever for food pellets will gradually stop pressing when the pellets stop coming. A child who learned that screaming in the grocery store leads to candy will eventually stop screaming if the candy never arrives.
What’s happening in the brain isn’t a simple deletion of the original memory. Research shows that extinction creates a new, competing association: the organism learns to inhibit the response in the specific context where reinforcement stopped. One way researchers describe this is straightforward: the animal learns “stop making this response in this context.” The original learning remains intact underneath, which is why extinguished behaviors can sometimes return.
The Extinction Burst
One of the most important things to know about extinction is that behavior almost always gets worse before it gets better. When reinforcement suddenly stops, there’s typically a temporary spike in the frequency, duration, or intensity of the behavior. This is called an extinction burst, and it can happen almost immediately after reinforcement is withdrawn.
Think of it this way: if a vending machine suddenly stops dispensing your drink after you insert money, you don’t calmly walk away. You press the button harder, press it multiple times, maybe shake the machine. That’s an extinction burst. The good news is these bursts are short-lived. They occur close to the onset of extinction and disappear with continued exposure. The bad news is that if someone gives in during the burst, they’ve just reinforced a more intense version of the behavior.
Frustration and Aggression During Extinction
The extinction burst isn’t the only side effect. The sudden loss of expected reinforcement is genuinely aversive, and it can trigger emotional responses like frustration and even aggression. Classic experiments showed that pigeons attacked nearby pigeons (and even stuffed pigeon models) at the onset of extinction after a period of food reinforcement. The transition from “this works” to “this no longer works” creates real emotional distress.
This frustration response matters in practical settings. A child whose tantrums are being extinguished through planned ignoring may escalate to hitting or throwing things. A pet whose begging at the table is being ignored may bark louder or paw more aggressively. Understanding that frustration is a normal, predictable part of the process helps caregivers and trainers stay consistent rather than giving in at the worst possible moment.
Why Some Behaviors Are Harder to Extinguish
Not all behaviors fade at the same rate. One of the most reliable findings in behavioral science is the partial reinforcement extinction effect: behaviors that were reinforced inconsistently are far more resistant to extinction than behaviors that were reinforced every single time.
This seems counterintuitive, but the logic is elegant. If a behavior was rewarded every time it occurred, the absence of reward is immediately noticeable. The organism detects the change quickly. But if the behavior was only sometimes rewarded, the early stages of extinction feel identical to normal conditions. The organism needs many more unreinforced attempts before it can detect that anything has changed. Several theories explain why this happens. One proposes that the frustration of occasional non-reward during training actually becomes associated with continued responding, so frustration during extinction energizes the behavior rather than suppressing it. Another suggests that the memory of unrewarded trials followed by rewarded trials teaches the organism to persist through dry spells.
This has real-world implications. A slot machine pays out on an unpredictable schedule, which is why gambling behavior is notoriously difficult to extinguish. A parent who usually ignores tantrums but occasionally gives in has accidentally created a partial reinforcement schedule, making the tantrums far more persistent than if they’d rewarded them every time.
Spontaneous Recovery and Relapse
Because extinction doesn’t erase the original learning, extinguished behaviors can reappear. The most common form of this is spontaneous recovery: after a period of time away from the extinction context, the behavior resurfaces at a reduced level. This happens because the new inhibitory learning is tied to a specific context, and the passage of time effectively changes the temporal context.
A related phenomenon is renewal, where an extinguished behavior returns when the organism is placed in a different physical environment from where extinction took place. If a behavior was learned in one room and extinguished in another, returning to the original room can bring the behavior back. This context dependence is a central feature of extinction. The brain links the “stop responding” signal to the specific setting where reinforcement was withdrawn, so changing the setting can release the old behavior from inhibition.
What Happens in the Brain
Extinction learning relies on a circuit connecting three brain structures. The amygdala, which stores the original emotional and motivational associations, is where the initial plasticity for extinction takes place. The prefrontal cortex then develops the ability to suppress activity in the amygdala, essentially putting the brakes on the old learned response. The hippocampus provides contextual information, helping the brain determine whether “now” matches the conditions under which extinction was learned. When the prefrontal cortex successfully inhibits the amygdala, the old behavior stays suppressed. When context changes or time passes, that inhibition can weaken, allowing the original response to resurface.
How Extinction Differs From Punishment
Extinction and punishment both reduce behavior, but they work through completely different mechanisms. In extinction, the reinforcer is simply withheld: nothing happens after the behavior. In punishment, a consequence is actively delivered, either something unpleasant is added or something valued is taken away. A child who throws a tantrum and gets no reaction is experiencing extinction. A child who throws a tantrum and loses screen time is experiencing punishment.
Interestingly, both extinction and punishment produce context-dependent learning, meaning behaviors suppressed through either method can return in a new environment. Research comparing the two has found that negative punishment (removing a reward contingent on the behavior) and extinction can produce equivalent renewal effects, suggesting that a similar interference mechanism is at work in both cases. The practical difference is that extinction avoids introducing aversive consequences, which makes it preferable in many therapeutic and educational settings.
Planned Ignoring in Parenting
One of the most common real-world applications of extinction is planned ignoring, a technique used by parents and caregivers to reduce attention-maintained problem behaviors. The principle is simple: if a behavior is sustained by attention, withdrawing all attention should cause the behavior to fade.
In practice, this means not making eye contact, not speaking to the child, not changing your facial expression or tone of voice, and not providing any objects during the problem behavior. You wait until the child is calm before re-engaging. If the child wants something but isn’t requesting it appropriately, you ignore the inappropriate request and redirect them to another activity.
Consistency is everything. Nationwide Children’s Hospital guidelines emphasize that once you begin planned ignoring for a specific behavior, you must continue it every time that behavior occurs. Giving in even once, especially during an extinction burst, teaches the child that escalation works. It’s also critical to limit planned ignoring to behaviors that are undesirable but harmless. Hitting, biting, or other dangerous behaviors require a different approach. Having another adult for support helps, because watching a child’s behavior escalate before it improves is genuinely stressful, and the temptation to give in is real.