Core Primitive
When you stop rewarding a behavior it temporarily intensifies before declining — expect this.
The vending machine that stole your dollar
You put a dollar into a vending machine and press B4. Nothing happens. You press B4 again — harder this time, as if the force of your finger on the button could communicate urgency to the mechanism behind the glass. Still nothing. You press it three more times in quick succession. You press B3, then B5, exploring neighboring buttons you never intended to select. You slam the side of the machine with your palm. You rock it slightly, tilting the front edge off the ground. You press the coin return. You press B4 one final time, holding it down for a full three seconds. Then you walk away.
What just happened was not a tantrum. It was not irrational. It was an extinction burst — one of the most predictable, well-documented, and universally experienced phenomena in behavioral science. You performed a behavior (inserting money and pressing a button) that has been reliably reinforced in the past (the machine delivered a snack). When the reinforcement failed to arrive, you did not immediately stop. You escalated. You increased the frequency of the behavior, the intensity of the behavior, and the variety of the behavior. You tried harder, tried faster, and tried different variations of the same fundamental action. Only after the burst ran its course did you disengage.
Every behavior you have ever tried to extinguish follows this same arc. And the burst is where most extinction attempts die — not because the discomfort is intolerable, but because people do not know it is coming and mistake it for evidence that extinction is not working.
What an extinction burst actually is
In Extinction requires removing the reward, you learned that extinction requires removing the reward that maintains a behavior. The mechanism sounds simple: stop reinforcing the behavior, and the behavior declines. But between "stop reinforcing" and "behavior declines" lies a valley that B.F. Skinner first documented in his operant conditioning chambers in the 1930s and that every subsequent researcher has confirmed. When reinforcement is first withdrawn, the behavior does not decrease. It increases.
Skinner observed this with his laboratory pigeons. A pigeon trained to peck a key for food pellets, upon finding that pecking no longer produced pellets, did not simply stop pecking. The pigeon pecked faster, pecked harder, and pecked with greater variability — sometimes striking the key at different angles, sometimes pecking at nearby areas of the chamber wall. The rate of responding surged before it fell. Skinner noted this pattern with precision but did not give it a formal name. Later researchers did: the extinction burst.
The formal definition is straightforward. An extinction burst is a temporary increase in the frequency, intensity, duration, or variability of a previously reinforced behavior immediately following the removal of reinforcement. The key word is temporary. The burst is not the new baseline. It is a spike — a last, escalated attempt by the behavioral system to produce the reward it expects. Dorothea Lerman and Brian Iwata, in their foundational research on extinction procedures, characterized the burst as consisting of several distinct features that co-occur.
First, frequency increases. The behavior happens more often than it did during the reinforcement period. A child who whined once to get a cookie, upon finding that whining no longer works, whines five times, ten times, continuously. A person who habitually checked email every twenty minutes, upon attempting to stop, finds the urge surfacing every five minutes.
Second, intensity increases. The behavior becomes louder, stronger, more forceful. The child does not just whine — the child screams. The vending-machine user does not just press the button — they slam the machine. Your urge to check your phone does not just linger — it produces genuine agitation, a restlessness that feels disproportionate to the situation.
Third, duration increases. Individual instances of the behavior last longer than they did before. A thought pattern you are trying to extinguish does not just flash and pass — it settles in and loops, occupying mental bandwidth for minutes rather than seconds.
Fourth, and most interesting, variability increases. The organism does not simply repeat the old behavior more aggressively. It generates novel variations — behaviors it has never performed before, all organized around the same functional goal of producing the missing reward. The pigeon pecks at new locations. The child who whined now tries bargaining, then crying, then throwing something, then being excessively sweet. Your mind, denied its habitual checking behavior, generates new rationalizations and "just this once" arguments it has never produced before. The burst is creative. The behavioral system is searching the space of possible actions for one that might restore the reward.
Why bursts happen: the logic of trying harder
Extinction bursts are not malfunctions. They are the correct response of a well-calibrated reinforcement-learning system to ambiguous reward withdrawal.
Consider the problem from the organism's perspective — and in this context, "organism" includes you. A behavior has been reinforced many times. The association between behavior and reward is strong. Now, on one occasion, the reward does not arrive. The system faces an inference problem: Is the reward permanently gone, or did something temporarily interfere? If the vending machine has reliably dispensed snacks a hundred times, one failure is far more likely to be a mechanical glitch than a permanent policy change. The rational response is to try again, try harder, and try variations, because any one of those variations might unstick the temporary glitch.
Hyung-Seok Goh and Brian Iwata, in their 1994 analysis of extinction burst characteristics, found that burst magnitude correlates with reinforcement history. Behaviors that have been reinforced many times and on a dense schedule produce larger bursts than behaviors reinforced few times or intermittently. This makes sense from the inference perspective: the stronger the evidence that the behavior "should" work, the harder the system pushes before concluding that the contingency has changed.
This is also why the burst eventually ends. The organism is not infinitely persistent. Each unreinforced attempt updates the system's estimate of whether the reward is truly gone. After enough unreinforced attempts — and this is critical — the system's estimate tips from "temporary glitch" to "genuine contingency change," and the behavior begins to decline. The burst is the period during which the system is gathering evidence for the new contingency. It is learning in progress. It feels like failure. It is actually the mechanism by which extinction works.
The catastrophic trap: reinforcing during the burst
Here is where theory becomes survival. The extinction burst creates a perverse incentive structure that, if you do not understand it, will cause you to inadvertently train a behavior that is far more resistant to extinction than the one you started with.
The scenario unfolds like this. You decide to stop reinforcing a behavior. The burst begins. The behavior escalates, creating pressure — emotional, social, logistical. At some point during the burst, usually at peak intensity, you give in. You check the phone. You give the child the cookie. You engage the compulsive thought.
You have just done something catastrophic. You have not merely failed to extinguish the behavior. You have taught the behavioral system that escalation works. Specifically, you have reinforced the behavior at the burst level of intensity, which means the system has now learned: "Normal-level behavior does not produce the reward, but burst-level behavior does." The next time you attempt extinction, the behavior will go directly to burst intensity and stay there longer, because that is the level at which reinforcement was last delivered.
This is the mechanism by which intermittent reinforcement creates persistent behavior, applied at the worst possible moment. Cooper, Heron, and Heward, in their comprehensive applied behavior analysis textbook, describe this as the single most common clinical error in extinction procedures. Practitioners who begin extinction, encounter the burst, become alarmed at the escalation, and reinstate the reward do not return to the pre-extinction baseline. They create a more intense and more resistant version of the behavior. The child who was given a cookie during a screaming tantrum has learned that screaming at that intensity is the price of a cookie. The next tantrum will reach that intensity faster and persist at that intensity longer.
The same mechanism operates when you are extinguishing your own behaviors. You decide to stop doomscrolling before bed. For three nights, the urge is manageable. On night four, the urge peaks — your mind races, the phone on your nightstand feels like it has its own gravitational field — and you pick it up "just to check one thing." You have now taught your reward system that burst-level craving is the price of the reward. The next attempt will be harder, not because you lack willpower, but because your system has updated its model of what intensity of effort the reward requires.
Predicting the burst: timing and magnitude
Bursts are not random. They follow predictable patterns you can anticipate and prepare for.
Lerman and Iwata's research established several empirical regularities. The burst typically begins within the first few instances of unreinforced behavior — often immediately after the first or second unreinforced trial. It does not build slowly over days. It hits fast. If you stop reinforcing a behavior today, the burst is most likely to peak today or tomorrow, not next week. Goh and Iwata found that in controlled settings, bursts typically resolved within the first extinction session, though they could recur at diminished intensity in subsequent sessions (a phenomenon related to spontaneous recovery, which Relapse is part of extinction will address).
The magnitude of the burst is proportional to the density and consistency of prior reinforcement. A behavior reinforced every single time produces a sharper but shorter burst than one reinforced intermittently. This is counterintuitive but important: variable reinforcement schedules make behaviors harder to extinguish overall, but their initial burst may be less dramatic because the system is already accustomed to some unrewarded trials. Continuously reinforced behaviors produce the most dramatic bursts because the system has never experienced a failure — the mismatch signal is maximal.
For practical purposes, ask yourself before beginning: How consistently has this behavior been reinforced? How long has the reinforcement history been? "Very consistently" and "a long time" means an intense but relatively brief burst. "Intermittently" and "a long time" means a less dramatic burst but a longer overall extinction timeline.
Surviving the burst: a preparation protocol
The burst is survivable, temporary, and predictable. But it requires advance preparation, because in the moment — when every signal in your nervous system is telling you to give in — you will not have the cognitive resources to reason your way through it. The time to prepare is before the burst arrives.
First, name it in advance. Before you begin any extinction attempt, write down this statement: "When I stop reinforcing this behavior, it will temporarily get worse before it gets better. The worsening is called an extinction burst, and it is evidence that extinction is working, not evidence that it is failing." This is not a motivational affirmation. It is a cognitive reframe you are installing before the burst compromises your ability to think clearly.
Second, predict the specifics. Write down what the burst will look like for this particular behavior — frequency increases, intensity escalation, novel variations, accompanying emotional states. The more specific your prediction, the less destabilizing the burst will be when it matches your expectations. Lerman and Iwata's work suggests that the variability component is the most disorienting aspect, because novel behavioral forms feel like a new problem rather than a predictable phase. Predicting them in advance neutralizes the disorientation.
Third, engineer the environment. Put the phone in a different room during the predicted burst window. Empty the house of snacks during the first 72 hours. Schedule the burst period during a time when absorbing activities are available. You are not relying on willpower to survive the burst. You are making reinforcement structurally unavailable.
Fourth, establish a commitment device. Tell someone what you are doing and when you expect the burst. Better yet, create a structural commitment — delete the app, lock the credit card in a timed safe, remove the option to give in. The burst will generate creative rationalizations ("just this once," "I will restart tomorrow"), and these are the burst expressing itself through your cognitive system. Structural commitments do not require you to argue with yourself.
Fifth, set a timer. Based on your burst prediction, set a literal timer. When the burst peaks and every moment feels like an hour, the timer provides an external anchor: this will end. You need to endure until the timer goes off, not forever. Bounded struggles are dramatically easier to survive than open-ended ones.
The Third Brain
An AI assistant becomes a critical ally during the extinction burst, not because it can remove the discomfort, but because it can hold the frame when your own frame is collapsing.
Before you begin an extinction attempt, describe the behavior, its reinforcement history, and your plan to your AI assistant. Ask it to generate a burst prediction: when will the burst likely peak, what forms might it take, and how long is the acute phase likely to last? The AI can draw on the empirical regularities from extinction research and apply them to your specific situation, giving you a more calibrated prediction than your intuition alone would produce.
During the burst itself, the AI serves as an external cognitive system. When the burst generates rationalizations — "this is not working," "I will just do it one more time and then really stop" — you can present them to the AI and ask: Is this a genuine insight, or is this the burst talking? An AI that knows you are in a predicted burst window can apply appropriate skepticism to your in-the-moment reasoning, functioning as a commitment device with natural language processing.
The AI can also help you log burst data in real time. Noting each urge's intensity and form transforms the burst from something you are enduring into something you are observing. This shift from first-person suffering to third-person documentation is itself a form of cognitive defusion (which Cognitive defusion will formalize). You are still in the burst, but you are also studying it, and the studying creates distance that makes the enduring more tolerable.
From burst to silence: what comes next
The burst ends. It always ends. The frequency declines, the intensity drops, the novel variations stop appearing, and the behavior settles into gradual fading — the extinction curve Skinner first documented, declining asymptotically toward zero.
But surviving the burst does not mean extinction is complete. It means you have passed the most dangerous checkpoint. The behavior may resurface at lower intensity in subsequent days (spontaneous recovery, which Relapse is part of extinction will address). It may re-emerge in novel contexts where your extinction learning does not generalize. And crucially, the behavior may feel suppressed rather than truly extinguished — which raises the question the next lesson takes on directly. Extinction eliminates behavior by removing its maintaining reinforcement. Suppression blocks behavior while the reinforcement contingency remains intact. They look identical from the outside. They are fundamentally different processes with fundamentally different outcomes. Extinction is not suppression teaches you how to tell them apart.
Sources:
- Skinner, B. F. (1938). The Behavior of Organisms: An Experimental Analysis. Appleton-Century-Crofts.
- Lerman, D. C., & Iwata, B. A. (1995). "Prevalence of the Extinction Burst and Its Attenuation During Treatment." Journal of Applied Behavior Analysis, 28(1), 93-94.
- Goh, H., & Iwata, B. A. (1994). "Behavioral Persistence and Variability During Extinction of Self-Injury Maintained by Escape." Journal of Applied Behavior Analysis, 27(2), 173-186.
- Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied Behavior Analysis (3rd ed.). Pearson.
- Lerman, D. C., Iwata, B. A., & Wallace, M. D. (1999). "Side Effects of Extinction: Prevalence of Bursting and Aggression During the Treatment of Self-Injurious Behavior." Journal of Applied Behavior Analysis, 32(1), 1-8.
- Bouton, M. E. (2004). "Context and Behavioral Processes in Extinction." Learning & Memory, 11(5), 485-494.
Frequently Asked Questions