Reward Extinction

A simple explanation

You stopped giving a behavior its reward. The behavior — the urge, the reaching, the small ritual — did not stop. Not at first. For a while it actually got worse. Then, unevenly, with the occasional brief return, it began to fade.

This is reward extinction. It is what Pavlov's bell looked like once the food stopped following it: salivation did not vanish in a clean curve. It spiked, then declined, then briefly came back, then faded for good. The same arc runs in human reward-learning. The cue you stopped feeding does not simply go quiet. It protests first.

The trouble is not the mechanism. The trouble is that almost everyone misreads the protest as failure.

An everyday example

You decide to stop checking your phone first thing in the morning. The first morning is hard but manageable. The second morning, the reach is worse than the day before — your hand goes for the phone before your eyes open. By day four you are convinced this is harder than you thought, that maybe you are uniquely bad at this, that the habit is somehow stronger than you realised.

By day nine, the reach has quieted. You forget to think about it. Two weeks later, on a stressful Tuesday, the reach is back — sharp and specific, as if no work had been done. You feel disproportionately defeated. You assume you are back to square one.

You are not. The first arc was the extinction burst. The Tuesday return was spontaneous recovery. The mechanism is running exactly as it does in the literature. The misread of either one is what derails most changes.

Why does my craving get worse right after I quit?

Because the Reward System, having learned that a particular cue reliably delivers a particular reward, treats the missing reward as a delivery failure rather than a change in the rules. Its first response is not to update the rule. Its first response is to escalate — to push the behavior harder, on the theory that the reward will arrive if only the effort increases.

This is the extinction burst. It is not a sign that the change is failing. It is a sign that the conditioning was real, that the System noticed the cut-off, and that it is doing the only thing it knows to do before it accepts the new contingency. The burst is the system's last argument for the old contract.

The burst typically peaks within the first few days to two weeks of a sustained removal, then declines. It is steepest in behaviors that were most variably rewarded, because variable schedules are the most resistant to extinction — the System has learned to keep trying through dry spells.

The behavioral loop

A non-linear loop, in six stages:

1. Conditioned baseline — a cue reliably triggers a behavior; the behavior reliably produces a reward. The System has built a working contract.
2. Cut-off — the reward is removed from the cue, deliberately or by circumstance. The cue still fires. The behavior still runs. The reward does not arrive.
3. Extinction burst — within hours to days, the behavior intensifies. The System is trying harder. Subjectively, the pull feels stronger than before you started.
4. Decline — if the cut-off holds, the behavior begins to fade, unevenly. Some days are nearly clean. Others have flares. The trend is downward but jagged.
5. Spontaneous recovery — after days or weeks of quiet, the behavior briefly returns, often triggered by stress, novelty, or proximity to the original cue. The return is real but short-lived if not re-rewarded.
6. Durable extinction — if neither the burst nor the recovery is re-rewarded, the cue eventually goes quiet. The behavior is not erased — the System remembers it — but it no longer fires automatically. The contract is closed.

Emotional drivers

Four feelings, often layered and almost always misread:

• A specific worse-than-before frustration during the burst — I'm getting weaker, not stronger.
• A faint self-suspicion that the change is not going to hold, peaking around days three to seven.
• A small grief at the loss of the old contract, often unnamed and mistaken for general low mood.
• A flat relief in the quiet stretches that the system reads, incorrectly, as done.

What your nervous system does

The Reward System routes through the dopaminergic prediction system. Under a stable reward schedule, the system fires on the anticipation of reward, not the reward itself — this is why cues feel motivating even before they pay out. When the reward is removed, the prediction error grows: the system expected something and got nothing. A growing prediction error is what increases the drive. The burst is, mechanically, the prediction error doing its job.

Over repeated dry runs, the system updates its prediction downward. The cue becomes less motivating. The drive falls. But the original learning is still on file. A stress spike, a sleep deficit, or a return to the original context can briefly raise the prediction back up — this is spontaneous recovery, and the body experiences it as a sharp, specific, almost vivid return of the old pull.

The DojoWell interpretation

Reward extinction is the technical name for the work of decoupling from a substitute. Every Atlas entry that names a substitute behavior — every loop where the System was asked for an original system and supplied a mimic — eventually requires extinction to close. The substitute does not vanish because you understood it. It vanishes because the cue stopped paying out, the burst ran, the recovery flares were not re-rewarded, and the System closed the contract.

This is why extinction is the canonical delayed_harvest density signature. The effort is high and front-loaded. The deposit is real but does not appear during the burst. To the system being asked to do the work, it looks for several days like effort with no deposit — the exact shape that makes most well-intentioned change stop on day five.

There is a second, sharper distinction in the MDT frame: extinction is not the same as suppression. Suppression is the System holding the behavior down by force — vigilance, willpower, gritted teeth. Suppression does not update the contract. The moment the vigilance drops, the behavior returns at full strength. Extinction changes the contract itself. The decoupling is durable because the System has learned a new rule, not because anyone is holding the line.

This is why suppression-based change reliably fails and extinction-based change reliably holds. The work is harder during the burst, but the harvest is durable. The substitute that gets extinguished — not suppressed — is the one that does not come back to claim the original system.

How do I stop a conditioned reward pull?

Not by fighting it. By letting it fire without paying it. The System needs to log enough unrewarded firings to update its prediction. Every time you fight the cue with willpower, you are suppressing. Every time you let the cue fire and notice it pass without delivering the reward, you are extinguishing. The difference, subjectively, is small. The difference, durably, is total.

Three principles, in order:

1. Hold the cut-off, especially during the burst. The burst is not evidence of failure. It is evidence the mechanism is running. Reading it correctly is the difference between day seven and day three.
2. Do not re-reward spontaneous recovery. A single payout during a recovery flare can fully reinstate the contract. The System was almost done. One slip resets the timer not because you are weak but because variable reinforcement is exactly what makes the old loop most durable.
3. Distinguish the cue from the system underneath it. The conditioned cue is what extinction handles. The original system the substitute was masking — completion, recognition, novelty, presence — still needs a real-reward path. Extinction without recovery leaves a vacuum the System will eventually fill.

Practical steps

1. Before you start, name the cue and the reward explicitly. When [cue], my system expects [reward]. Most failed extinctions are vague about what is actually being decoupled.
2. Predict the burst in advance. Mark on a calendar that days three to ten will likely feel worse than day one. The prediction itself is protective — the burst that was anticipated cannot be misread as failure.
3. For the first two weeks, hold a single line: the cue may fire; the reward will not arrive. Do not try to also fix the underlying ask. Two changes at once stacks the load and risks both.
4. Plan for one spontaneous recovery, two to six weeks in. Decide now what you will do when it arrives. The System does not warn you. A single pre-made decision spares you a slip.
5. Track unrewarded firings rather than abstinence days. Each firing-without-reward is a unit of work the System is logging. Counting them reframes the burst from cost to progress.

Reflection questions

• Which cue in your life is most clearly still being paid out, even though you have decided to stop?
• When you last tried to change a behavior, did you stop during what was almost certainly the extinction burst?
• Have you mistaken a spontaneous recovery for "I'm back to square one" — and what would change if you read it as the brief return that the literature predicts?
• Where in your life have you been suppressing a behavior for years without ever letting the cue fire unrewarded long enough for extinction to do its work?

Frequently Asked Questions