Dopamine, Reward & How Habits Hijack You

Everything you have been told about dopamine is roughly one word wrong — but it is the load-bearing word. Dopamine is not what pleasure feels like. It is what wanting feels like. It is the signal your brain uses to mark something as worth going and getting again, and it fires hardest not when you are happy but when you are surprised. Below you can watch a dopamine burst walk backwards in time from a reward to the cue that predicts it — and then watch what happens when the reward you were promised does not arrive.

Try this: scroll to the Reward-Prediction-Error Lab below the brain and press 2 · Learning. Watch the dopamine spike physically walk backwards from the reward to the cue. Then press 4 · Reward withheld — and watch it dive below baseline. That dip is disappointment. That dip is craving. Then press 🎰 Variable reward and watch the error never settle.

The Science in Plain Language

1. Wanting and liking are two different systems — and dopamine only runs one of them

The cleanest way to see this is to take dopamine away and watch what survives. Mice engineered so that they cannot make dopamine at all stop eating. They will sit beside food and starve, and they die within weeks unless they are given L-DOPA. But if you put sugar in their mouths, they make exactly the same happy, rhythmic liking faces that a normal mouse makes. They still enjoy it. They have simply lost the thing that makes them go and get it.

The mirror image also works. Rats whose accumbens dopamine has been depleted are not miserable and not anhedonic — they still eat and they still prefer the tastier food. But asked to press a lever many times, or climb a barrier, to reach that tastier food, they give up and settle for the bland chow lying free on the floor. Dopamine is what buys effort. This is the core of Kent Berridge’s incentive salience theory: dopamine paints a thing with “go get that,” and that paint is not the same substance as the pleasure of having it.

Actual liking — the sweetness itself — is generated in surprisingly small hedonic hotspots in the nucleus accumbens shell and the ventral pallidum, and it runs on opioid and endocannabinoid signalling, not dopamine. Amphetamine injected into the accumbens makes an animal want a reward more without making it like the reward more. And that gap — wanting that has come loose from liking — is exactly what a person with an addiction is describing when they tell you, truthfully and with no self-pity, that they stopped enjoying it years ago and still cannot stop.

2. Prediction error: the spike that walks backwards in time

Wolfram Schultz put electrodes into the dopamine neurons of monkeys and found something no one expected. The neurons did not fire in proportion to how good the juice was. They fired in proportion to how much better things turned out than the animal expected.

Unexpected juice: a hard burst. Now sound a tone one second before the juice, every time, for a few hundred trials. Slowly, the burst leaves the juice and moves to the tone. Eventually the fully predicted juice arrives and the dopamine neurons say nothing at all — a flat line, as though nothing happened. The juice is exactly as sweet as it ever was. Dopamine simply has no comment, because there is no news.

And then the cruel, perfect experiment: sound the tone, and withhold the juice. At the precise millisecond the reward was due, the dopamine neurons pause — they drop below their normal background firing. That dip is a negative number. It is the brain computing what I got, minus what I expected, and getting a deficit. That dip is disappointment. It is also, played on a loop, craving.

One honest note about the animation above. The spike sliding smoothly backwards along the timeline is a teaching device — press ⇆ Show the real recordings and you will see what the data actually look like: the response to the reward shrinks while the response to the cue grows. The result is the same, and it is the important one: the dopamine signal migrates to the earliest reliable predictor of the reward. Your brain is not tracking pleasure. It is tracking surprise, and it is trying to run out of surprises.

3. Why uncertainty is the most powerful trainer ever devised

If dopamine tracked pleasure, then the most addictive thing in the world would be the most pleasurable thing in the world. It isn’t. The most addictive thing in the world is a reward you can’t predict.

Fiorillo, Tobler and Schultz showed this directly in 2003. When a cue predicts a reward with probability p, the burst at the cue scales with the expected value — as you would expect. But a second signal appears, a slow ramp of dopamine activity climbing between the cue and the moment of truth, and that ramp is largest when p = 0.5. Maximum uncertainty. A coin flip. The dopamine system leans furthest forward exactly when it has no idea what is about to happen.

B. F. Skinner had already found the behavioural half of this: a variable-ratio schedule — reward every nth press on average, but you never know which press — produces the highest, steadiest, most stubbornly extinction-resistant responding of any schedule he tested. It is the single most effective way to make an animal keep doing something.

Now list the things engineered on a variable-ratio schedule: a slot machine. A loot box. A pull-to-refresh feed that might have something good in it. A notification that might be someone who loves you or might be nothing. An infinite scroll where the next video might be the one. None of these are addictive because they are wonderful. They are addictive because they never let the prediction error reach zero. The teaching signal is never allowed to switch off, so the learning never completes, so you never stop. That is not a metaphor and it is not an accident — it is a design pattern, and it is chosen on purpose.

4. Inside the synapse: built, packed, released, recovered, destroyed

Dopamine starts as the amino acid tyrosine, which you eat, and which your body can also make from phenylalanine. Then:

1. Tyrosine → L-DOPA, by tyrosine hydroxylase (TH). This is the rate-limiting step — the narrow neck of the whole bottle. TH needs iron (Fe²⁺) in its active site and tetrahydrobiopterin (BH₄) as its cofactor, plus oxygen.
2. L-DOPA → dopamine, by aromatic L-amino acid decarboxylase (AADC), which needs vitamin B₆ in its active form, pyridoxal 5′-phosphate.
3. VMAT2 pumps the finished dopamine into vesicles, which protects it from being eaten by MAO and gets it ready to be fired.
4. An action potential arrives, vesicles fuse, and dopamine crosses the cleft to hit D1 receptors (the “go” direct pathway) and D2 receptors (the “stop” indirect pathway). D2 receptors also sit on the terminal itself as autoreceptors, telling the neuron to ease off — a built-in volume knob.
5. The dopamine transporter (DAT) then vacuums most of it straight back up for re-use. What escapes is broken down by MAO (to DOPAC) and COMT (to 3-MT), and ends up as HVA.

Two details with real clinical teeth. First: in the prefrontal cortex there is very little DAT, so COMT does most of the clearing — which is why the common COMT Val158Met variant noticeably shifts prefrontal dopamine and working memory while barely touching the striatum. Second: in Parkinson’s disease, the dopamine neurons of the substantia nigra die, and the treatment is L-DOPA rather than dopamine itself — because dopamine cannot cross the blood–brain barrier and L-DOPA can. It is given together with carbidopa, which blocks AADC outside the brain so the L-DOPA survives the journey. The whole pathway in the diagram above is, quite literally, a prescription.

And because TH needs iron: iron status matters. Low brain iron is implicated in restless legs syndrome, which responds both to iron repletion and to dopamine agonists — one of the tidiest demonstrations that a mineral you can measure sits inside a circuit you can feel.

5. What the drugs are actually doing (each one is a different sabotage)

Cocaine plugs the DAT. Dopamine gets released normally — it just cannot be recovered. It pools in the cleft and keeps hammering the receptors long after it should have been cleared.

Amphetamine is stranger and more violent. It is itself carried into the terminal by DAT, then it makes DAT run backwards, so dopamine is actively pumped out into the cleft. At the same time it collapses the gradient across VMAT2, so the vesicles empty their dopamine into the cell’s interior, where it can be shoved out through the reversed transporter. The result is dopamine release that does not need an action potential at all. The neuron is not firing. It is being drained.

Nicotine works upstream. It binds nicotinic acetylcholine receptors sitting on the VTA dopamine neurons themselves and depolarises them into burst firing — it presses the accelerator directly.

Opioids do the opposite and get the same result: the VTA is held in check by inhibitory GABA interneurons, and opioids switch those interneurons off. Take the foot off the brake and the car speeds up without anyone touching the accelerator — that is disinhibition. Alcohol is messier (it acts on GABA-A receptors, NMDA receptors, endogenous opioid release and more at once), but part of what it does to the VTA is that same trick.

One more thing, and it matters more than almost anything else on this page: the speed of the rise is as important as the size of it. The same molecule is far more addictive when it reaches the brain fast. A slowly absorbed oral dose produces a gentle ramp; smoked or injected, it produces a near-vertical wall. This is a large part of why a therapeutic, extended-release stimulant taken orally for ADHD is pharmacologically not the same event as stimulant misuse — and it is the honest answer to a question a lot of people are too embarrassed to ask out loud.

6. Downregulation: tolerance, anhedonia, and chasing baseline

Hammer any receptor system hard enough for long enough and it will defend itself. Under chronic overstimulation the striatum removes D2 receptors from the membrane — you can watch them vanish in the synapse view above. Brain-imaging (PET) studies consistently find lower striatal D2/D3 receptor availability in people with cocaine, methamphetamine, alcohol and heroin addiction than in matched comparison subjects, along with a blunted dopamine release to a given stimulus. The percentage sounds modest. Its effect is not.

Because two things happen at once, and the second one is the cruel one:

1. Tolerance. The same drug now produces less signal. So the dose climbs.
2. Anhedonia. Everything ordinary now produces almost none. Coffee with a friend, a walk, a song you used to love, your own child telling you about their day — all of it goes quiet at the same time.

Which means the thing everyone assumes is happening is not happening. The person is not chasing a high. Long ago they stopped chasing a high. They are chasing baseline. They are trying to feel the way you feel right now, reading this, for free.

Now the hopeful half, and it has to be told exactly straight or it is worthless. It recovers. Nora Volkow’s imaging work in people who had used methamphetamine found that dopamine-transporter binding — badly reduced during use — had not meaningfully recovered after about a month of abstinence, but had substantially recovered after roughly nine months or more (and even then, some cognitive deficits lagged behind). Studies of D2 receptor availability in cocaine users have found it still reduced after months of detoxification. So: receptor availability recovers slowly — over months, not days.

Read that as the good news it is. If you are three weeks in and the world still looks grey and flat and pointless, that is not failure and it is not proof that you are broken. That is the timeline. That is what the middle of it looks like from the inside. The receptors come back on a schedule that has nothing to do with how hard you are trying — and they do come back.

7. How a want becomes a habit: the migration from ventral to dorsal

At the start, a behaviour is goal-directed. It runs through the ventral striatum — the nucleus accumbens — and you do it because you want the outcome. If the outcome stopped being worth having, you would stop.

With enough repetition, control migrates upward and outward into the dorsal striatum, and the behaviour changes character completely. It becomes stimulus–response: the cue fires, the action happens, and the action no longer depends on wanting the outcome at all. The classic laboratory demonstration is outcome devaluation — make the reward worthless, and a goal-directed animal stops, while a habitual one carries right on performing. Barry Everitt and Trevor Robbins traced exactly this ventral-to-dorsal shift as drug-seeking became compulsive.

This is why “just decide to stop” keeps failing, and why the failure is so humiliating and so misread. The decision-making machinery is no longer the machinery running the behaviour. You are asking the prefrontal cortex to veto something that is not being routed through the prefrontal cortex any more. The hand reaches for the phone before there is a “you” involved.

Two consequences worth taking seriously. First: changing the environment beats out-willing the cue, every time, because it kills the trigger before the loop even starts. Leaving the phone in another room is not a confession of weakness — it is the correct engineering solution to a stimulus–response problem. Second: habits are not deleted, they are overwritten. The old track stays in the ground. That is why relapse under stress is so common, and it is a fact about striatal wiring, not a verdict on your character.

8. The honest supplement section

The precursors are real. Tyrosine and phenylalanine are genuinely the raw material. The cofactors are real. AADC genuinely requires vitamin B₆; tyrosine hydroxylase genuinely requires iron and BH₄; BH₄ availability is tied to folate status; and COMT can only break dopamine down using a methyl group donated by SAMe, which comes out of the methylation cycle. Every one of those is a real dependency, and a real deficiency in any of them is a real bottleneck worth finding and fixing.

And yet: taking tyrosine will not raise dopamine in a well-fed person. This is the part the supplement aisle leaves out. Tyrosine hydroxylase is already saturated with tyrosine at normal dietary concentrations, and its output is controlled by feedback inhibition and by phosphorylation — not by how much substrate you throw at it. Pouring more tyrosine onto a saturated, feedback-regulated enzyme achieves precisely nothing. The bottleneck is not the raw material; the bottleneck is the enzyme, and the enzyme is not asking for your help.

The genuine exception is narrow and worth knowing: under acute, severe catecholamine demand — sustained cold, prolonged sleep deprivation, high acute stress, long unbroken cognitive load — catecholamines can be depleted faster than they are replaced, and in those conditions tyrosine supplementation has produced modest cognitive benefits in laboratory and military stress studies. In a rested, well-fed person sitting comfortably, it does not. That is the entire rule, and it generalises: if you are deficient, correcting the deficiency helps, because a deficiency is a real bottleneck. If you are not, it does nothing. (Iron is the one genuinely worth checking — ferritin, if you have restless legs, unexplained fatigue, heavy periods, or a plant-based diet. That is a lab test, not a wellness product.)

“Dopamine fasting” is a misnomer, and the name has done real harm. You cannot fast from a neurotransmitter. If you genuinely lowered your brain dopamine you would not become serene and focused — you would become slow, rigid, and unable to start moving. That is Parkinson’s disease, not enlightenment. But the idea buried underneath the silly name is completely sound: deliberately cutting your exposure to supernormal, unpredictable, variable rewards genuinely works. It is stimulus control, it is decades old, it is one of the most reliable behavioural interventions there is, and it deserves a better name than the one it got. Keep the practice. Drop the pseudoscience. They are separable, and it is worth the effort of separating them.

9. What actually helps — and a note for anyone who needs it

Sleep. This is the least glamorous item on the list and the most load-bearing. Sleep deprivation has been shown to downregulate D2/D3 receptor availability in the striatum. Protecting your sleep is, quite literally, protecting your receptors. Nothing you can buy comes close.

Daylight and a stable rhythm. Not because sunlight “releases dopamine” — be suspicious of anyone who says that — but because this entire circuit rides on a circadian schedule, and a wrecked schedule wrecks the system riding on it.

Exercise. In animal studies regular exercise increases striatal D2 receptor expression, and in humans it is among the best-supported non-drug interventions for both depression and craving. It is also, conveniently, an effortful goal with a delayed payoff — which is the next item.

Effortful goals with delayed payoffs. Nobody wants to hear this one, so here is why it is not a moral lecture but a mechanism. Dopamine is the currency of effort — it is what makes an animal willing to climb the barrier for the better food. Every time you do something hard and it pays off later, you are training this system that effort predicts reward. Every hour on a variable, effortless, instant-reward feed trains it in exactly the opposite direction: that reward arrives without effort, unpredictably, and right now. You are not choosing between virtue and vice. You are choosing which rule your prediction machine gets to learn.

Real, physically present other people. Not a substitute for them.

And treatment, if that is what this is. Addiction is treatable, and there is medication that genuinely works: buprenorphine and methadone for opioid use disorder (they reduce mortality — among the most robust findings in all of medicine), naltrexone and acamprosate for alcohol, varenicline and nicotine replacement for smoking. And contingency management — small, immediate, certain rewards for verified abstinence — is the most effective treatment there is for stimulant use disorder. Notice what it is: it is this exact learning system, turned around and pointed the other way. The machine that got you here is the same machine that gets you out.

If you have read this far and recognised yourself somewhere in it, please hear the last thing plainly. What is happening in your brain is not a character defect. It is a healthy, well-functioning learning system doing precisely what it evolved to do — on an input it never evolved to meet. Addiction is a learning disorder in a normal brain. Learning can be re-done. It takes months, and it takes help, and both of those are completely allowed.

In the United States, the SAMHSA National Helpline — 1-800-662-4357 — is free, confidential, and staffed 24 hours a day, in English and Spanish. If you are in crisis, call or text 988. Outside the US, your national health service will have an equivalent. There is no version of this story where asking for help was the wrong move.

↑ Back to the animation

Connections

• All Interactive Visualizations
• Addiction — a learning disorder in a healthy brain
• Alcohol Use Disorder — releasing the VTA brake
• Gambling Disorder — the variable-ratio schedule
• Internet Gaming Disorder — engineered uncertainty
• ADHD — dopamine, effort and the speed of the rise
• Depression — anhedonia and the flattened reward signal
• Parkinson’s Disease — when the dopamine neurons die
• Restless Legs Syndrome — iron, dopamine and the TH bottleneck
• Tyrosine — the precursor (and why loading it does nothing)
• Phenylalanine — one step further upstream
• GABA — the inhibitory brake on the VTA
• Vitamin B6 — the cofactor AADC cannot work without
• Folate (B9) — BH₄ recycling and the methylation cycle
• Iron — sits in the active site of tyrosine hydroxylase
• How a Nerve Signal Travels — the spike that triggers release
• The Methylation Cycle — where COMT gets its methyl group
• Sleep & Circadian Rhythm — protecting your D2 receptors
• Stress & the Cortisol Axis — why relapse clusters around stress