The pathway — what the neuroscience actually shows, and where it stops

The shared circuit

Every substance or behaviour that has earned the addiction label engages the same core neural architecture: the mesocorticolimbic dopamine system. Dopaminergic cell bodies sit in the ventral tegmental area of the midbrain and project upward to the nucleus accumbens, the amygdala, and the medial prefrontal cortex. When something biologically salient happens — food when you are hungry, sex when you are aroused, a familiar drug cue when you have used before — these neurons fire and release dopamine downstream. The circuit is ancient. It is also general-purpose. It is the same circuit that fires when a teenager opens a notification.

Drugs of abuse — alcohol, opioids, stimulants, nicotine — all converge on this circuit by different molecular routes. Some bind to receptors that disinhibit dopamine release (opioids, via mu-receptors on inhibitory interneurons in the VTA). Some block reuptake (cocaine). Some directly stimulate release (amphetamines). Hyperpalatable foods engage the same pathway through gustatory and post-ingestive signalling. Sexual stimuli including pornography do so through ascending reward signalling from cortical processing of visual cues. The convergence is not metaphor; it is anatomy.

The fact that the circuit is shared does not mean every stimulus that engages it produces addiction. The circuit also fires when you eat a tomato, or kiss your spouse, or finish a piece of work you are proud of. What separates the addictive from the merely rewarding is what happens with repeated, intense, predictable activation — and how the brain reorganises around it.

Wanting versus liking — Berridge

The most important conceptual move in addiction neuroscience of the last thirty years is the dissociation between wanting and liking. Kent Berridge and Terry Robinson at Michigan made the case in the early 1990s and have been refining it ever since; the canonical 30-years-on retrospective is Robinson & Berridge in Annual Review of Psychology, 2025.

The dopamine pulse in the nucleus accumbens does not encode pleasure. It encodes incentive salience — the motivational pull toward a cue, the "this matters, go get it" signal. Pleasure itself — the actual hedonic hit — is mediated by smaller, more fragile opioid and endocannabinoid hotspots and is not dopamine-dependent. You can lesion the dopamine system in a rat and the rat still likes sugar (it makes the same facial expressions); it just stops wanting it.

This matters because it explains the central clinical fact about addiction that the willpower frame cannot. Repeated drug use sensitises the wanting circuit — it grows stronger, more reactive to cues — without sensitising liking. The drug user wants more and more, and enjoys it less and less. Long-term addicts routinely report that the substance stopped being fun years ago. They use because the wanting system has been pulled out of register with the liking system. This is the lived experience of every recovered alcoholic describing why they kept drinking after it had ceased to be pleasurable. It is not a moral failure to enjoy something less and pursue it more. It is the predictable output of a brain in which two normally co-operative systems have come apart.

Allostasis — Koob's dark side

The second load-bearing concept is the opponent-process / allostatic theory of addiction, developed by George Koob and Michel Le Moal across the late 1990s and 2000s. Foundational papers: Koob & Le Moal, Neuropsychopharmacology, 2001; Koob & Le Moal, Philosophical Transactions of the Royal Society B, 2008.

Early drug use is positively reinforced — the high feels good, the user wants more. Chronic use, by definition repeated and intense, recruits the brain's anti-reward systems — corticotropin-releasing factor, dynorphin, noradrenaline in the extended amygdala — which overcompensate for the recurrent dopamine surges. The brain re-baselines. The hedonic set-point resets to a new, lower level. This is allostasis — the maintenance of stability through change. The user is no longer dosing to feel good; they are dosing to climb back to neutral.

The clinical implication is precise. Late-stage addiction is dominated by negative reinforcement — using to escape the dysphoria of being sober. Not pleasure-seeking, pain-avoidance. The substance becomes the only thing that makes the new baseline tolerable. Anyone who has worked with a chronic alcoholic in withdrawal has watched this in person: the drink is no longer wanted because it is enjoyable, it is wanted because its absence is unbearable. This is a measurable neuroendocrine state, not a moral one.

D2 downregulation — Volkow's image

The third foundational finding is empirical, and it comes from positron emission tomography. Nora Volkow's NIDA group, beginning in the early 1990s, scanned the brains of cocaine, methamphetamine, alcohol and heroin users. All four populations showed reduced striatal dopamine D2 receptor availability compared to controls — and the D2 reduction correlated with reduced metabolism in the orbitofrontal cortex and anterior cingulate, the executive regions that would otherwise apply the brake.

Key papers: Volkow et al., cocaine users, 1993; Volkow et al., methamphetamine users, 2001; the synthesis is Goldstein & Volkow, Nature Reviews Neuroscience, 2011, which proposed the iRISA model: impaired Response Inhibition and Salience Attribution. The addicted brain over-values the drug cue and under-recruits the prefrontal stop signal. Both halves of the regulatory circuit are demonstrably blunted, on imaging, in living human addicts. The brake fails as the accelerator sticks.

Set-point — the obesity case

Obesity sits awkwardly in the addiction taxonomy because food is not a drug, exposure is universal, and abstinence is impossible. But the underlying biology of weight defence is as robust as anything in the addiction literature, and it is what makes the "just eat less" instruction so often fail.

The canonical paper is Leibel, Rosenbaum & Hirsch, NEJM, 1995: Changes in Energy Expenditure Resulting from Altered Body Weight. Rockefeller metabolic ward, controlled isocaloric feeding, lean and obese subjects, weight maintained at 10 to 20 per cent below their habitual mass. The result, replicated since: total energy expenditure dropped by 300 to 400 kilocalories per day below what mass loss alone predicted. The reduced body actively burned less than its new size predicted, while hunger-signalling hormones rose. The defence persisted at follow-up years later (Rosenbaum & Leibel, 2016).

This is not metaphor and not character. It is a homeostatic system defending its prior set-point, and it does so indefinitely. The "fat patient" the lab tech mentions on the previous page is not failing to muster the same self-discipline her grandmother had. She is being asked to override a hypothalamic defended state, every day, for the rest of her life. A small fraction of patients can do this. The published five-year sustained loss rates of behavioural-only obesity interventions are in single digits. The drug works precisely because it acts on the same system that is doing the defending — it shifts the defended set-point downward, rather than asking the patient to white-knuckle their way around it. The drug is not cheating biology. It is meeting biology where biology is.

Alcohol — the cleanest fit

Of the four conditions on this site, alcohol is the closest fit to the classical addiction model. DSM-5 collapsed the old abuse/dependence split into a single Alcohol Use Disorder, diagnosed on a severity spectrum: mild (2–3 symptoms), moderate (4–5), severe (6 or more). US 12-month prevalence is approximately 13.9 per cent of adults, with lifetime prevalence near 29.1 per cent (Grant et al., NESARC-III, JAMA Psychiatry, 2015).

The reason alcohol fits the model cleanly is that it engages essentially every component: dopaminergic reward via VTA disinhibition, GABA-mediated anxiolysis, glutamate downregulation with rebound on withdrawal, opioid release, opponent-process allostasis on a textbook timescale, and a well-documented physical withdrawal syndrome that includes seizure and death risk. The clinical picture is unambiguous in severe AUD.

The political awkwardness is also unambiguous. Alcohol is legal, normalised, embedded in social ritual. The overwhelming majority of drinkers are not addicted. The disease label is hard to apply consistently when the substance is on sale at the petrol station. This is the same tension that runs through the food and pornography sections below: the substance is ordinary, the addicted subset is small, but the biology of the addicted subset is real. The willpower frame collapses the distinction; the disease-for-everyone frame inflates it. Neither holds together.

Opioids — dependence is not addiction

The most precisely understood mechanism of any addictive class, and the most commonly muddled in public discussion. Opioids are full or partial agonists at the mu opioid receptor, hyperpolarising GABAergic interneurons in the VTA and disinhibiting dopamine release in the nucleus accumbens (StatPearls Mu Receptors review). The acute effect is analgesia, sedation, euphoria. The chronic effect is receptor desensitisation, beta-arrestin recruitment, cAMP up-regulation, and the predictable physical-dependence picture: tolerance, withdrawal, dose escalation.

The critical distinction that journalism routinely fails: physical dependence is not addiction. A cancer patient on chronic morphine is physically dependent; they are not addicted. A post-surgical patient on a two-week opioid taper develops mild dependence; they almost never become addicted. Dependence is predictable, dose-related neuroadaptation that resolves with taper. Addiction is the compulsive behavioural-motivational construct — using despite harm, prioritising the drug, loss of control. The two often co-occur. They are not the same thing, and the failure to separate them has driven a generation of bad pain-management policy.

The OxyContin / Purdue history is the cautionary case. Purdue Pharma marketed OxyContin from 1996 with the demonstrably false claim that its slow-release formulation prevented addiction. Aggressive GP-targeted sales, fabricated pain-relief data, and downplayed dependence are widely credited as the inception of the US opioid crisis. Purdue dissolved in 2021 after criminal pleas. The same regulatory architecture is now in front of GLP-1 promotion — a fact the lab tech is right to flag, even if her conclusion does not follow from it.

Food — the contested case

The most actively contested label in the field. Ashley Gearhardt's group at Michigan developed the Yale Food Addiction Scale in 2009 and a DSM-5-aligned 2.0 version in 2016, applying the substance-use-disorder criteria to hyperpalatable foods — engineered high-fat, high-refined-carbohydrate, salt-enhanced combinations rarely found in nature. David Kessler, former FDA commissioner, coined conditioned hypereating in The End of Overeating (2009). Michael Moss's Salt Sugar Fat (2013) documents food-industry executives engineering "bliss point" formulations to maximise consumption.

The Gearhardt push, restated in BMJ 2023, is for institutional recognition of ultra-processed food addiction (UPFA) as a substance-based addictive disorder. Recognition has not arrived. Food addiction is not a DSM-5 diagnosis. Critics — and they are serious critics, not industry shills — point out three things: food is essential and abstinence is impossible (which breaks the standard model); the YFAS criteria overlap heavily with binge-eating disorder; and prevalence estimates around 14 per cent of adults are inflated by that overlap.

The honest position is that the underlying mechanism — hyperpalatable foods engaging the same reward circuit as drugs of abuse, in some individuals producing compulsive use patterns that pass the addiction screening — is real and replicable. The clinical category is not yet earned at the institutional level. The set-point story above is more important than the addiction story for the policy question on Mounjaro: even if "food addiction" turns out to be wrong as a clinical label, the metabolic defence of body weight is the load-bearing biology, and it is not contested.

Pornography — same circuit, unearned label

The most contested label in the entire field. There is no DSM-5 diagnosis for pornography or sex addiction. The hypersexual disorder proposal was rejected in 2013 on grounds of insufficient empirical support, weak inter-rater reliability, and the foreseeable forensic misuse risk (Reid & Kafka, 2014).

The ICD-11 (WHO, 2018) added Compulsive Sexual Behaviour Disorder — but explicitly classified it among impulse-control disorders, not addictive disorders. The WHO working group's stated position was conservative wait-and-see: the evidence base is not yet equivalent to gambling or substance disorders (Kraus et al., World Psychiatry, 2018).

The neural evidence is suggestive. Valerie Voon's Cambridge group scanned compulsive sexual behaviour patients in PLoS One, 2014 and found elevated activity in ventral striatum, dorsal anterior cingulate and amygdala while viewing pornography — the same circuit lit by drug cues in addicted users. Crucially, compulsive users reported higher wanting but no higher liking. A textbook Berridge incentive-sensitization profile. Voon herself cautioned that this does not prove pornography is intrinsically addictive.

The sceptics — Nicole Prause, David Ley — have a peer-reviewed standing. They argue the construct is unfalsifiable, the neural findings inconsistent with addiction models, and the diagnosis disproportionately weaponised against gay men and religious users. Their work is itself contested. The honest summary for a site like this: the brain mechanism is plausible and partially documented; the clinical diagnosis is not yet earned at institutional level; do not assert porn addiction as established, but do not deny that compulsive use in some individuals produces real measurable distress on the same circuit.

Four frames — Volkow, Lewis, Szalavitz, Hart

Four working scientists, four readings of the same evidence base. Each is taken seriously inside the field. The disagreement is not "real scientists versus cranks". It is "what does the data mean".

Volkow — the brain-disease orthodoxy

Volkow, Koob & McLellan, NEJM, 2016: addiction is "a chronic, relapsing brain disease" comparable to diabetes or hypertension, with biological, behavioural and environmental causes. The Director of NIDA. Controls the largest addiction-research budget in the world. The disease frame's strongest institutional advocate. Volkow's defence: conceptualising addiction as a brain disease "challenged views that addiction occurs as a result of moral failing or lack of willpower" and unlocked insurance parity for treatment. The frame's pragmatic case is stigma reduction and access to medical care.

Lewis — addiction as deep learning, not disease

Cognitive neuroscientist, former heroin user, author of The Biology of Desire: Why Addiction Is Not a Disease (2015). Lewis accepts the brain-change findings — D2 downregulation, accumbens sensitisation, prefrontal hypometabolism — and re-interprets them. These are neuroplasticity doing what neuroplasticity always does: deepening grooves around repeated, salient, emotionally rewarding behaviour. The brain that learned the addiction can learn its way out. The disease label, in his view, produces learned helplessness in the people it labels and is empirically loose. Lewis, Neuroethics, 2017 is the academic statement.

Szalavitz — learning disorder with developmental and social roots

Recovery-experienced harm-reduction journalist, author of Unbroken Brain (2016). Szalavitz unifies the learning frame with the developmental-neuroscience evidence: addiction emerges from the interaction of temperament (often high sensitivity, anxiety, trauma exposure) with the timing of first heavy use during the still-developing adolescent brain. Closer to ADHD or autism than to diabetes. Harm reduction first, abstinence-only counterproductive, the carceral apparatus actively prevents the learning that ends addiction.

Hart — addiction is the exception, not the rule

Columbia neuroscientist, author of Drug Use for Grown-Ups (2021). Hart's position is the sharpest. Most users of even the hardest drugs — including heroin — are not addicted. The destructive cases are driven by poverty, trauma and prohibition's own harms, not by pharmacology. The addiction frame pathologises ordinary adult pleasure-seeking and lands disproportionately on Black communities. Hart is himself an open responsible heroin user, a tenured Ivy professor making the autobiographical point that drug use and a flourishing life are compatible. His policy line is full legalisation.

The peer-reviewed in-house critique

The disease model has been challenged from inside the field on its own terms. Hall, Carter & Forlini, Lancet Psychiatry, 2015 — The brain disease model of addiction: is it supported by the evidence and has it delivered on its promises? — concluded that the BDMA is not supported by animal and neuroimaging evidence to the extent its advocates suggest, has not delivered more effective treatments, and has had only modest effects on policy. This is a load-bearing internal critique.

The mental toll mechanism

Five reinforcing layers, each documented, that constitute the lived experience of addiction.

Anhedonia. D2 downregulation and ventral-striatal hyposensitivity mean non-drug rewards — food, sex, money, friendship — generate progressively weaker hedonic signal. The world greys out around the one bright pixel of the drug. Reported clinically as flat affect, loss of interest in previously enjoyed activities, emotional blunting. Pre-dates the drug career sometimes; deepens with it.

Executive function decline. Reduced orbitofrontal and anterior cingulate metabolism translates clinically into impaired decision-making, poor impulse control, blunted future-orientation. Measurable on PET; measurable on standardised neuropsychological testing; measurable in everyday function. The brake fails as the accelerator sticks.

Allostatic load. Chronic recruitment of corticotropin-releasing factor, dynorphin and noradrenergic anti-reward systems produces a persistent dysphoric baseline — anxiety, irritability, sleep disruption, elevated stress reactivity. This itself drives further use as the user doses to escape the very state the previous dose produced.

Shame and identity spiral. Repeated failure to control use triggers self-attribution: I am the problem, not the behaviour. Shame drives concealment, concealment drives isolation, isolation removes the social regulation that would otherwise interrupt use. The willpower frame on this site's previous page is, ironically, the single largest accelerant of the shame spiral, because it tells the addict that the problem is a defect in their character.

Social isolation. Withdrawal from family, work, friends — pushed by avoidance of judgement, pulled by use crowding out relationships — removes the strongest natural protective factor against relapse. Loneliness amplifies craving in laboratory cue-reactivity studies. The shame spiral and the isolation spiral feed each other.

These five layers are mutually reinforcing. Anhedonia drives use; use deepens executive impairment; impaired control produces shame; shame produces isolation; isolation amplifies anhedonia. The loop is the toll. It is also the reason interventions that address only one layer — only the drug, only the cognition, only the social environment — under-perform. The strongest treatment apparatuses address several at once.

The next page asks the cost question. →