Wanting vs Liking

What it is

In the late 1980s, researchers at the University of Michigan destroyed 98 to 99 per cent of the dopamine in a group of rats' brains. The rats stopped eating. Food sat in front of them and they would have starved beside it. Terry Robinson and Kent Berridge then put a drop of sugar water on the animals' tongues. The rats produced the same rhythmic tongue protrusions rats always produce for sweetness. The pleasure reaction was intact. What had gone was whatever made them cross the cage to get it.

Robinson and Berridge named the pursuit half of this pair incentive salience, more commonly called wanting, and the pleasure half hedonic impact, or liking. Dopamine, on the model most people still carry, is the pleasure chemical, the thing behind a "hit". Robinson and Berridge's account, developed across their 1993 paper on incentive-sensitization and formalised as a general dissociation in 1998, says otherwise. Their thirty-year review, published in the Annual Review of Psychology in 2025, states it with little hedging: dopamine mediates "incentive motivation ('wanting')... but not their hedonic impact (liking)". Dopamine antagonists and near-total lesions leave liking reactions to sweetness unchanged, even as they gut the pursuit of it.

The two systems are not built to the same scale. Berridge's lab has mapped the sites where opioid stimulation genuinely amplifies pleasure reactions, the hedonic hotspots, and they run to roughly a cubic millimetre each in a rat brain, tucked inside the nucleus accumbens shell and the ventral pallidum. Move the injection a millimetre outside a hotspot and the animal eats more without liking it more. Wanting responds across most of the surrounding region. Liking responds only inside the hotspot. Berridge's own summary of the asymmetry, quoted in Adam Alter's Irresistible, is worth keeping in mind: wanting is "much more robust and big and broad and powerful. Liking is anatomically tiny and fragile."

The correction that most popular writing on this gets wrong is what happens to liking once wanting takes over. The version that circulates is that an addict grinds on joylessly, chasing something they stopped enjoying years ago. Robinson and Berridge reject this reading of their own theory. Their 2025 review reasserts that wanting "grows disproportionally due to sensitization", then specifies "whether liking either declines or stays the same". They note that a figure in their original 1993 paper implied a decline in liking and caused years of downstream confusion. King and colleagues, in a 2021 study cited in the 2025 review, followed people who went on to develop alcohol use disorder: wanting rose over time, while liking "remained high and stable". The problem cannot be found by asking whether someone still enjoys the thing. Usually, they do.

Sensitization, once established, is unusually durable. Robinson and Berridge describe it as "extremely long-lasting, persisting for months to years in both rats... and people, even after drug use was stopped", with cue-triggered hyperreactivity observed after decades. Tolerance can mask it for days or weeks, after which, in their words, "sensitization reemerges at full strength within a few weeks as tolerance dissipates". A pull can look like it is fading at precisely the point it is still fully present, which is a poor moment to conclude that anything has been resolved.

The honest limit of the theory matters as much as its findings. Robinson and Berridge extend incentive-sensitization beyond drugs, with visible hedging, to binge eating, compulsive gambling and compulsive sexual behaviour, citing neuroimaging evidence of cue-triggered hyperreactivity in some individuals in those groups. They never extend it to apps or social media. Applying the wanting and liking split to product engagement metrics, streak counts, daily active users, session length, is an argument by analogy, not a claim the authors make themselves, and any page or article that uses the framework this way should say so plainly rather than borrow the theory's authority without borrowing its caution.

In effect

The dopamine-depleted rats (Robinson and Berridge)

The clearest research case is the one the theory is built on. Rats with 98 to 99 per cent of their brain dopamine destroyed stopped eating entirely and would have starved next to a full bowl of food. When researchers placed sugar water directly on the animals' tongues, bypassing the need to seek it out, the rats produced the same rhythmic tongue protrusions that rats always produce in response to something sweet, the standard measure of hedonic reaction in this line of research. The pleasure response had not been touched. What had been destroyed was the system that turns a pleasure into a pursuit. The result is the reason the theory exists in a form specific enough to be tested rather than just argued for: it separates the two questions, does the animal like this, and does the animal want this, and shows an intervention that answers them differently.

Reopening a phone at one in the morning

Away from the laboratory, most people already have a private version of this. A feed gets refreshed, or an app reopened, well past the point where anyone watching would call it enjoyable, and often past the point where the person doing it would call it enjoyable either. The behaviour continues without a decision being renewed each time. Robinson and Berridge's framework gives this a name rather than a diagnosis: the wanting system can keep generating pursuit on its own, detached from whatever liking is or is not doing underneath it. This is exactly the gap King and colleagues documented in people developing alcohol use disorder, wanting climbing while liking held steady, and it is recognisable in miniature in anyone's own late-night scrolling, without requiring that the scrolling qualify as an addiction to make the point.

A product dashboard that only has half the instrument

Daily active users, session frequency, return rate, time in app, streak length: every one of these records how reliably something pulls a person back, which makes each of them a wanting metric. Satisfaction surveys and net promoter scores sit closest to liking, and in most product organisations they run quarterly, owned by a different team, treated as a supplement rather than the other half of the picture. A team can run a year of clean experiments, watch engagement climb every quarter, and generate no evidence at all about whether anyone is happier using the product. Ben Grosser's Demetricator, a browser extension that hides Facebook's like and follower counts, works this same logic from the opposite direction: remove the wanting signal from view and see what is left of the experience once it is gone. Neither the dashboard nor the extension proves Robinson and Berridge's neuroscience applies to software. What they illustrate is the asymmetry the neuroscience describes: pursuit is cheap to instrument and pleasure is not, so organisations that measure what is easy end up measuring wanting almost by default.

What it does not say

The dissociation is not a claim that pleasure is illusory, or that dopamine is irrelevant to reward, only that dopamine's role is pursuit rather than pleasure specifically. It is also not the joyless-addict story that circulates in popular accounts: Robinson and Berridge are explicit that liking can remain high and stable even as wanting climbs, and they say their own 1993 paper contributed to the confusion by implying a decline that later work did not support.

The theory's authors extend incentive-sensitization to food, gambling and compulsive sexual behaviour, but always with visible hedging, "some" individuals, cue-triggered reactivity that "may be" a marker of vulnerability, and they state plainly that the theory was developed for drug addiction rather than behavioural addiction in general. They never mention apps or social media. Nothing in Robinson and Berridge's published work has been tested on a product dashboard, a notification system or a feed. Using wanting and liking to explain engagement metrics, as this site does elsewhere, is an argument by analogy borrowed from a much better-evidenced neuroscience literature, not a finding that literature contains.

The authors also list limits inside their own theory. Dopamine is, in their words, "only one link in the causal chain", alongside glutamate, opioid, GABA, orexin and endocannabinoid systems. Other researchers, including George and Koob, argue sensitization matters mainly early in addiction rather than throughout its course, and choice theorists dispute whether addiction is compulsive in the way the theory implies; Robinson and Berridge concede it is not compulsive in the classic sense. The wanting and liking dissociation itself is about as well established as anything in reward neuroscience. Everything built on top of it, including its use on this site, should be held to a lower standard of certainty than the dissociation itself.

Who developed it

Terry Robinson is first author on both the 1993 incentive-sensitization paper and the 2025 thirty-year review; Kent Berridge, his University of Michigan colleague, led much of the hedonic hotspot mapping that established where liking is generated. The two names are usually cited together because the dissociation, and the addiction theory built on it, is joint work developed across three decades of shared publications.

---