A Common Currency

Valuation, Preference, and the Orbitofrontal and Ventromedial Cortex

The problem of unlike things

The last unit handed us a precise and unfinished question. The ventral striatum, we said, learns the value of what just happened and spends that value to bias which action the basal ganglia will release. But it does this without ever representing value as such — without holding an option up, apart from any action, and asking what it is worth. Our worked example made the limitation vivid: a shift in drive flipped a contest between eating and throwing, but nothing in that story weighed two genuine alternatives and judged between them. The drive changed; the architecture resolved the new contest; no comparison was made.

Now ask the animal to do something harder. It is not hungry in a kitchen with one sandwich. It is in a world that offers, at this moment, several incompatible goods: there is food some distance off, water in another direction, a potential mate across the clearing, and unexplored ground beyond. It cannot pursue them all. To choose, it must compare them — and here is the difficulty that organizes this entire section. These things are not alike. Food and water and a mate and a patch of unknown territory are incommensurable; they answer different needs, arrive on different timescales, and have nothing obvious in common that would let one be measured against another. You cannot subtract a mate from a meal. And yet the animal must somehow decide, which means it must somehow place these unlike goods on a single scale where they can be compared — where one can come out ahead.

That scale is what we will call a common currency: a shared dimension of worth onto which qualitatively different options are converted, so that a controller can do the one thing it must do to choose well — rank them. The claim of this section is that the brain computes such a currency, that the orbitofrontal and ventromedial prefrontal cortex are central to computing it, and that this is precisely the representational capacity the striatum lacked. The striatum told us a reward was coming and taught us how to get it. This cortex tells us what an option is worth, on a scale that lets it be set against another.

It is worth seeing at the outset why a common currency is not optional. A controller facing incommensurable options has only two possibilities. Either it converts them to a shared scale and compares, or it does not — in which case it needs some other rule for choosing, and every such rule turns out to smuggle in a chooser. “Do whatever you most prefer” only relocates the problem: preference is the ranking we are trying to explain. “Follow a fixed priority — always eat before drinking” fails the moment the animal is parched and full. The honest alternatives to a common currency all require an agent who already knows what it wants — which is to say, they require the very homunculus this book has spent five units declining to install. A common currency is what choice among unlike things looks like without that agent: not a judge who weighs, but a cortex that converts everything to one axis, on which the largest simply wins. We are, once again, looking for the mechanism that does the work a little inner person would otherwise have to do.

Value in single neurons: the orbitofrontal cortex

The most direct evidence that the brain computes something like this comes from listening to single neurons in the orbitofrontal cortex while an animal makes exactly the kind of comparison we have described.

The decisive experiments are Camillo Padoa-Schioppa and John Assad’s, working with monkeys offered choices between two flavored drinks. Before recording, the experimenters did something essential: they measured how much the monkey valued each drink relative to the other, by finding the trade-off at which the animal was indifferent. A monkey might prefer grape juice to apple juice so strongly that it would give up several drops of apple to get one drop of grape — and by varying the amounts on offer, one can locate the point of indifference, the exchange rate at which the two options are subjectively equal. (Suppose it falls around four to one: four drops of the less-preferred drink are worth one drop of the preferred.) This matters because it lets the experimenter separate the amount of a drink from its value to the animal, which the indifference point has now quantified. Then they recorded, and asked what the orbitofrontal neurons were tracking.

What they found is the heart of the matter. Some neurons fired in proportion to the value of an offered good — not to its identity, not to the action the monkey would take to get it, but to how much that quantity of that drink was worth to the animal, measured in the common currency the indifference point had revealed. A neuron coding the value of the offered grape juice responded similarly whether the value came from a little of the strongly preferred drink or more of a weakly preferred one, as if it had already done the conversion. Other neurons tracked the value of the option the monkey was actually choosing, and others the difference between the two offers. The population, in other words, looked very much as though it were representing offers on a scale of worth — exactly the representation a controller would need in order to compare unlike things.

Two features of this finding deserve emphasis, because they are what make it more than another reward signal. First, the code appears to be about the good, not the action. The orbitofrontal neurons tracked what an option was worth before, and somewhat independently of, the movement that would secure it — a value attached to the thing rather than to the reaching. This is a different kind of signal from anything in the last unit, where value was always expressed as a bias on action selection. Here value is represented in its own right, abstracted from what you would do about it. (We will see that this “good-based” reading is itself contested — whether value is ever truly divorced from a prospective action is a live question — but the basic result, that orbitofrontal firing grades with subjective worth, is solid.) Second, the scale is relative. The neurons were not reporting some absolute, context-free quantity of pleasantness; they were reporting worth as established by the animal’s own preferences, the same worth the indifference procedure had measured behaviorally. The cortex and the behavior were reading off the same currency.

The same currency for money, faces, and wine

If the orbitofrontal and ventromedial cortex really compute a common currency, then a strong prediction follows, and it is testable in humans: the same valuation signal should appear across goods that have nothing physically in common. A scale that can compare unlike things must, by definition, be indifferent to what the things are. Money and food and a smiling face are about as unlike as rewards get; if one neural signal tracks the worth of all of them, that is the currency showing itself.

This is close to what the human imaging literature reports. Brian Knutson’s monetary incentive task — built deliberately as a human echo of Schultz’s monkey work — separates the moment a reward is anticipated from the moment it is delivered, and the two moments emphasize different nodes: the ventral striatum responds vigorously when a reward is signaled and in prospect, while the ventromedial prefrontal cortex is especially engaged as the outcome itself is registered and valued, scaling with how much is at stake. Crucially, the same circuitry is recruited when the reward is not money at all. A version of the task in which the “reward” is the chance to see an attractive face drives the ventral striatum in anticipation much as money does; another in which participants watch a charity receive money, or choose to give to it themselves, engages the same valuation regions. Monetary, social, and even vicarious or charitable rewards converge on overlapping machinery — which is just what “common currency” predicts: the substance of the reward washes out, and what remains is its worth.

The most systematic statement of this comes from a large meta-analysis by Bartra and colleagues, who pooled more than two hundred imaging studies of subjective value. A consistent system emerged whose activity scaled linearly with how much a person valued an outcome — and that system centered on the ventral striatum and the ventromedial prefrontal cortex, the very pair the last unit and this one keep returning to. Several details of the finding matter for our argument. The system responded to both primary rewards (food, taste) and secondary ones (money), as a genuine common currency must. It responded both when a reward was delivered and when a cue merely predicted one, tying anticipation and outcome to the same scale. And — a subtle but telling point — it tracked positive subjective value specifically, rising with how good an outcome was, rather than responding to extreme values of either sign. A signal that fired equally for very good and very bad outcomes would not be coding value at all; it would be coding intensity, or salience, or arousal. That this system distinguishes the two is part of what licenses calling it a value signal in the first place.

Underneath all of this is a particular idea about what value is, and it is worth naming because it frames everything in the section. The worth being represented here is not a property of the good itself. There is no quantity of “value” residing in a drop of grape juice or a dollar, waiting to be read off. Value is subjective — it is the importance the option holds for this animal, given its current state and history, for the achievement of its ends. The same sandwich was worth crossing the room for before dinner and not after; the drink was worth four-to-one to this monkey and might be worth two-to-one to another. The common currency is a currency of subjective worth, computed relative to the animal, which is exactly why it has to be computed at all rather than simply sensed. A homunculus could be handed the values; a brain has to construct them.

The currency, gamed

There is a backhanded way to confirm that a signal really is the brain’s measure of worth: show that you can manipulate it by changing what something means to a person, while changing nothing about the thing itself. If the valuation signal can be driven by mere belief about an option’s worth — by its price, its label, its brand — then it is tracking subjective value in the fullest sense, value as constructed rather than value as sensed. The literature that does this most pointedly grew up, fittingly, around marketing.

Consider the brand experiment that started much of this interest. People with a strong preference for one cola over another were given the drinks while their brains were imaged, and the ventromedial response tracked not just the taste but the brand knowledge: the same cola produced a stronger valuation signal in someone who knew and preferred that brand than in someone who did not. The flavor on the tongue was identical; what differed was what the brand meant to the drinker, and the valuation signal followed the meaning. Antonio Rangel and colleagues made the point sharper still with wine. They gave people the same wine under different price tags and asked them how much they enjoyed it. People reported the wine as more pleasant when it was labeled expensive than when the very same wine was labeled cheap — and the orbitofrontal and ventromedial signal tracked the reported pleasure, rising with the price on the label rather than with anything in the glass. The cortex was not reporting the chemistry of the wine. It was reporting what the wine was worth to a person who believed it was costly.

It is tempting to read these studies cynically, as exposés of how easily we are fooled, and that reading has fueled a small industry of “neuromarketing.” But the deeper lesson is the one this section has been building toward. The valuation signal is supposed to be manipulable by meaning, because the quantity it represents is subjective worth, and subjective worth genuinely is partly constituted by belief and context. The expensive label does not merely change a report; on this evidence it changes the experience — the wine is, to that person in that moment, more pleasant. A signal that ignored price and reported only the chemistry would be a worse measure of value, not a better one, because value was never in the chemistry. That the currency can be gamed is not a flaw in it. It is what it looks like for a currency to be genuinely subjective.

Worth is not fixed: updating what a thing is worth now

One more property of this system needs introducing here, because it distinguishes the ventral frontal cortex from a simple lookup table and because later sections will lean on it. The worth of a thing is not stamped on it permanently. It depends on the animal’s state — a satiating food loses its value as the animal fills, exactly as the last unit’s flip required — and it can change when the world changes the rules. A cue that has predicted reward may abruptly stop; a good that was valuable may become worthless or even aversive. A valuation system worth having must update, tracking not what an option was worth in general but what it is worth now.

This updating is something the orbitofrontal cortex in particular seems built to do, and it is part of why damage there is so disabling in ways a loss of “intelligence” would not predict. An animal whose orbitofrontal cortex is compromised can continue to treat an option as valuable after it has ceased to be — failing to adjust when the contingencies reverse, persisting with a choice whose worth has collapsed. We will meet the human face of this failure in the syndromes ahead: people who can state what they ought to value yet keep choosing as though the old worth still held. For now the point is structural. The common currency is not a fixed table of prices. It is a running estimate of subjective worth, continuously revised against the animal’s changing state and the world’s changing contingencies — which is exactly the kind of estimate an anticipatory controller, trying to act on a future that keeps moving, would need.

Where this leaves us, and what it sets up

Step back and see what this cortex has added to the machine of the last unit. The striatum could learn that a reward was coming and could bias action toward getting it, but it could not hold two unlike goods apart from action and ask which was worth more. The orbitofrontal and ventromedial cortex supplies exactly that: a representation of subjective worth on a common scale — visible in single neurons that grade with relative value, in a human valuation system that tracks worth across money and food and faces and wine, and in the manipulability of that system by meaning, which is the signature of a currency that is genuinely subjective. With this representation in hand, an animal can do what the striatum alone could not: compare incommensurable options and rank them. And it can do so without an inner judge, because the ranking falls out of the conversion — put everything on one axis and the largest wins. The currency is the chooser.

But notice what this section has and has not done. We have shown how the brain represents what an option is worth. We have not shown how a merely possible option — a future that has not arrived, a consequence that has not happened — comes to carry such a value in the first place. The wine was in the glass; the juice was on offer; the money was on the screen. In each case the good was present, or its arrival was imminent and cued. The harder problem, the one the whole unit is circling, is how worth attaches to things that are not present: to outcomes the animal must imagine, to futures it must simulate before it can value them. Representing the worth of an option in front of you is one thing. Feeling the worth of a possibility — letting an imagined consequence pull or repel you before it is real — is another, and it will turn out to draw on the same ventral cortex’s oldest inheritance, its origins in the body and the emotions. That is the next problem: not what an option is worth, but how a future gets to be felt.

There is also a second thread this section has deliberately left dangling. We have spoken throughout as though the animal already has the options to compare — as though food and water and a mate were simply given, waiting to be ranked. But the options had to come from somewhere. A controller can only value the futures it has managed to represent, and it can only represent the ones it has gathered the materials to build. Where those materials come from — the wandering that stocks the animal’s map of what the world affords, against needs it is not yet feeling — is a thread we will pick up when we turn to exploration. For now, we have a currency. We turn next to how the options priced in that currency come to be felt before they are real.

What we are sure of, and what is still open

As in the last unit, it is worth separating the settled core from the frontier.

What is well established. The orbitofrontal and ventromedial prefrontal cortex carry signals that scale with the subjective value of options, in monkeys (single neurons grading with relative, preference-defined worth, as in Padoa-Schioppa and Assad) and in humans (a valuation system, centered on ventromedial prefrontal cortex and ventral striatum, whose activity tracks subjective value across qualitatively different rewards, as in Bartra and colleagues’ meta-analysis). This valuation signal spans primary and secondary rewards and applies both to delivered rewards and to cues predicting them. It tracks subjective worth, manipulable by belief and context — price, label, brand — rather than the physical properties of the good, which is the expected signature of a value signal rather than a defect in it. And the worth represented is state-dependent and updatable, not a fixed property of the good; the orbitofrontal cortex in particular is important for revising value when contingencies change.

What remains contested or unsettled. Whether value in the orbitofrontal cortex is truly represented in a “good-based,” action-independent form — abstracted from the movement that would secure it — or whether value is always, at bottom, tied to a prospective action, remains debated; Padoa-Schioppa’s abstract reading is influential but not uncontested. The “common currency” itself should be held with some care. It is genuinely uncertain whether there is a single scale of value or several partially separate ones, computed in distinct but spatially intermingled circuits that coarse imaging cannot resolve — the same resolution problem we met with the direct and indirect pathways of the striatum, and one that applies with full force here, since functionally distinct microcircuits in this cortex may be interdigitated below the scale fMRI can see. There is, too, a standing methodological worry: because the ventromedial cortex responds to so many pleasant and valued things, inferring that it “computes value” from its activation risks circularity, and distinguishing genuine value coding from correlated quantities like salience and arousal is a continuing problem rather than a solved one. The identification of this cortex as a valuation system is strong and well-replicated; the precise form of the value it represents, and whether “common currency” is one scale or a convenient name for several, is an active frontier.