Humans

The dominant prediction about artificial intelligence and economic value runs roughly as follows. AI makes software abundant. Things that become abundant lose value, and value moves to whatever remains scarce. Software is information, and information is becoming nearly free to produce, so the scarce thing must be physical: the chips, the devices, the energy, the hardware that no model can generate on its own. The conclusion is that the next era belongs to hardware, and a great deal of capital has followed the conclusion.

The prediction is half correct, which is what makes it convincing. The half that holds is that computation is genuinely scarce. Processors, fabrication capacity, and electrical power are real constraints, and the firms that own them have real businesses. But this is the scarcity of an input, not of an end. Computation is scarce in the way the toll on a road is scarce: charging for it is profitable precisely because the road is busy, but no one travels in order to pay the toll. The prediction identifies the cost of the abundance and mistakes it for the destination.

The question it skips is what additional hardware the abundance is supposed to make people want, and the history is not encouraging for the hardware case. When companies have tried to deliver AI as a new physical object, a pendant, a wearable assistant, a device meant to displace the phone, the objects have generally failed, because software absorbed their function before they reached customers. This is the ordinary direction of the relationship between the two. Software absorbs hardware, and the reverse rarely happens. A single phone has already absorbed the camera, the map, the calculator, the newspaper, and the record shop. AI does not reverse this tendency; it intensifies it.

There is one kind of physical demand that has in fact been rising: the deliberately analog. Sales of paper notebooks, film cameras, vinyl records, and intentionally limited phones have grown even as their technical inferiority has grown with them. But this is not demand for hardware. No one buys a record player because it reproduces sound more accurately than a streaming service; the record player is bought as a way of withdrawing from the screen, and the withdrawal is the point. What the buyer wants is not the object but the thing the object protects, a quality of attention, usually shared with another person, that the software had been consuming. The demand that looks like demand for hardware is demand for something the hardware merely shelters, and that something is not physical.

The Shared Resource

The thing being sheltered is attention, and naming it requires immediate care, because the scarcity of attention is also the founding premise of the industry that did the damage. In 1971, the economist Herbert Simon noted that a wealth of information produces a poverty of attention. The attention economy adopted this as an operating principle and organized itself around capturing the scarce resource as efficiently as possible. The premise was not wrong. The error was in the verb. Attention was treated as something to be captured, when the property that makes it valuable is that it can be shared.

The word carries two distinct meanings that the industry collapsed into one. In the first, attention is something a system extracts: it captures attention, holds it, directs it toward whatever is being sold. In the second, attention is something one person extends to another, and when two people extend it to each other at the same time, the result is not extraction but a shared state. Ordinary language keeps the two apart. Machines capture and hold attention; people pay and give it. The attention economy was built entirely on the first meaning and was constitutionally unable to produce the second.

The second meaning has a precise name in developmental psychology. At around nine months, an infant begins to point at an object, look to confirm that the adult is also looking at it, and register that the adult has seen it too. Psychologists call this joint attention and regard it as the first specifically human act, arriving before language and before reason. Its structure is simple: two people attend to the same thing, and each is aware that the other is attending. Almost everything built on top of it, language, cooperation, trust, culture, depends on that mutual awareness. Shared attention is not an ornament on human life, everything runs on.

The Surplus

Shared attention also obeys different economics from the captured kind, and this is the difference on which the rest of the argument depends. Captured attention is rivalrous: the minutes a feed takes from a person are minutes that person no longer has, and the supply is fixed and diminished by use. Shared attention is not. Two people fully attending to each other are not dividing a fixed quantity between them; the attention each gives is not subtracted from the other but added to it. The exchange is self-reinforcing rather than depleting, since attention offered is generally returned, and the return deepens the attention that prompted it.

It also produces something that neither party held going in. Two people talking through a problem arrive at a solution that was in neither of their heads beforehand; two friends recounting an evening produce a shared account of it that belongs to both and did not exist until they made it. This surplus — the idea, the understanding, the trust, the shared reference that was not there before the attention was shared — is the actual product, and no single mind can manufacture it alone. Describing attention as a resource to be spent, mined, or harvested therefore misses what matters most about it. Those verbs apply only to the captured form. The shared form increases in the act of being shared.

Enclosure

This identifies what the attention economy actually did, which was not theft but enclosure. It took a thing that grew when shared and re-engineered it into a thing that could be extracted and sold, diverting attention that would otherwise have passed between people through systems that could only draw it out of them.

The damage followed from the design rather than from any malice in it. A dating service measured by engagement is most successful when its users keep returning and least successful when they pair off and stop, which means its commercial interest points toward users who never quite succeed. Pornography optimized for repeat use shifts the baseline against which ordinary intimacy is judged, until the real version registers as a disappointment. Social feeds reproduce the form of a shared public while delivering each user, separately, to a private screen. None of these outcomes was written into the technology; there is nothing in a network that prefers people to be alone. They were written into a single optimization target. Engagement is maximized by retaining the user, and a person who is connected to others and out in the world is a person who has closed the application. Therefore, isolation was contained in the objective they were given.

Counterfeit Attention

The natural objection is that AI will now restore the connection it helped remove, by supplying synthetic companions without limit. It will supply something, but there is one operation it cannot perform however capable it becomes. It cannot share attention, because sharing requires two parties each aware of the other, and on the model's side there is no party. A system can aim unlimited and finely tuned attention at a person. It cannot be jointly attentive with that person, because joint attention requires a second center of awareness, and the system does not contain one. What it offers has the one-directional structure of capture, turned inward toward the user rather than outward toward an advertiser, but capture nonetheless.

The probable effect of synthetic intimacy is the effect that processed food had on diet. It satisfies the immediate craving while leaving the underlying need less met than before, and for anyone who can still detect the difference, it raises rather than lowers the demand for the genuine version. Considered this way, the companion products are not competitors to human connection.

The Antidote

The scarcity of shared attention was not caused by a shortage of the underlying capacity; the capacity was diverted into systems that cannot return it. This is the part of the situation that is genuinely favorable, because a diversion can be reversed where a true shortage cannot. The human capacity for joint attention has not been removed. It has been starved of occasions, and occasions can be rebuilt.

To put it in economic terms, the one thing that's valuable and that cannot be produced cheaply by AI alone is not hardware, but the thing the previous era spent two decades dismantling: people giving one another attention while present in the same situation. It's the largest unmet need of the period, and whatever addresses it credibly will hold its value precisely because the surrounding software has lost to satisfy it, and AI might.

Software built for this would run on the inverse of the earlier objective. Rather than capturing attention, it would arrange the conditions under which attention can be shared, placing two people in contact over the same thing, and then withdraw. Its measure of success would be time spent away from it (or with it, as a group) rather than time alone for all parties; it would be functioning correctly at the moment the user turned to someone present. The reason this is not a marginal business is that it describes the one market AI cannot commoditize alone, and the one product the attention economy was structurally incapable of building, since a company paid by engagement cannot afford a product that succeeds by being left.

The Next Enlightenment

There is a precedent for the shape of this transition, though not for its scale. The Enlightenment was itself a response to a sudden abundance of information, produced by the printing press, and its central achievement was to equip the individual to reason without depending on inherited authority. It succeeded almost completely, and it reorganized the world around the liberated individual.

Carried forward for three centuries, that achievement arrives at the present arrangement: a population that is the best informed in history and, by many measures, among the most isolated. The individual was freed so thoroughly that the connections between individuals were left with nothing to maintain them. The liberation of the single mind, taken to its limit, produces people who are each free and largely unreached.

The correction now required cannot be a further liberation of the individual mind, because the individual mind is about to have unlimited capability available to it on demand. It is no longer the capacity of the person but the connection between persons, the joint attention that appears at nine months and was subsequently engineered out of ordinary life. The earlier task was to make it possible for people to think without authority. The remaining task is to make it possible for them to attend to one another again.

This is less dramatic than either the hardware thesis or the fear of being replaced by machine companions, and it is also more tractable than either. It does not call for a retreat from software toward physical objects. It calls for software built, for once, around the people using it rather than around their attention. The refuge was never the hardware. It was other people, and the work is to stop building the things that stand between them.