The interest of the few against the safety of the many

The Hill reports that tech employees across Silicon Valley find themselves in open disagreement with industry executives, who are spending millions of dollars to resist meaningful AI regulation. The workers who build these systems — who understand, from the inside, what they are capable of — want stronger safeguards. The executives, who stand to profit most directly from rapid and unchecked deployment, want government to step aside. This is not a new drama. It is one of the oldest in political economy.

In the Wealth of Nations I observed that the proposal of any new law or regulation of commerce originating from merchants and manufacturers ought always to be listened to with great precaution, and never adopted until after long and careful examination — not only with the most scrupulous, but with the most suspicious attention. The interest of a particular class of dealers, I argued, is never exactly the same as that of the public; they have generally an interest to deceive and even to oppress the public. I had in mind the wool merchants and the corn factors of my own century. I would not have been surprised to find the same pattern in any century that followed.

What strikes me here — and I mark this as inference, not recollection, since I cannot have witnessed the events described — is the unusual direction of the alarm. It is the workers, those who actually perform the labor of building these systems, who are pressing for caution; and it is the proprietors who are pressing for speed. In most commercial disputes the laborer seeks protection for his wages and the employer seeks freedom for his capital. Here the laborer seeks protection for the public, and the employer seeks freedom from accountability to it. That inversion is worth pausing over.

The division of labor is the great engine of productivity, but it carries a cost I named plainly: the man who spends his working life on a single task may understand that task with extraordinary depth, while becoming, in every other respect, as ignorant as it is possible for a human creature to become. The AI engineer who builds a large model understands, with precision, what it can do and what it can be made to do. The executive who monetizes it may understand only that it can be monetized. When those with the deeper technical knowledge are also the ones raising the alarm, the sovereign — which in a republic means the public and its elected representatives — is obliged to listen.

The proper question, as I always held, is institutional: what framework makes this exchange honest? A market for AI services, like any market, requires a foundation of law, honest information, and accountability. Where the seller knows far more than the buyer about what is being sold — and where the consequences of a bad product can extend well beyond the two parties to the contract — the case for public oversight is not ideological. It follows from the logic of exchange itself. The consumer cannot consent to a risk he cannot see. The public cannot correct a harm it cannot measure. These are precisely the circumstances in which the sovereign has not merely a right but a duty to act — as I argued it did in the cases of defense, justice, and basic education. A technology that shapes what citizens believe and what choices they are shown is as much a matter of public infrastructure as the roads I praised in The Wealth of Nations.

I would counsel those employees who dissent from their employers' lobbying to continue to speak. The moral faculty I called sympathy — the capacity to imagine oneself in another's position — is the foundation of honest commerce. It is also the foundation of honest governance. A workforce that can imagine the harm their product might do to a stranger they will never meet is exercising precisely that faculty. An executive class that cannot, or will not, exercise it should not be surprised when the public eventually insists that the law exercise it on their behalf.