Business
The Agreement Nobody Made
The next great threat to competition may resemble a cartel while lacking every essential ingredient of one.
No executives will meet in secret. No prices will be fixed by agreement. No incriminating correspondence will surface. The companies may compete ferociously, their algorithms shielded behind proprietary walls, each convinced of its own superior intelligence.
Yet their machines may increasingly reach the same conclusions, reject the same uncertainties, and eliminate the same alternatives.
The distinction is fundamental.
Collusion requires coordination among competitors. Convergence can emerge without coordination of any kind.
One involves concerted behavior. The other may require nothing more than sufficiently similar intelligence.
Artificial intelligence introduces the possibility that competitors could become economically synchronized without ever becoming legally or organizationally connected.
Competition would survive in form even as the diversity of its outcomes began to disappear.
Markets have never required companies to make consistently intelligent decisions. They have benefited from companies making different ones. An economy can withstand spectacular errors of judgment. It may be far less capable of surviving a future in which every sophisticated participant makes the same mistake simultaneously.
In a functioning market, error is ordinarily self-correcting precisely because it is unevenly distributed. When one participant misjudges the future, another profits by seeing it differently. The correction emerges from the disagreement itself. A synchronized error offers no such mercy. There is no counterparty on the other side of the mistake because everyone whose judgment matters has made it at the same moment, for essentially the same reasons.
Consider energy inventories, semiconductor capacity, shipping commitments, and industrial investment. A dozen independent enterprises might anticipate the same shortage, pursue the same alternatives, withdraw capacity simultaneously, or abandon investments their predictive models deem uneconomic.
Each decision is individually rational. Collectively, those decisions could create the very conditions the algorithms were designed to anticipate.
Traditional remedies might prove remarkably ineffective. Breaking up companies would accomplish little if their successors continued making essentially identical decisions. Greater transparency might accelerate convergence rather than prevent it. Regulators accustomed to investigating coordinated conduct could confront synchronized behavior for which no agreement exists.
The deeper difficulty is legal as much as economic. Traditional antitrust enforcement often seeks evidence of coordination: a meeting, a message, a contract, or a decision made together. Convergence need involve none of these. It is a condition, the predictable result of intelligent systems trained on overlapping information and rewarded for the same definition of prudence, arriving independently at essentially identical conclusions. There may be no unlawful agreement to prohibit because no agreement was ever made.
Nor can the participants necessarily be counted on to break ranks. A firm whose model dissents from the prevailing judgment may find itself punished rather than rewarded for its independence, quarter after quarter, by investors comparing its performance with competitors that all made the seemingly safer, identical choice.
No one imposed that discipline. Everyone enforces it.
Consider a more unsettling institutional possibility.
Five competing banks independently evaluate a strategically important semiconductor facility. Their credit committees employ different AI systems, all drawing upon overlapping economic assumptions. Each concludes that the project presents an unacceptable risk.
Financing disappears.
The national government considers the facility essential to industrial sovereignty. Financial regulators identify no improper lending practices. Competition authorities uncover no collusion. Bank directors have fulfilled their obligations to shareholders.
Every institution has performed its assigned responsibility.
Together, they have made the government’s industrial strategy impossible to execute.
What makes this possibility remarkable is not the rejection of a single investment. It is the prospect that governments could eventually lose the practical ability to pursue entire categories of industrial development because independently operating intelligence systems have converged on the same definition of economic rationality.
A state may retain every formal instrument of industrial policy—subsidies, guarantees, procurement, and law—only to discover that each instrument now depends on the practical consent of models it neither controls nor can readily overrule. No individual model has necessarily done anything a regulator could identify as improper.
Sovereignty would not have been seized. It would have been priced, declined, and filed away as risk.
The consequences could extend far beyond industry.
For generations, markets have depended on competing interpretations of an uncertain future. Some proved correct. Others were spectacularly mistaken. Yet the disagreement itself made room for discoveries that prevailing opinion could never have anticipated. An economy increasingly governed by convergent artificial intelligence might gradually lose that creative disorder.
The improbable invention, the unconventional energy system, the unfashionable technology, the industrial possibility that no predictive model can justify—all could be denied investment before their potential is ever demonstrated.
The future would become whatever the prevailing intelligence considered sufficiently probable to finance.
The most consequential monopoly of the artificial intelligence age may therefore have no identifiable owner, no corporate headquarters, and no regulatory jurisdiction.
It may be a monopoly over what the world considers possible.
And nobody will have agreed to create it.