Democracy in Artificial Intelligence

With each day, it seems, artificial intelligence (AI) arrives in a new sphere of American life. This is a revolution, but an iterative one, advancing by stages—and not unlike the past revolutions of the automobile, of mass electrification, and of cellular devices and connectivity. In its application, AI is general-purpose: It will become—indeed, has become—intertwined in the operations of innumerable devices, platforms, tools, and systems, rendering human endeavors more efficient and human efforts more potent. As did electricity in the 20th century, AI will become ubiquitous in the 21st.

All revolutions have their counterrevolutionaries—be they royalists opposing democratic advances or Luddites smashing stocking frames. So too does the AI revolution, but those who wish the technology to succeed in facilitating human flourishing have produced all manner of proposals to mitigate its disruptions and relieve the misgivings of the American people with respect to how it may transform the world they know. These plans proceed, however, from diverse premises, and their prescriptions deviate accordingly. One of the latest, a white paper published by OpenAI, the proprietor of ChatGPT, takes a strikingly ambitious posture: “History shows that democratic societies can respond to technological upheaval with ambition: reimagining the social contract, mediating between capital and labor, and encouraging broad distribution of the benefits of technological progress while preserving pluralism, constitutional checks and balances, and freedom to innovate,” the document reads.

The reader learns that innovation must be managed, funded, and directed by the state. OpenAI favors a whole-of-society effort to remake society itself, changing fundamental facts of economic life and doing away with many of the forms of the ancien régime. Regulation, subsidization, wealth redistribution, a reimagination of the relationship between labor and capital, and the institution of an expanded safety net will be the tools of the craftsmen of the new order. America must “act collectively, at scale.”

The logic and language of OpenAI’s white paper summons to mind Alexis de Tocqueville’s warning against the soft, friendly sort of despotism to which democratic nations are susceptible. Such a government acts as a schoolmaster, Tocqueville wrote: “It is absolute, detailed, regular, far-seeing, and mild. It would resemble paternal power if, like that, it had for its object to prepare men for manhood; but on the contrary, it seeks only to keep them fixed irrevocably in childhood.” It arrives in the guise of a democratically sanctioned regime: “each citizen…can still fancy that in obeying he submits only to himself.” And yet, it is a despotism, nonetheless, wrests liberty from the people and puts the means of power to an “administrative despotism,” rendering the citizen unfree and dependent on the state.

The word democratic recurs throughout OpenAI’s report, appearing 13 times—e.g., AI must “remain[]…aligned with democratic values.” The word we appears 18 times, and the report opens with the section heading: “Let’s Talk.” The tone is warm and fraternal, even neighborly. After all, to this mode of thought, government is simply the name we give to the things we choose to do together.

This is a cheering formulation, but it does not describe the realities of governance. The acts of any democratic government are carried out by the representatives of transient and mutable political coalitions, animated by diverse opinions, passions, and interests. Those representatives, subject to all the difficulties of human imperfection and public-choice factors, deploy the coercive force of the state to compel obedience. 50.1 percent of the population may agree—for a short time, at least—that a certain thing ought to be done, but only rarely do such determinations amount to an expression of “the will of the people” or some similar abstraction. People, being people, disagree with one another, and free societies clear away a large zone of liberty within which citizens, as individuals and within their communities, can pursue the good life as best they can. Of course, certain questions must be answered by public policy, and in such cases only one answer can be given. But that is reason to limit the activities of the government and leave most questions to other institutions: to families, churches, private businesses, etc. The arrangements of civil society and the market more perfectly allow “us” to get what “we” want, because individuals and communities may decide questions for themselves, even though, inevitably, some will decide differently from others. Efforts to “combine centralization and the sovereignty of the people” might “give…some respite” to some, Tocqueville wrote, but this settlement little resembles true liberty.

“At OpenAI, we believe we should navigate it through a democratic process that gives people real power to shape the AI future they want, and prepare for a range of possible outcomes while building the capacity to adapt,” the report reads. In this sense, democratic control means conformity; all must bend to the dictates of the state. Tocqueville’s soft democratic despotism “covers [society’s] surface with a network of small, complicated, painstaking, uniform rules through which the most original minds and the most vigorous souls cannot clear a way to surpass the crowd.” Entrepreneurial experimentation and discovery, the fundamental activities of a market economy, cannot survive if American innovators must work within the confines of what is imagined to be good and desirable by politicians and bureaucrats.

The company professes to value “the freedom to innovate,” but this freedom as conceived in the white paper is reduced to the freedom to innovate in ways of which society approves. It ought to be noted that OpenAI has in the past adopted a more genuinely free-market position, but OpenAI now proposes to transform innovation—the domain of the entrepreneur, free to follow whatever course seems to him to be most useful and profitable—into a collectivized teleocratic venture. “As advanced AI makes more decisions that affect people’s lives, societies need shared clarity about what these systems are supposed to do, what values should guide them, and how well they are performing,” the report reads (emphasis added). Not just the parameters of permitted conduct but the ends of innovation themselves are to be defined by the dictates of the whole. Such determinations, OpenAI says, cannot be entrusted to “engineers or executives behind closed doors”—all oars must pull in the same direction under the supervision of the pilots of the ship of state. This model would, to borrow the 1938 words of the New Deal critic Garet Garrett, transfer “the ultimate power of initiative…from the hands of private enterprise to government.”

The scope of the AI revolution and the sheer power of the technology itself raise the stakes, so to speak, of the debates over its regulation—for both the regulators and the regulated. For businesses and citizens, the regulation of AI will control the tools that, increasingly, are coming to define modern life. For politicians and bureaucrats, the immense potential of AI makes the prospect of enforcing their will upon private industry far more attractive.

And when the government issues fiats, companies that resist must be put down.

In February, the Department of Defense designated Anthropic, the developer of Claude, a “Supply-Chain Risk to National Security.” Anthropic objected to the use of AI for the purposes of domestic surveillance and wholly autonomous weapons and wished to retain the technical ability to restrain the government’s use of its tools. This the Trump administration could not brook. And rushing past the sensible remedy—terminating its contract with Anthropic in favor of another AI developer—the government chose a punitive (and likely illegal) measure whose object was to devastate the offending firm. It did not satisfy the administration to find a satisfactory private-sector partner; it desired control.

The Anthropic dispute illustrates the failures of imagining “democratic” control of AI as something lovely or communal (in the proper sense of the word). “We” did not seek to subdue Anthropic; it was the act of the Trump administration and the officials who staff it, who acted on the basis of incentives and for reasons that attach to them and them alone, and not to the American people. The act of governing does not transubstantiate whoever happens to hold office into the body of the American people. To assume as much requires a metaphysical sleight of hand that better reflects the abstractions of Rousseau’s theories than the truth of self-government.

The dispute also returns the argument to Tocqueville. The Frenchman wrote of the leveling effects of democracy, which, with its penchant to centralize power in a single domineering authority, tends also to destroy the mediating institutions that stand between the individual and the state. Without these protective barriers, the individual is like a Parisian workman confronting a counterrevolutionary battalion alone in the middle of a deserted square: he is exposed, isolated, and, by comparison to his adversary, weak. Thus situated, he cannot resist the state’s force.

Voluntary association is the free man’s defense against tyranny. “A political, industrial, commercial, or even scientific and literary association is an enlightened and powerful citizen whom one can neither bend at will nor oppress in the dark and who, in defending its particular rights against the exigencies of power, saves common freedoms,” Tocqueville argued. Even a business such as Anthropic, bound by its own interest and bottom line, is a locus of power that serves as a counterweight against the power of the government. As AI becomes an ever-more-ubiquitous presence in American life, this sector in particular must remain free to stand athwart the designs of the men of system in Washington, D.C. Subjecting the development and deployment of AI to an overbearing regulatory regime or viciously punishing noncompliant developers does not create more liberty, but centralizes unchecked power in the federal government.

The very ubiquitousness of AI gives cause for an unflinching vigilance against centralized and collectivist control, “democratic” or otherwise. The trajectory of its development will influence how businesses operate and how citizens gather and process knowledge. Its regulation, therefore, is of far more consequence than the regulation of most other industries. “I would, I think, have loved freedom in all times,” Tocqueville wrote; “but I feel myself inclined to adore it in the time we are in.”