Fears about artificial intelligence have gone mainstream, and an intense panic now grips the minds of the press, policymakers, and the public. A familiar cast of technocratic leaders—Barack Obama, Rahm Emanuel, Kamala Harris, Bernie Sanders, Elizabeth Warren, and even King Charles—have chimed in with calls for some sort of “AI pause” or other steps to give government major new powers to ensure proper “alignment” of AI models, algorithmic applications, and advanced computation.
It’s a good moment to recall some timeless lessons from social thinker Friedrich Hayek. Today’s AI alignment debate contains echoes of the old economic calculation debate and the planner’s mentality that viewed society as a math problem to be solved. Hayek (1899–1992) spent his life demolishing that reasoning, while warning of the “fatal conceit” of elites who thought they could reorder entire economic systems and remake human nature through freedom-crushing, top-down interventions.
Finally, a reason to check your email.
Sign up for our free newsletter today.
Previously, AI safety discussions were hidden from public view, confined to nerdy tech conferences, obscure internet discussion boards, and the plots of sci-fi stories and movies. That all changed following the splashy resignations of some AI lab employees, worried that developers like Anthropic and OpenAI may have lost control of their creations. Over 1,300 AI lab employees also signed a letter calling for “pacing the frontier” of AI development, and the CEOs of many major labs have called for government assistance in slowing the advancement of recursive self-improvement (RSI) or artificial general intelligence (AGI).
The AI debate increasingly features even more apocalyptic overtones and extreme proposals than did debates during the nuclear era. Regulatory ideas include moratoria on AI development potentially enforced by new global regulatory treaties and bureaucracies; mandated “kill switches” that let government take control of AI systems; nationalization of leading labs or “golden share” socialization schemes to give governments ownership stakes in leading firms; “FDA for AI”-like licensing regimes; and more. At the state level, more than 1,800 AI bills have been floated to regulate virtually every facet of computational innovation and algorithmically enabled speech, creating an unprecedented patchwork of technocratic red tape.
Central to all these ideas are fears about AI “alignment”—the process by which labs try to ensure that AI systems behave in ways consistent with human goals and values. Many fear that a misaligned AI could lead to catastrophe, requiring aggressive government intervention.
This is where a dose of Hayekian thinking is particularly useful. Hayek would tell us that AI alignment is not a computable math equation, and that efforts to treat it as such will lead to the sort of authoritarian thinking and proposals we witnessed with previous efforts to remake economies and humanity according to some grand design.
In his 1974 Nobel Prize acceptance speech, Hayek warned of the “pretence of knowledge,” or the belief that experts understand society well enough to direct everyone else’s choices according to their supposedly enlightened plans. He argued that the “limits to his knowledge ought indeed to teach the student of society a lesson of humility which should guard him against becoming an accomplice in men’s fatal striving to control society—a striving which makes him not only a tyrant over his fellows, but which may well make him the destroyer of a civilization which no brain has designed but which has grown from the free efforts of millions of individuals.”
Hayek teaches us that technical “alignment” can be a pretext for dangerous political control of our economy and liberties. Empowering global elites to guide humanity’s acceptable risks and technological future raises other risks in the process. Aligned with whose values? Imposed by whose authority? And at what cost? “Progress by its very nature cannot be planned,” Hayek wrote in The Constitution of Liberty, but instead “civilization [is] the accumulated hard-earned result of trial and error,” and “the sum of experience.”
Other scholars advanced similar reasoning to explain how safety itself is an evolutionary discovery process of continuous learning and improvement—a sort of spontaneous order in Hayekian terms. Safety emerges through decentralized experimentation, continuous feedback, knowledge gained from failures, and ongoing societal resilience. This was the great lesson of political scientist Aaron Wildavsky’s 1988 book Searching for Safety, which warned of the dangers of “trial without error” as compared with ongoing experimentation and iteration.
Safety emerges as a byproduct of experience, incremental improvements, and learning, not top-down, anticipatory controls. “If you can do nothing without knowing first how it will turn out, you cannot do anything at all,” Wildavsky argued. Worse yet: if society fails to advance the technological frontier, the “very lack of change may itself be dangerous in forgoing chances to reduce existing hazards,” leading to an overall decline in human well-being.
Today’s developers of AI and advanced computational systems instinctively appreciate these lessons, too. “Safety is paramount,” says Nvidia CEO Jensen Huang. “In a lot of ways, it’s job one, however safety is an engineering problem. If you build a product or a service and you’re not confident in its functionality, capability or safety, then don’t release it.”
Yes, law will still play an important role in the future of AI. But real AI safety will require a more boring but pragmatic blueprint. We already have many norms, rules, and legal processes to guide us—especially in various heavily regulated sectors like health, transportation, and finance. But what will work best, and ensure the greatest degree of both long-term safety and innovation, is what the great Hayekian legal scholar Richard A. Epstein called “simple rules for a complex world.”
Some of these steps include bolstered cybersecurity coordination; continuous communication about vulnerabilities; hardening digital infrastructure; constant software patching; ongoing incident reporting; and consensus-driven safety and security standards and best practices. Most of these are already in place today but are constantly being refined. Independent model testing and certification bodies will also become more prevalent, and AI liability and insurance will play a role, too, as Judge Glock recently observed in City Journal.
Aligning technical systems with human values in any context is not something that can be preemptively decreed through global treaties or paperwork-intensive, bureaucratic box-checking exercises. AI safety is an ongoing journey, not a one-and-done, ex ante silver-bullet scheme. As Hayek taught us long ago, “the reliance on abstract rules is a device we have learned to use because our reason is insufficient to master the full detail of complex reality.”