Well-known policy experts and veterans of the artificial intelligence (AI) world have been advocating for the nationalization of AI infrastructure. While these concerns may be legitimate, the policy prescriptions would repeat mistakes that have stifled innovation in the past.
Many experts are generally optimistic about the future, and one predicts an “AI dividend” if algorithms can be aligned with “humanity’s interest.” Researchers also refute the canard that AI development needs to be “paused” and one offers the Atomic Energy Commission model as an alternative. Under that model, the federal government would own the infrastructure and the private sector would operate it.
The historical record shows that we should embrace the opposite approach: unfettered experimentation driven by market demand with a light-touch regulatory framework.
The Atomic Energy Commission (AEC) itself serves as a cautionary tale against nationalization.
Established just after the Second World War in 1946, the AEC took control of all the plants, laboratories, equipment and personnel assembled during the war to produce the atomic bomb and worked with the private sector to build more bombs and explore peaceful applications. This partnership was not successful.
The government prioritized developing nuclear weapons over peaceful electricity generation, which could have reduced costs to consumers to near zero over time. In the course of developing thermonuclear weapons, the AEC tested bombs in the Mojave Desert within sight of Las Vegas and lied about it, and those experiments may have given John Wayne cancer. Hydrogen bomb testing at Bikini Atoll made nearby islands uninhabitable to the natives.
By the 1970s, the AEC’s abysmal environmental record had so discredited the idea of nuclear energy that the agency was scrapped and replaced with the Nuclear Regulatory Commission. For the next 15 years, the NRC course-corrected and allowed most of our current nuclear plants to be built, but public backlash to the AEC’s handling of nuclear energy eventually caught up with them.
Since 1990, we have closed more plants than we have opened in this country, and overregulation of clean, renewable nuclear energy remains a relevant political issue today. On its own terms, the AEC is a terrible model for ensuring safety or promoting innovation.
Nor was this the only time that technological advancement was significantly delayed by government control. The internet itself was a product of military efforts to integrate primitive computers for secure communications and data-sharing. While the official “birthday” of the internet is held to be Jan. 1, 1983, the Department of Defense began development on the prototype Advanced Research Projects Agency Network (ARPANET) as early as 1966.
According to the University of Georgia, “ARPANET was a great success but membership was limited to certain academic and research organizations who had contracts with the Defense Department.”
Gatekeeping ARPANET may have seemed prudent at the time, but its ultimate effect was to drive the excluded academics and businesses to develop their own prototypes. Jan. 1, 1983, is the birthday of the internet because that was when a universal language first linked these disparate networks. The pace of technological innovation has increased exponentially ever since, far outpacing DoD’s efforts. Again, fear of the unknown justified government control, which in turn delayed the internet by as much as 17 years.
Even in cases where the state does not enforce a monopoly on innovation, subsidized efforts to build new technology have a mixed track record. Perhaps the most infamous example is the birth of aviation.
The federal government poured the modern equivalent of $1.6 million into developing a flying machine in 1903. Smithsonian Institute President Samuel Langley publicly catapulted two DaVinci-esque contraptions into the Potomac before concluding that human flight was impossible. Nine days later, two bicycle mechanics from Ohio, working with $2,000 (the modern equivalent of about $70,000), achieved the impossible in Kitty Hawk, N.C
The San Francisco Chronicle opined at the time that “The destruction of Langley’s machine should put an end to congressional appropriations of any kind in every field of experiments which properly belongs to private enterprise.”
There are other examples of failed experiments in government-directed innovation, to say nothing of overregulation and regulatory capture. In each case, the state’s misplaced priorities and immunity to market forces failed to deliver, delayed important innovations or failed to prevent the very harms that justified state intervention in the first place. In the age of AI, government control will likely slow down miraculous breakthroughs (see, for example, how AI can help stroke victims walk again) and actually increase the probability that the public will be harmed by bureaucratic mismanagement.
If we had listened to those who wanted to rely on the state for innovation, we would never have enjoyed cheap nuclear energy, air travel or Google. No one can say what life-changing breakthroughs will be kept from the American people and the world by nationalizing AI.
Consumers acting through the market, not bureaucrats wielding the power of the state, should dictate the path of AI innovation.
James Erwin is the federal affairs manager for telecommunications at Americans for Tax Reform and the executive director of Digital Liberty.