Does AI need to be regulated? The short answer is yes.
But that’s not the relevant question. First we need to know what, in fact, it’s possible to regulate, and then how it can be done without stifling innovation or risking “regulatory capture” by the loudest voices with the deepest pockets. The one thing you can take to the bank is that there’s no one in government that can or should try to figure that out. The entire range of players creating the disruption will need to be the ones to frame any path toward sensible regulation.
Finally, a reason to check your email.
Sign up for our free newsletter today.
One can make a reasonable market-centric argument for doing nothing and instead apply to new regimes the existing laws and regulations that are part of an already-vast regulatory state. But the political momentum for new regulation is likely unstoppable.
You can see it in the data center resistance movement, which has metastasized as a political issue—one that has swept up all the anxieties about AI. There is no better evidence of that than a recent New York Times opinion piece, “Hating Data Centers Is a Winning Issue.” It calls the issue the “Democrats’ greatest untapped opportunity” to take on the “behemoth A.I. interests.”
While ginned-up hype and meddling by the professional grievance industry is fueling data-center resistance, substantive issues are nonetheless creating reasonable concerns. Thus, we also find some Republicans embracing the resistance. Missouri Senator Josh Hawley recently published a kind of AI manifesto with a clear class message: “So I am asking my party to choose to stand with our moral covenant and not with the barons.”
Doubtless the senator knows that the phrase “robber barons” originated in the nineteenth century with rivals of Cornelius Vanderbilt trying to stoke outrage over the then-unprecedented political and economic power of Gilded Age rail and steel titans. As one writer put it in The Atlantic in 1881: “In less than the ordinary span of a life-time, our railroads have brought upon us the worst labor disturbance, the greatest of monopolies, and the most formidable combination of money and brains that ever overshadowed a state. The time has come to face the fact that the forces of capital and industry have outgrown the forces of our government.” Does that framing sound familiar?
Consequential technologies always become political. Thus far, however, no one has found a path to soothe anxieties. As a recent Ispsos poll found, a majority of both Democrats and Republicans don’t know which party has a “better plan” for dealing with AI. While there is clearly no magic-bullet solution for the array of AI issues, there is one path that offers the best prospect for sensible measures: self-regulation.
Self-regulation may sound like an oxymoron, but it’s an old and often successful means for juggling the challenges created by rapidly emerging technologies or industries. A key enabler for effective self-regulation—and the biggest hurdle—is the need for the purveyors of the disruption to collaborate. The problem today is that the AI-tech community is wired for competition, not collaboration. Given anti-AI political momentum, however, there is an urgent need to collaborate in order to ensure the freedom to compete.
As George Washington University law professor Aram A. Gavoor has written, when “industries adopt genuinely effective self-regulatory regimes, the threat of external regulation correspondingly diminishes. Companies engage in voluntary governance in part to stave off federal or state intervention.”
To understand the prospect for self-regulation, and what is and isn’t different this time, every executive in the tech industry needs to read Gavoor’s paper. It is a magnum opus not only for its presentation of the history and structures of effective self-regulation but also because it is the first such guide encompassing the many recent regulatory attempts made by presidential administrations and Congress, especially in light of recent Supreme Court decisions that, as Gavoor notes, have “reduced the administrative state’s interpretive and enforcement leverage.”
Gavoor’s paper is no mere academic exercise: it describes how self-regulation “can be economically rational rather than merely aspirational,” and how “codification can support federal preemption where state regulation becomes conflicting or excessive.” In short, it offers a path that is as close to a get-out-of-jail-free card as is humanly—and politically—possible.
Governments always want to regulate. The origin of regulating commerce, technologies, and human behaviors is coincident with the dawn of government itself. Nearly 4,000 years ago, for example, one finds the famous Babylonian Code of Hammurabi, which had, among its hundreds of rules, perhaps the earliest example of technology regulation: in that case, relating to building construction, which established that if a building collapsed and killed its owner, the “builder shall be put to death.”
The industries that have created the world of AI will not escape the regulatory urge. Indeed, the emergence of every consequential technology has triggered not only new classes of regulations but also often entirely new government regulatory entities. The railroads brought the Interstate Commerce Commission; radio, the Federal Radio Commission (later the FCC); pharmaceuticals, the Bureau of Chemistry (later the FDA); aircraft, the FAA; the atomic pile, the Nuclear Regulatory Commission; and so on.
History also reveals a significant time lag from invention to regulation. It took 50 years after Edison’s first power station for Congress to create the Federal Power Commission, and 50 years after the Wright Brothers before President Dwight D. Eisenhower signed into law the Federal Aviation Administration. It’s been almost 50 years since the invention of “machine learning,” which arguably began with a seminal 1986 paper by the computer scientist Geoffrey Hinton, credited as the godfather of AI. If the correct benchmark for AI’s birth is not the idea in the math but the physical tool—the 1993 founding of Nvidia and the graphics processing unit (GPU)—then the AI industry has a little time before Congress adds yet another edifice to the regulatory state.
Of course, every new technology is different. The reason any new technology succeeds is precisely because it is different and thus offers benefits, many unforeseen, that people want. Equally true, all technologies bring unintended side-effects, trade-offs, and often risks. Social and cultural side effects are invariably the most challenging to mitigate. The invention of the car ended the drayage industry and led directly to the collapse of more than 10,000 companies that manufactured carriages, about which little could be done. It also led to car accidents and pollution. All of which spawned new classes of regulation.
The factors leading to a car crash are different from those that cause a horse-and-buggy crash or a computer crash. Design standards and professional codes of conduct can ameliorate new technical risks but can never eliminate them, much less obviate bad behaviors—thus, the need for regulation.
But as Gavoor notes in his paper: “Statutory processes move slowly, political coalitions fracture quickly, and technical definitions expire and fail to continue serving the markets they seek to govern. The result is a recurring cycle of over- and under-correction, in which regulators’ ambition exceeds administrative capacity, and regulatory text ages faster than the technology it purports to discipline.” The solution, again, lies in self-regulation that originates with the fast-moving industries themselves.
In the wide array of approaches to self-regulation, one finds a continuum from professional ethics to professional and technical standards and then, at the far end of the spectrum, to private self-regulatory organizations to which Congress grants enforcement authority. Perhaps the most salient example of the latter is FINRA—the Financial Industries Regulatory Authority, a privately run but SEC-endorsed oversight body. According to the SEC: “The creation in FINRA of a super regulator was part of a broader attempt to understand, and therefore effectively regulate, the whole of the ever more fragmented securities market.” The digital market is similarly fragmented and complex. It is not hard to imagine a Digital Technology Regulatory Authority (DITRA?) organized and designed by industry and then congressionally granted SEC-style oversight and enforcement authority.
Gavoor’s paper reviews the contemporary state of play regarding various AI regulatory and legislative initiatives and proposals, at both the state and federal level, and outlines the business case and an “institutional design to achieve meaningful outcomes” via self-regulation. He concludes, correctly, that self-regulation “offers the best prospect for protecting the public without suffocating the very innovation on which the future of AI depends.”
It has not escaped anyone’s attention that our current technological disruption is centered on thinking machines rather than on mechanical ones. One is tempted to repurpose René Descartes’s well-recognized maxim, cogito, ergo sum with cogito, ergo moderari—I think, therefore I regulate. Let’s hope that Big Tech is thinking about all this and takes the time to read, and heed, Gavoor’s roadmap.