AI is a general-purpose technology, but often the motivations for regulation are more about specific applications. Still, there may be some areas where policy certainty is needed to allow the market and investment to flourish.

Unfortunately, AI has become the new villain in many policy narratives that focus only on its potential downsides rather than on the benefits it is already delivering. Technologies we are used to, like talk-to-text and autocomplete, are no longer thought of as AI, yet they are. Certain technologies that consumers do not directly see, such as some tools that catch fraud or help respond to natural disasters, are not often recognized as AI, but they are too. Regulatory restraints can and do affect such tools as much as the generative AI chatbots that have become synonymous with AI.

Pacing problem / In the preceding article, Randall Mayes concludes, “The risks of moving too fast are real, but so are the risks of moving too slowly.” Moving quickly toward regulation is a more European, “precautionary” approach that is calibrated to the technology of today, limiting not only potential harm but also future innovations and options that do not fit neatly into what the regulations permit. Of course, this regulatory straitjacket affects more than just AI.

As with many technologies before it, the “pacing problem” has shown that technology often moves faster than policy does. In many cases, this supposed problem is a benefit, allowing innovators and consumers to access new technologies with minimal government or regulatory intervention and allowing technology to evolve to meet the demands of consumers rather than bureaucrats. However, in some cases—particularly where industry is already regulated, such as health or transportation—this can indeed create problems as regulation is unable to adapt to the speed of technology.

The pacing problem also means policymakers should reconsider existing regulation. Since policy change often moves slowly, static policy can hinder innovation and technologies that it was not designed to accommodate. For example, AI may, in some cases, provide better and safer methods that call into question whether existing regulations are still needed. The response should not be to force new technology to comply with regulations that it renders outdated, but rather to consider whether such regulations are needed in the first place. If not carefully considered, outdated regulation could discourage important innovation that it was never intended to affect.

Focus on harm / As noted, because of the broad nature of AI, there is unlikely to be a single regulatory approach. Much of the debate over AI regulation and technology is better understood in terms of the particular applications of the technology. As a result, often the true question should not be how to regulate AI generally, but rather how to prevent a specific harm or respond to a specific concern given a potential application of the technology. In some cases, regulators have already affirmatively stated that using AI to engage in an unlawful act, such as fraud or discrimination, does not undo the burdens on the individual. In other cases, AI may raise concerns about civil rights and civil liberties that call for careful consideration of how to restrain potential government abuse. As a result, what regulatory tools—if any—are appropriate is better considered in light of the application than the technology more generally.

When it comes to AI and other general-purpose technologies, regulation should be reserved for clearly agreed-upon, irreversible, and catastrophic harms, and it should be proportionally responsive to the likelihood of such risks. Policy should focus not only on preventing worst-case scenarios but also on creating an ecosystem that allows for best-case scenarios that could never be foreseen. Similarly, more adaptive frameworks, such as voluntary consultations and standards, can evolve with risks and technological changes more readily than top-down approaches based only on the technology of the day. As Mayes notes: “The goal is not risk elimination. No technology worth having is risk-free. The goal is proportionality: matching the stringency of regulatory response to the severity and probability of harm, while preserving the innovation capacity that makes beneficial outcomes possible in the first place.”

The risks of over-regulation should not be underestimated. AI is moving exceptionally fast, and policy seems to move ever more slowly. This risks regulation that could lock in the technologies of 2026, preventing future improvements that might naturally alleviate such problems, even when regulation does move quickly. Regulators’ crystal balls are rarely able to predict the future that innovators can provide consumers if allowed.