Jennifer Huddleston
There has been a recent resurgence in fears about AI. A range of voices, from Sen. Bernie Sanders and Steve Bannon to former engineers and leaders of various AI labs, have been calling for a pause in AI development. Many of these calls are accompanied by science fiction-esque doomsday scenarios about the worst that could happen. Yet a rush to regulate out of fear could have longer-term negative consequences than the risk it claims to address.
The last week has focused a lot on the potential risks of AI, distracting from the ways this broad-purpose technology is already transforming our lives in ways we don’t think about or see. We no longer think of autocomplete or text-to-speech as AI, even though they are. We often don’t see the remarkable medical research that is being powered by AI until it directly impacts us or a loved one. AI is opening new doors for entrepreneurship and creating fresh opportunities in skilled trades. Any conversation about how to govern this technology should consider two truths: yes, the risks could be real and worth taking seriously, but so are the benefits already showing up in people’s lives and livelihoods.
So, what should be done?
First, it is important to note that any particular developer or lab could decide to pause their own progress. Nothing stops OpenAI or Anthropic from choosing not to develop certain aspects of their product that they deem too risky or think would be harmful. Different companies will make different public choices about where exactly to draw lines. Other companies may react, adjust, and sometimes push back based on their own products. That back-and-forth is how industry norms take shape in a fast-moving field. The decision to draw similar lines at the same time doesn’t prove a need for regulation; it shows that norms around some of the most pressing risks can be resolved in other ways.
For example, when a company slows down to work through a specific, well-defined risk—cybersecurity vulnerabilities, for example—that’s a company managing its own product responsibly. It can course-correct, iterate, and resume once it has addressed the concern. Industry self-regulation and companies collaborating on shared standards around thorny issues like cybersecurity can provide the flexibility needed to deal with problems in real time while maintaining beneficial development. Unlike policy, norms and self-regulation let the industry adapt and create best practices as the technology and the risks evolve, rather than locking everyone into rules written for a snapshot in time. It also allows different choices for companies with different products and more targeted solutions within each company.
A government-mandated pause would raise several concerns and likely fail to fulfill its alleged safety improvements. Government rules are slow to write and slower to change. A regulatory framework built for today’s models may actively hinder tomorrow’s, preventing what could be better or safer responses.
In a highly competitive, fast-evolving industry, government intervention risks tilting the playing field—intentionally or not—toward incumbents or politically favored players at the expense of innovation. Calls by industry leaders to engage in more formal regulation should also give pause to what it could mean for the future of the industry. Today’s leaders may want regulation, as it could lock in their preferred developments and their market position. Meanwhile, smaller players would find it more difficult to enter the market and challenge rivals.
Even if it seems like this is for “safety,” a deeper look can reveal how it could lead to regulatory capture for the industry leaders at the expense of a more competitive market that would better benefit consumers. While there may be risks to address on issues like cybersecurity, emerging industry norms and existing policy tools are likely to be able to respond to most, if not all, of these.
It’s also worth remembering that this isn’t happening in a vacuum. AI development is a global, competitive race. Pausing US AI development could fail to provide adequate cybersecurity defenses against adversaries who do not seem likely to restrict themselves similarly. It could also impact the diffusion of the US AI stack if other countries’ companies are capable of providing better or more cutting-edge products. In the long run, this could raise concerns about the values of a free society when it comes to the development and deployment of AI technologies. In the absence of competitive US products, this risks leading to AI products being developed by those with a more totalitarian intent.
Additionally, AI is a uniquely accessible tool to collect and understand information. Because of this, government restrictions around AI could end up being restrictions on free expression or access to information. This can set a dangerous precedent by interfering with the development of a technology and laying the groundwork for broader censorship.
As Yogi Berra once said, “It’s tough to make predictions, especially about the future.” We shouldn’t presume only the risks as we try to imagine the future. AI is far from the first technology where concern about risks led to fears of doom. Twenty-six years ago, we faced fears that the Y2K bug would bring about the apocalypse. Some of the same leaders were concerned that GPT‑2 was too dangerous for release in 2019 despite very minor capabilities by today’s standards. Doomsday predictions about “thinking machines” go back to the 1950s and 1960s. But the reality is that we’ve found better solutions to the concerns that balance the risks while allowing technology to progress in beneficial ways without shutting down innovation.
