Policies to ensure public benefits from the adoption of artificial intelligence bear resemblance to policies designed to protect communities from climate change and the energy transition.
At this year’s CERAWeek, the energy industry’s largest annual gathering, talk on the main stage focused on the United States’ recent actions in Iran and Venezuela. But on the side stages and behind the scenes, another topic stood above all others: artificial intelligence (AI). How would the energy industry meet surging power demand from AI data centers? How would AI tools unlock new energy resources? How can energy and technology companies forge new partnerships to “win the AI race”?
However, on the topic of AI governance—how policymakers can and should manage the meteoric rise of this technology—CERAWeek attendees had little to say. Although AI technologies are growing and evolving at astonishing speed, governance of the AI sector has primarily been hands off; “let it rip” has been the general attitude among most federal policymakers. But this laissez-faire approach introduces risks that range from the secular (like higher electricity bills) to the profound (like the risk of human extinction).
In the past 20 years or so, the US energy sector has experienced rapid technological changes, including the shale revolution and the rise of battery storage, wind energy, and solar energy. These experiences, coupled with decades of effort to address the risks posed by climate change, provide a perspective that can help policymakers, industry, and the public build a more thoughtful governance structure around the deployment of AI infrastructure and services.
Here, I’ll lay out four lessons from the evolution of energy and climate policy, then discuss their implications for AI governance.
Existential Risk in Climate and Artificial Intelligence
In recent months, some prominent voices in Silicon Valley; Washington, DC; and elsewhere have made the case that building an AI “superintelligence”—an intelligence that can accomplish in hours or days what might otherwise take people years or decades—poses an existential risk to humanity. Science fiction authors have imagined these scenarios for years, but some experts argue that today’s race to scale up AI could mean unleashing catastrophe in the next several years. At the extreme end, “catastrophe” could mean the annihilation of humanity, but more plausible outcomes such as massive job losses, the release of powerful bioweapons, widespread disruptions to critical infrastructure, and many other scenarios should raise alarm bells for policymakers.
How likely are these scenarios to occur? I don’t know. But even if there’s a small chance they do, policymakers need to begin building guardrails to reduce the risk that AI tools unleash chaos.
Climate economists have told this story before. In 2014, Martin Weitzman published “Fat Tails and the Social Cost of Carbon,” making the point that the risks from climate change are uncertain enough, and potentially catastrophic enough, to justify robust public policies to reduce the chance that we render our planet uninhabitable. In other words, even if it’s unlikely that climate change decimates human civilization, the fact that it plausibly could justifies an aggressive response.
Credit: Mike Mareen / Shutterstock
In this way, AI is like climate change on steroids. Like climate change, AI poses an existential risk—even if that risk is small. Like climate change, AI poses global risks. And like climate change, the impacts of this risk could be irreversible. Chances are, a superintelligent genie isn’t going to want to go back into its bottle.
But unlike climate change, where the greatest risks lie decades—if not centuries—in the future, the scenarios painted by some AI experts involve catastrophic risks in the next several years. This imminence means that efforts to reduce AI risk must begin immediately. Although some AI companies clearly are concerned with the issues I outline, a lack of federal governance means that each company has no incentive to slow its breakneck race to superintelligence.
Policymakers got a fresh wake-up call when Anthropic alerted the world to the power of its Mythos model, which has the potential to disrupt infrastructure that underpins modern life, such as power grids and banking systems. With luck, policymakers will take the development of Mythos for the warning shot that it is. And there are some signs that they are: new reporting suggests the White House may review new AI models before companies publicly release them. Such efforts may be helpful, but they are a far cry from legislation drafted by Congress that has the potential to be far more transparent and consistent.
In short, decisions about whether and how to publicly release technologies that could upend modern life should not be left to C-suite executives, however well-intentioned they might be. Instead, legislators need to work with the AI industry to develop clear guardrails and rules of the road that ensure AI delivers its promised benefits (which, to be clear, could be massive) without bringing the profound damage that AI clearly is capable of causing.
How Artificial Intelligence Can Learn from Energy Infrastructure
Over the past 20 years or so, energy infrastructure has increasingly become a matter of local contention. Whether it’s the Dakota Access Pipeline, fracking in Colorado, solar energy in Ohio, or renewable energy resources and storage in New York, energy projects of all stripes have faced significant local opposition. Although each case is unique, much local opposition comes down to two simple realities: people want a say over what gets built in their communities, and they don’t want big changes unless those changes come with significant benefits for the community.
Although fossil energy sources bring significant environmental and health risks, coal, oil, and gas production generally receives substantial public support in the communities where the industries operate. In most cases, this support stems from the fact that the industry provides high-paying jobs, large royalty payments to landowners, and public revenue that supports local schools and other essential services.
Recognizing the importance of these local benefits, wind and solar project developers have increasingly (though not universally) focused on consulting with communities and delivering them meaningful economic benefits. For example, one large solar project in Indiana that met with significant initial opposition is now providing millions of dollars in property taxes and land-lease payments, sourcing steel from local vendors, and reducing land use conflict by facilitating animal grazing alongside its panels (an approach known as “agrivoltaics”).
As AI data centers proliferate across the country, local resistance has grown and will continue to grow, unless project developers recognize the importance of local engagement and local benefits. According to one recent report, local opposition has halted or delayed roughly $156 billion in potential data center investment across the United States.
AI developers have a variety of options for delivering local benefits, including those discussed above, plus much more. For example, developers can seek to hire locally, avoid large tax abatements, and support local civic institutions or events. A recent guidebook from the University of Michigan’s Center for EmPowering Communities makes additional recommendations for host communities to consider in conversations about siting data centers.
In short, seeking community input and delivering local benefits are essential when building major infrastructure, whether it’s a wind farm, oil field, or data center.
Energy and Artificial Intelligence Can Build a More Prosperous Future
Beyond providing local economic benefits, the rapid growth and sheer scale of the AI industry mean that it can deliver national economic benefits at a much larger scale. Here, we have another insight from the energy industry, specifically oil and gas.
Although most major AI companies are not yet profitable, some corporate leaders such as OpenAI CEO Sam Altman have speculated that AI companies could become so powerful as to “capture the light cone of all future value in the universe.” Whether or not this comes to pass, investors are betting hundreds of billions of dollars that AI will lead to massive profits in the years and decades to come. Nvidia, the leading manufacturer of high-end AI microchips, is the world’s most valuable company, with a market cap of over $5 trillion.
The impacts of AI, if not planned carefully, may well swamp the impacts of an energy transition.
In the context of oil and gas, historically one of the world’s most profitable industries, governments often capture some of the industry’s take to benefit citizens. The most forward-thinking governments have invested a portion of their oil and gas revenue into permanent savings funds that are invested to generate a return in perpetuity. Several US states, such as Alaska, New Mexico, North Dakota, and Wyoming, have embraced this model, as have countries like Norway and Saudi Arabia, which boast two of the largest wealth funds in the world.
If AI generates enormous profits, the federal government could apply a new “token tax” or update the tax policy for capital gains, then invest some of those revenues into a permanent fund that can provide economic benefits for generations to come. Indeed, both Anthropic and OpenAI have explicitly called for such a fund.
Establishing a wealth fund will be particularly important if AI technologies displace millions of workers. By investing in the long-term future, policymakers not only would distribute the economic benefits of AI, but also increase the probability that the public views AI as a wealth creator rather than a job destroyer.
How Planning for the Energy Transition Can Inform the Impacts of Artificial Intelligence
For years, I’ve been doing research on how a changing energy system will affect energy workers and the communities where they live. A transition to a net-zero economy, if not planned carefully, will likely result in disproportionate negative impacts for workers and communities who are dependent on coal, oil, gas, and related industries (plus lots of other challenges).
The impacts of AI, if not planned carefully, may well swamp the impacts of an energy transition. If policymakers fail to plan for the job losses associated with AI-induced automation and efficiency gains, tens of millions of workers—ranging from truck drivers to accountants to energy policy researchers—may find themselves unemployed with little or no safety net. To be clear, the extent of AI-induced job loss remains a matter of debate, but my (somewhat amateur) perspective is that this technology is moving—and could move—so much more rapidly than previous economic transitions. Workers and institutions will struggle to adapt to the rate of change, likely leading to large-scale job displacement.
Avoiding such an outcome would provide two major benefits. First and most obviously, good planning would avoid the enormous human suffering that would result from such massive job displacement. Jobs are fundamental to people’s lives, and job loss results in much more than forgone earnings.
Second, the public backlash associated with job losses on the order of the Great Depression would surely result in elections where voters reward candidates who run on anti-AI platforms. Signs of this backlash are already visible across the United States, with more than a dozen states recently considering moratoria on the construction of new data centers, along with a federal proposal to do the same.
Lessons from research on the energy transition point to a variety of policy options, including more robust wage insurance for displaced workers and their families, broader public health coverage, aligned job-training programs with growing industries, and even a universal basic income.
From Ideas to Legislation
Although research on energy and climate change has brought to light a variety of policy options that can improve societal outcomes, many of these policies are resting on bookshelves, gathering dust. From carbon prices to stronger social safety nets, the list of welfare-enhancing but unadopted policies grows longer each year at the federal level.
Will AI policy go the same way? If so, the impacts will be severe, possibly even catastrophic. AI companies have pushed for years for the federal government to intervene to reduce risk, protect national security, and prepare workers for potential job losses—not just because it’s the right thing to do, but because it’s the only way these companies can deliver the returns their investors demand.
