And little wonder: Everyone wants what they have. Investors, policymakers, and economists believe that America’s ongoing economic growth and competitiveness depend on ample access to this company’s products. The sitting Republican president has staked his administration on making sure Americans can get as much of it as they want — regulations be damned.
But there is a problem. One of the company’s researchers has become convinced that the company’s products are dangerous — so harmful, in fact, that their continued use and growth trajectory portends catastrophic risk for humanity. He attempts to alert the company’s executives to this fact. What happens next?
Perhaps you know the story. In the late 1970s and early 1980s, Exxon’s internal scientists concluded that the ongoing growth of fossil fuels would raise global temperatures and have “potentially catastrophic” effects on the planet’s climate. They presented these results to Exxon’s executives. A senior scientist warned that humanity had a brief window — “five to 10 years” — before “the need for hard choices regarding changes in energy strategies might become critical.”
Exxon led a large research effort into climate change, affirming its scientific validity. But then in the late 1980s, its CEO decided to go in the other direction. Its executives chose not to warn the public about climate change — and instead began a successful disinformation campaign meant to convince the public that climate change was not settled science.
But what if things had gone differently? We’re getting a taste of that pathway now. Last week, Sam Coxon, a researcher at the artificial intelligence company Anthropic, resigned because he feared the AI industry was too close to building an “out of control” intelligence. He quit his job just a few months before his corporate equity would have vested, giving up what would have likely been life-changing wealth to warn about what he believes to be existential risks. Humanity only had a brief period of time — perhaps a year — to steer the technology to a better path, he said.
Anthropic researchers who remain at the company affirmed his analysis. “We really do earnestly believe AI could kill all humans,” a senior scientist at the company posted on the social network X.
But this time, Anthropic’s CEO, Dario Amodei, did not respond as Exxon’s leadership did three decades ago. Instead, Amodei basically agreed with Coxon: He asked for the government to regulate artificial intelligence and “pace the frontier,” meaning that it should enforce a slower rate of cutting-edge artificial intelligence development.
I’ve thought of these two examples over the past few days as I’ve tried to make sense of the surge in public concern about AI and existential risk.
It seems to me that climate change is looming over the AI conversation and shaping the assumptions, outlook, and behavior of many key players and observers. President Trump, of course, is reading from the old playbook and has deemed AI to be a “hoax”; Coxon, appearing on Fox News, has downplayed climate change’s existential risk as compared to runaway AI. Yet even beyond those reruns and revisions, the analogy goes deeper: Just as nuclear non-proliferation agreements structured early attempts to regulate global greenhouse emissions, climate policy is now shaping how people understand AI risk.
And not for lack of cause. In some important ways, the problems — or alleged problems, depending on your perspective on AI — resemble each other. For instance, because technology exists in a global commons, any successful AI diplomacy must involve the United States and China. And since China’s AI development currently lags the United States, American politicians must persuade China that their proposals to regulate AI are not just concealed attempts to restrain China’s development.
This dynamic has long bedeviled climate negotiations, too. Since economic growth has (until very recently) required fossil fuels, China and other middle-income countries have long feared that any global climate treaty would constrain their future economic development. The Kyoto Protocol tried to finesse this problem by splitting countries into two groups, rich and not-rich; the Paris Agreement did it by imposing no collective restrictions on fossil fuel consumption at all.
Neither approach has worked, exactly, but each offer examples, counterexamples, and tools for thought. Perhaps the Montreal Protocol, which has successfully limited global production of the pollutants destroying stratospheric ozone — and has shown how to stop the growth of a dangerous but hard-to-manufacture technology that presents near-term existential risk — is a superior model.
There is at least one big way the two risks differ. Climate change is a chemical problem that arises from the size and scale of global fossil fuel consumption. Scientists have known that the greenhouse effect is real since the early 20th century. Climate change’s physics are rudimentary enough that Exxon’s in-house scientists could predict the path of future warming with some accuracy. It is a verifiable risk.
AI’s alleged existential risks, on the other hand, emerge from a lab pushing the technological frontier too far and drilling, like Tolkien’s dwarves, too deep. AI concern relies not on empirical observations, but on a story about exponential change and runaway growth. In this way, it’s a harder risk to predict, and a harder one to accept.
Climate advocates have long wondered what would have happened if Exxon’s leaders had embraced reality and warned the public in the 1980s that global warming is real and caused by fossil fuels. Inside Climate News once called it a “road not taken.” I can’t help but wonder if we’re watching it.
The Smith Mountain Project, a 600-megawatt pumped storage facility in Virginia. Education Images/Universal Images Group via Getty Images