What climate change politics can teach us about the AI apocalypse debate: Philip Bump
Barack Obama came to the White House in 2009 with a sweeping mandate to overhaul American politics. The country was battling a deep recession and mired in war; Obama was elected to shift directions.
One way in which he promised to do so was by building out the capacity of industries centered on climate change. It aimed at killing three or four birds with one stone: get Americans back to work, address an environmental crisis that had gained national attention a few years prior — thanks to a blockbuster Hollywood movie — and beat foreign competitors to what promised to be a lucrative market.
But then the politics of climate change shifted. Republican legislators went from joining bipartisan calls for action to dismissing warming temperatures as a hoax or overstated. Obama’s “green jobs czar” resigned in the face of pressure from right-wing media. A federal fund that invested in green energy companies was lambasted for one company’s failure, though the fund ended up earning a profit. Industries that faced pressure from efforts to reduce carbon dioxide and other emissions, like the oil and gas companies that thrived under Obama’s predecessor, poured millions into federal campaigns.
Advocates insisted that the U.S. couldn’t afford for its industries to be beaten by China, even as activists warned that an existential threat was looming. Billions of dollars were at stake. Congress settled for inaction.
Perhaps this sounds familiar.
In the past few weeks, the nation’s attention has been gripped by a new threat of annihilation, one that also feels as though it was plucked from a movie script. Jacob Coxon, a former employee of two of the leading artificial intelligence companies, OpenAI and Anthropic, stated publicly that the “people building AI earnestly believe that it could kill us all by the end of the decade.” Landing shortly after news that AI agents (that is, autonomous AI actors) had taken over elements of a German website without initial detection, Coxon’s warning gained international attention.
The current president’s response? All that’s needed to prevent AI development from triggering the end times is “a STRONG AND SMART (High IQ!) PRESIDENT” which, he boasted, America has “in spades.” Besides, he said while in Ireland to attend a golf tournament at one of his privately owned properties, he wanted to stay ahead of China on development “because whoever wins AI wins.”
Before we go too far here, it’s worth admitting that the parallels between the push for action on climate change and on AI are imperfect, with arrows pointing at times in opposite directions. It is also worth adjudicating just how seriously each threat of apocalyptic outcomes should be taken.
To understand the threat posed by AI, I reached out to John Herrman, a writer at New York magazine whose instincts on and understanding of technology I’ve long trusted. He noted that AI advocates and employees of AI firms often come into the discussion with a belief that AI could trigger some sort of collapse, a belief stemming from decades-old works like I.J. Good’s 1965 essay about a self-improving, super-intelligent machine. Meaning that the concerns often preceded exposure to the actual tools.
Few of the apocalyptic scenarios that are discussed are apocalyptic in the end-all-human-life sense. An autonomous AI taking control of a weapons manufacturing facility, for example. An AI siphoning off electricity from the grid to ensure its own survival. But some are, like an AI developing and deploying a virus that could infect and kill tens of millions of people.
Your response to these theoretical threats probably mirrors mine: there’s a significant gap between “an AI theorizes a deadly virus” and “that virus is now aboard your red-eye flight.” The director of a biolab recently made this exact point, in fact, writing on social media that “[t]here are many physical checkpoints you can implement easily that prevent dangerous biological work from happening.” And even if it did happen, even if an AI, say, used social engineering to mimic a biolab worker’s boss on a phone call, a lot of other dominoes would have to fall in very particular ways for this to occur. Any vague worst-case scenario that looms, we can think of a way in which it can stumble.
But this, Herrman pointed out, is the problem. We’re limited to what we can think of.
If artificial intelligence truly begins to outpace human cognition, either at the urging of AI companies or by establishing itself outside of human control (which also is harder than it might seem), it may think of things we wouldn’t. (Well, “think,” anyway. As someone who majored in philosophy in college, I recognize the temptation to adjudicate the nature of what’s happening but, as the behaviorists might argue, the difference between actual superhuman thought and simulated superhuman thought is probably unimportant.)
What we can’t predict is the unknown unknown here — which is both possibly true and begging the question. “AI will kill us all because it will think of novel ways to kill us all” is a tautology but not necessarily an incorrect one.
Summarizing one of Coxon’s points, Herrman used another Hollywood analogy to explain the long-tail risk.
“Imagine yourself as a monkey,” he said. “And that’s it — that’s sort of the beginning and end of the argument. We’ve all seen ‘Planet of the Apes’; we know deep down that the monkeys are never going to take over, and we also know that our attitude toward them is, best case scenario, making a sanctuary. And that’s it.”
The inevitable follow-up question is why anyone would keep building AI, given this risk they see as possible. Herrman analogized the situation to the nuclear proliferation of the Cold War, multiple sides understanding the risk of what they’re doing but seeing their own development as essential for counterbalancing that risk. In 2017, Anthropic co-founder Dario Amodei said that “although AI has a number of dangers, actually I think, you know, if we never built AI, if we don’t build AI for 100 years or 200 years, I’m very worried about whether civilization will actually survive.”
It’s much easier to describe the risks of climate change, though scenarios for extreme warming also bleed into the unpredictable. More massive storms due to the enormous amounts of water that can be stored in warmer air. Coastal flooding due to melting glaciers and oceans that expand as they heat up. Extensive droughts as heat sucks water from the land. Shifts in habitation, food availability and weather patterns.
Here, too, humans will survive, but at significant cost. Unlike the AI threats, though, climate risks have already begun to manifest. We are already much further down the road toward climate disaster than we are to AI apocalypse.
In each case, though, we have the sober words of President Donald Trump assuring us that these concerns are fake, a hoax. He’s infamous for dismissing climate change, of course. Just this week, he announced that he was “breaking another hoax,” the idea that “AI is going to take over, consume, and destroy the World.” We must push forward on AI, he has said, in order to stay one step ahead of China.
In both cases — climate and AI — Trump is on the side of the moneyed interests that want to push forward despite the risks. One significant difference is that those raising concerns about those risks include, in the AI case, the moneyed interests themselves. Another significant difference is that China served as a foil for combating climate change 20 years ago and now serves as a foil for racing forward on artificial intelligence — out of concern that China will achieve the riskiest toolset first.
The moneyed interests won the climate fight. Arguments prioritizing short-term economic benefits over distant threats won out. America pulled back on climate investment (save a surge during the Biden administration) and China came to dominate clean-energy manufacturing. In a grating turn for those who’ve been tracking the issue over time, Trump and his allies used the fact that so much of the infrastructure of solar and wind energy originates overseas as an argument against investing in those systems.
So why are we pushing forward on AI, despite talk of distant threats? In part because those threats are much less obviously concrete than future flooding and droughts. In part because the short-term and medium-term economic benefits are much more concrete. Mostly because it has the moneyed interests on its side.
Besides, the hand-wavy who-knows-what-AI-will-do argument works the other direction, too. Maybe AI will figure out how to block a rogue AI. Maybe, despite the electricity AI systems guzzle, those systems will defeat climate change.
You know those thought experiments that are like Would you take $1 billion now in return for a 1% chance you’ll be hit by a train tomorrow? Just think of modern society as one of those, except that you’re not getting the $1 billion.
This article originally published at What climate change politics can teach us about the AI apocalypse debate: Philip Bump.