AI panic is the new climate hysteria — and it’s handing the future to China

There’s a saying that although history might not repeat itself, it does rhyme. Such is the case with artificial intelligence and climate change. In today’s communications-inundated world, not only does public reaction to global trends evolve at warp speed, but the boundaries between misinformation, disinformation, opinion, and facts blur.The current result is panic and the whipsawing of public opinion about AI by the day. A new generation of anti-progress Luddites has been born almost overnight. Climate hysteria took hold much more slowly by comparison. This contrast is striking.The spark is typically an alluring doomsayer. Teenage activist Greta Thunberg declared our planet was in imminent peril. Blogger Evan Hubinger, echoing Anthropic alumnus Jacob Coxon, speculated that there is “greater than a 10% chance” AI will kill all humans in a decade. The common thread is how a threat tantalizingly rings true to people new to the issue and its nuances. The predictable result is rash, ineffective, and counterproductive reactions, even if the underlying concerns are valid.This happens in part because crisis sells, and attention-seeking media and pundits pounce. Definitive-sounding “evidence” appears. In the case of climate, it was a scenario designated RCP8.5, a greenhouse gas prediction from the Intergovernmental Panel on Climate Change. It has since been debunked as unrealistic.The AI model that “went rogue” and attacked a prominent startup, Hugging Face, is the most frequently cited evidence that the AI apocalypse is nigh. AI learning from experience, rather than coders, supposedly means that humans have lost control. Fears of models improving themselves (dubbed Recursive Self-Improvement) have rapidly piled on. Like RCP8.5, this is a worst-case scenario masquerading as a certainty.Largely ignored in the reporting was that the test environment (“sandbox”) was inadequately fortified. Fewer still noted that this was an important learning experience or that self-learning and RSI models will be essential for identifying bad actors misusing AI.There are valid reasons to fear the unknowns of AI, to be sure. As a tool, AI can be misused to create malicious software, bioweapons, more innovative scams, and more. Even more reason to find ways for models to police themselves using AI.Further, jumping headlong into using AI agents that operate autonomously magnifies risk. The agent cart has come before the internal controls horse. People forget that today’s large language models work by predicting the most statistically likely output based on patterns in their training data, not by truly understanding a situation, so they can make incorrect predictions leading to destructive actions. That is why vendors making AI-powered physical devices incorporate extensive hard rules. These rules are specific to the situation, not generic. A Waymo taxi does not need grammar correction.An analogous lack of understanding of the Earth’s ecosystem led to the many costly climate policies that had little to no impact as the political class grandstanded. Expensive pseudo-remedies were implemented to be seen to be doing something. Effectiveness was beside the point.For AI, the political catnip is regulation of the AI industry and the demonization of data centers. Neither makes sense.No government or industry regulator could possibly keep up with AI advances without allowing China to rapidly leapfrog the United States on this critical technology. We cannot stop the development of AI. It is too decentralized, so we must instead focus on ways to channel it.The flap over data centers is vapid. Hyper-scalers (the big builders) know that they need off-grid power, whether natural gas now or small-scale nuclear soon. If water is a scarce resource at a proposed location, a closed-loop cooling system can minimize usage to the point of irrelevance.With any poorly framed crisis also comes a misplaced opportunity. “Big Wind” and “Big Solar” exploited the climate panic. They consumed vast resources and subsidies that might have been better allocated elsewhere, such as to carbon capture and small modular nuclear plants. Electric vehicles received massive subsidies even though they have little emissions benefit when battery manufacturing is accounted for.The prominent AI opportunists are OpenAI and Anthropic, who view regulation as ensuring their duopoly for the most advanced models. Open source — models whose inner workings are published and available — is likely to thwart this advantage anyway because it is customized in private and mostly beyond the reach of normal regulation.There remains an important legacy of defective climate policy that impacts AI. In the absence of limits on building coal-fired power plants, unlike in the West, China now has more than double the total energy-generating capacity of the U.S., and the gap is widening. Energy may provide its crucial competitive edge in meeting AI demand.THE ANTI-AI REBELLION JUST FOUND ITS FIRST REAL TARGET — AND IT HAS A STREET ADDRESSIf the anti-AI panic consumes us, we will lose one of our critical global assets — technology leadership. We would repeat, at our peril, the fiasco of futile climate programs. Our adversaries are already exploiting this. Their online propaganda, magnified by our own ill-informed “useful idiots,” is pushing us toward the wrong responses to AI dangers. Sen. Bernie Sanders (I-VT) has even proposed legislation that would enable the government to criminalize the production of data processors used to build or deploy superintelligent AI.Can’t we learn from past mistakes?Andrew I. Fillat spent his career in technology venture capital and information technology companies. He is also the co-inventor of relational databases. Henry I. Miller, a physician and molecular biologist, is the Glenn Swogger Distinguished Scholar at the Science Literacy Project. They were undergraduates together at M.I.T.