我们曾拯救过世界：臭氧洞对AI的启示

It might destroy the world, despite passing every known safety test. If we wait for a “warning shot” before we act, it might be too late. And action requires global coordination, because if anyone makes it, everyone dies. Sound familiar?It should, because it already happened half a century ago, with chlorofluorocarbons (CFCs). Despite seemingly impossible odds, we got our act together and completely solved the problem through unprecedentedly successful international coordination. The Montreal Protocol banning CFCs, signed 39 years ago today, is the only treaty that has ever been ratified by every single country in the entire world.Total Montreal protocol victoryMaking AI go well is going to be a lot harder than fixing the ozone hole. Nonetheless, the similarity is uncanny, and we don’t have any other choice. Understanding how we did the impossible once before may teach us something about how to do it again.The theory is bornThe year is 1973. The slow televised unraveling of the Nixon administration is already well underway. DDT finally got banned last year by the newly created EPA. A river got so polluted that it literally caught on fire.The Cuyahoga River FireEnvironmentalism looms large in the public consciousness, and trust in the government’s competence is at a nadir. There are some people explicitly trying to anticipate catastrophic environmental risks, but beyond the known quantities like DDT, it’s hard to predict much more than the abstract possibility of delayed effects and catastrophic threshold effects from as-yet unknown pollutants.[1]Scientific discovery is often serendipitous. At an obscure conference, during a coffee break (the most important part of any conference), a Professor F. Sherwood Rowland of UC Irvine would by chance learn about some intriguing new findings from a meteorologist, who found evidence that these gases called CFC were accumulating in the atmosphere. Something about this direction piqued Rowland’s interest, and he filed it away in his mind.After this direction collects dust for about a year, he assigns it to one of his postdocs, Mario Molina. Molina works through the theory of what happens to CFCs in the atmosphere, not expecting to find anything super interesting. Instead, he discovers something alarming: the chlorine from the CFCs will break off in the upper atmosphere, and then catalyze reactions that destroy large amounts of ozone.Without the ozone layer, nothing prevents intense UV rays from sterilizing everything under the sun.[2] Vast swathes of vegetation would be obliterated, walking around outside for too long would become a death sentence, and billions would die.After checking and rechecking the results with Rowland, the result is undeniable. Something needed to be done, and quick.“The work is going well,” [Rowland] told his wife, Joan. “But it looks like the end of the world.”The world reactsDespite an initial media reaction comparable in volume to the average cricket, their persistence pays off, and their theory eventually catches the attention of the nation. Industry lashes back, describing the new theory as ridiculous sci-fi theorizing. CFCs had already established an illustrious (not to mention, lucrative) decades-long career of being inert nontoxic chemicals. Why throw them out because of some theory? Every safety eval had shown that CFCs were far safer than their decidedly non-inert predecessors, which would happily burst into flames, or non-theoretically rearrange the insides of the lungs of unlucky families with leaky refrigerators.State governments became the first to regulate this new danger. Within a year, Oregon enacted a ban on CFCs in aerosols—[3]comprising a small fraction of all CFC usage, to be fair, but substantial nonetheless. Other states followed quickly. Industry officials viewed these bills with distrust. After all, how could a patchwork of local regulations ever solve a global issue?More effective than these early state regulations, however, was consumer choice. Despite its rocky start, public awareness of the CFC issue kept mounting. Concerned members of the public launch a boycott of spray cans and other CFC-containing products, putting a huge dent in sales many years before any federal or international regulations were introduced.This decimation of the CFC market due to boycotts was one of the key enablers of later regulation—it’s easier to ban something that’s already on the way out than something that’s rapidly growing.CFC sales per year (Solomon, 2024). The hole left by the Molina-Rowland paper is clearly visible. Just think, we could have been exploring ultra skin cancer by now.After successful state legislation, the next step was to go federal. Many scientific debates, some political finagling, and one report from a National Academy of Sciences panel later, the EPA finally announced a ban of some “nonessential” uses of CFCs in aerosols, such as spray cans. While this didn’t solve the core problem—the “essential” uses were going to be a lot harder to replace—it was a big step in the right direction.Clearly, the only reasonable next move would be to expand federal regulations to cover all CFC uses, and then to propose and ratify an international agreement, solving the problem once and for all, and then everyone stood up and clapped. That treaty negotiator’s name? Albert Einstein.Just kidding! After the 1978 federal ban on nonessential aerosol CFCs, the field of CFC notkilleveryoneism enters what one contemporary advocate for regulations refers to ominously as The Dark Years.The dark yearsAfter these early wins, the next 8 years were a rough time for ozone safety advocates. A frustrating stalemate ensued: The government wasn’t willing to fully ban such a big industry without airtight evidence, but there was no empirical evidence that the ozone layer was diminishing.The researchers continued to insist that the potential harm was so serious that prudence demanded immediate action even just on theory. There could be a decades-long delay between CFC emission and ozone layer damage, so we might already be past the point of no return by then.Caught in this impossible dilemma, the only thing the researchers could do was to further refine the theory. The original Rowland-Molina theory was simple: the chlorine catalyzes the decomposition of ozone into oxygen. As with all simple theories, this one is useful but incomplete.Every few months, someone would discover a new chemical reaction relevant to ozone depletion, or run a more detailed computer simulation, and the ozone depletion estimates would swing around wildly. Rowland and Molina themselves were even responsible for discovering a reaction that made the models suggest bizarrely that CFCs would increase ozone. Despite the inevitable weaponization this data would find in the hands of DuPont, they decided out of a sense of scientific honor that it would be wrong to withhold publishing it. Although later research suggested these reactions were not enough to exculpate CFCs, their commitment to scientific integrity earned Rowland and Molina a good deal of respect, even from those who disagreed.These mercurial predictions did not exactly earn the trust of policymakers. Even though chemistry is a very mature scientific field, it turns out that predicting things from theory is just really hard. It didn’t help that a lot of ozone research came from industry trying to muddy the waters, though even the industry researchers themselves often weren’t so confident that all was going to be well.Industry scientists are used to letting the PR people talk for them. But privately: “But there were nights, over drinks in hotel bars, when industry scientists seemed as concerned as anyone.” “Look, I’m a person. I have children. I want to have grandchildren. We’re interested in the future, too.”With the dawn of the ‘80s and the anti-regulation Reagan administration, the possibility of anyone ever doing anything felt more and more distant with every day.A