SerpApi.Org
In 2019, the AI researcher Rich Sutton published a short essay called “The Bitter Lesson.” Seventy years of AI research, he argued, kept demonstrating the same thing: the most effective ways to train AI rely on raw, brute-force computing power — simply scaling up — rather than clever approaches built on human insight. It didn’t matter how elegant researchers’ work was. More compute would win every time. It’s a bitter lesson because, well, no one wants to accept that even the best efforts can go unrewarded.
This summer, climate change has been teaching its own kind of bitter lesson. Last week smoke from out-of-control Canadian wildfires reached more than 120 million Americans, sending out air quality alerts across 18-plus states from Minnesota to Virginia. Those wildfires, which were made more likely by the effects of climate change, followed record-breaking heat waves in Europe that contributed to the deaths of thousands of people.
Disasters like these inevitably renew calls for climate action, but no current US emissions policy, no amount of carbon cutting now, could clear the air, just as nothing Europe has done on climate — and it has done more than any other region, at no small cost — could protect it from those killer temperatures. Not this summer, not this decade, not the one after. The steps countries take now to reduce carbon emissions — to mitigate climate change — will not protect them from the present-day effects of climate change.
That’s the bitter lesson of climate change, and 2026 is the summer it has become unignorable. The dangerous effects we’ve been warned about are here, locked in on every timescale that matters to the people living through them. We can’t prevent the worst from happening right now, so we need to figure out how we’re going to adapt to it.
The smoke we can’t stop
Canadian wildfire smoke is as close to an unstoppable disaster as exists today. The fires are Canada’s, the ignition is mostly lightning, the fuel is a subcontinent of boreal forest no fire agency can manage, and the whole system sits outside US jurisdiction. While strategies like prescribed burning and forest thinning can help reduce the wildfires that crop up in the western US, the sheer scale of Canadian forests makes such management far more difficult, maybe impossible. There is no timescale on which American mitigation clears American air of Canadian wildfire smoke.
What this means is that for this wildfire smoke, at this moment, adaptation may be all there is.
American law has already conceded the point. Under the Clean Air Act’s Exceptional Events Rule, a state can petition the EPA to strike wildfire-smoke days from the record used to judge whether its air is legally clean. In other words, the official response to one of the country’s fastest-growing air-quality threats is to delete it from the ledger.
It’s true that climate attribution science ties heat waves to warming with the highest confidence of any extreme, while wildfire is messier — a compound event filtered through ignition, fuel, and fire weather. But fire seasons are lengthening, fuels are drying, and the most extreme wildfire events on Earth more than doubled between 2003 and 2023. The pattern is intensifying fire in particular regions — the boreal north among them — and the smoke goes where the wind says.
And that smoke is worse than it looks. Wildfire PM2.5 — the microscopic particles that are particularly dangerous — is roughly three to four times more toxic to the lungs, per unit of mass, than ordinary urban particulate matter. During major smoke events, pediatric asthma visits can spike by more than 30 percent, and Harvard researchers have found the cardiorespiratory damage persists for months after the sky turns blue again. That’s especially dangerous for vulnerable populations like babies and pregnant women.
We can’t affect the winds, we struggle to stop the fires, and any current attempt to reduce warming enough to make those infernos less likely won’t be felt for years. But that doesn’t mean we’re helpless to protect ourselves from wildfire smoke, any more than we are from extreme heat.
We can purify indoor air, and it costs less than you probably think. Strap four furnace filters to a box fan — creating a Corsi-Rosenthal box — and indoor PM2.5 drops by roughly half; in testing during actual wildfire smoke, a DIY unit cut fine particles by about 56 percent in a large room and close to 99 percent in a small sealed one.
For heat, the toolkit is even better known — and one of the best proofs it works is in, of all places, France. The historic 2003 heat wave killed some 15,000 people there, most of them elderly and alone. The national heat plan built in its aftermath — tiered alerts, registries of vulnerable residents, check-in calls, cool rooms — cut the death toll of comparable heat waves by roughly 90 percent: when France hit its all-time record of 114.8°F in 2019, fewer than 1,500 died, per the Christian Science Monitor. A recent European analysis estimates a rerun of 2003 would now kill 77 percent fewer people in France than in a world that never adapted.
So technologies exist to protect ourselves. What’s too often missing are the standards, money, and the drive to deploy protection before the disaster arrives instead of after.
Catching up to today
When it comes to wildfire smoke, some governments are ahead of the game — and some aren’t. Washington state, Oregon, and California now regulate outdoor workers’ smoke exposure by air-quality index, under rules like Cal/OSHA’s wildfire-smoke standard; California funds “clean air centers” with upgraded filtration where people can shelter on bad-air days; Colorado’s Clean Air for Schools program ships free HEPA units to K-12 classrooms. A purifier humming in a classroom is essentially doing for smoke what a cooling center does for heat.
When it comes to heat, some parts of Europe are catching up. Barcelona now runs a network of nearly 400 “climate shelters” — libraries, museums, retrofitted schoolyards — with the goal of putting one within a 10-minute walk of every resident. For the first time the city is keeping libraries open through August to do it, which, as anyone who has visited Europe in the late summer knows, is no small thing.
So adaptation exists, and it is saving lives. But this summer has also shown what happens when cities, states, and countries aren’t ready.
In the US, the Midwestern and Northeastern states that spent last week under smoke alerts have almost none of the protections against wildfire smoke that have become more common in Western states, and no one is required to build it, because the country has no enforceable national indoor air-quality standard at all. In Europe, national heat plans are still reluctant to reach for the most powerful tool on the shelf. About one European household in five has air conditioning, compared to nearly nine in ten in the US, in large part because much of the continent still treats the machine as a moral defect.
In both cases, rich regions that have largely taken climate mitigation more seriously than some of their neighbors aren’t pursuing adaptations they need. Which really means not taking climate change as seriously as they should.
Accepting the bitter lesson
For a generation, adaptation was the word climate advocates were often reluctant to say out loud. In his 1992 book Earth in the Balance, Al Gore called adaptation “a kind of laziness, an arrogant faith in our ability to react in time to save our skins.” (He later changed his mind.) And in fairness, “we’ll just adapt” has often been a delaying tactic, an easy out for polluters.
But the ground has shifted under that caution, because the projections did exactly what projections are supposed to do: they came true.
Longer fire seasons, deadlier heat waves, smoke reaching places that never used to see it — this summer is the confirmation. And that changes what seriousness requires. If you’ve spent decades warning people that the effects of climate change are coming, you can’t treat readying for those effects as a distraction from the real work. None of this argues for easing up on emissions — each ton that isn’t emitted is adaptation nobody has to build. But preparing for what you predicted is what believing your own prediction looks like.
For AI researchers, Rich Sutton’s lesson was bitter for a specific reason. Those who resisted its conclusion weren’t foolish; they resisted it because it seemed to undercut all their hard work. They resisted it because it didn’t seem fair, just as it’s not fair that countries and states that have in many ways done much to mitigate climate change find themselves just as vulnerable, if not more so, than peers that have done comparatively little.
But here’s the other half of Sutton’s lesson: the researchers who accepted it won. They stopped defending their cleverness and built what worked. That option is open to climate politics too. The atmosphere isn’t fair, but it is consistent, and it will deliver next summer’s version of this one on schedule. We can be ready. The bitter lesson is only bitter until you have the will to act on it.
A version of this story originally appeared in the Future Perfect newsletter. Sign up here!
