The Experiment Nobody Meant to Run

Humanity is changing the chemistry of the atmosphere. Not deliberately—no one set out to do this—but as the accumulated exhaust of two centuries of industrial civilization. Every furnace, engine, and cleared hectare of forest adds molecules that absorb infrared radiation. The planet retains more heat than it sheds. It warms.

A century ago, atmospheric CO₂ stood near 305 parts per million, already elevated from the pre-industrial 280 but not yet remarkable. It has now passed 425 ppm, higher than at any point in several million years, and rising by roughly 2.5 ppm annually.¹ Average surface temperature is up more than 1.3°C against the 1850–1900 baseline.

None of this is contested among people who study it. The greenhouse effect was described by Joseph Fourier in the 1820s, measured by John Tyndall in 1859, and quantified by Svante Arrhenius in 1896—who calculated that doubling CO₂ would warm the Earth by roughly 5°C, an estimate that sits inside the range modern models produce with vastly more sophisticated methods. The warming was predicted before it was observed, and then it was observed.

This chapter is the history: the science that identified the problem, the industrialization that caused it, the politics that delayed the response, and the economics that have finally begun to change the trajectory.

2026 Snapshot — Where Things Stand

The Physical Record

CO₂ concentration is approximately 425 ppm. Warming is approximately 1.3°C. Sea level has risen roughly 20 centimeters since 1900, with the rate accelerating as thermal expansion is joined by land ice loss.

Emissions run near 40 gigatons of CO₂ annually from fossil fuels, and roughly 53 gigatons of CO₂-equivalent once land use, methane, nitrous oxide, and fluorinated gases are included.⁶ The distinction matters more than it appears: methane is a far more potent warming agent over short horizons, which makes it the highest-leverage near-term target and the one most often omitted from headline figures.

Arctic sea ice extent has declined roughly 40 percent since 1980. The oceans have absorbed most of the excess heat and roughly a quarter of the emitted carbon, becoming about 30 percent more acidic in the process—an effect that proceeds independently of temperature and that no amount of sunlight management would address.

The Response

The Paris Agreement, adopted in December 2015 and in force since November 2016, has been signed by 196 parties committing to hold warming well below 2°C and preferably to 1.5°C.⁵ Its architecture is voluntary: countries set their own targets, report progress, and are expected to ratchet upward.

Current policies put the world on a trajectory toward roughly 2.5–3°C by 2100.² That figure deserves a careful reading, because it is genuinely better than it was. A decade ago, business-as-usual projections ran to 4°C or beyond. The improvement came less from diplomacy than from the collapse in solar and battery costs described in Chapters 10 and 12—the trajectory bent because clean energy got cheap, not because treaties got stronger.

Clean energy investment exceeded $1.7 trillion globally in 2023, surpassing fossil fuel investment for the first time.⁹ That crossover is the single most encouraging number in this chapter.

Notable Players

The IPCC synthesizes rather than conducts research, and its assessment reports have grown steadily more confident: from "the balance of evidence suggests" in 1995 to AR6's flat statement in 2021 that "it is unequivocal that human influence has warmed the atmosphere, ocean and land."⁴ Its process is consensus-based across governments, which means its conclusions are systematically conservative—a fact worth knowing when its projections are described as alarmist.

National research institutions—NOAA, NASA GISS, the UK Met Office, Max Planck—maintain the temperature records and climate models that everything else rests on.

The major emitters are the decisive actors. China accounts for roughly 30 percent of global CO₂ and has committed to peak before 2030 and reach net-zero by 2060.⁷ It is simultaneously the largest emitter and by a wide margin the largest manufacturer of solar panels, batteries, and electric vehicles, which makes the simple framing of China as climate obstacle inadequate. The EU has the most consistent policy and a 2050 net-zero target. The US oscillates with administrations; the Inflation Reduction Act of 2022 contained roughly $370 billion in climate provisions, the largest US climate investment ever made.⁸

Fossil fuel incumbents—ExxonMobil, Shell, BP, Saudi Aramco, Gazprom—retain enormous political influence and reserves whose value depends on those reserves being burned.

Hard-to-abate industry is where the remaining difficulty concentrates: steel, cement, and chemicals, whose emissions come substantially from process chemistry rather than energy and therefore cannot be solved by clean electricity alone.

Climate finance has grown into a real sector, with green bond issuance exceeding $500 billion annually, alongside advocacy organizations and youth movements that changed the political salience of the issue faster than any institutional actor.

The Science Story

Discovery, 1824–1958

Fourier identified that the atmosphere retains heat. Tyndall measured which gases do it. Arrhenius calculated how much, and—working by hand, in 1896—produced a sensitivity estimate that modern supercomputers have not substantially improved upon.

For sixty years this remained a curiosity, partly because nobody could establish that CO₂ was actually accumulating. Then in 1958 Charles Keeling began continuous measurement at Mauna Loa, and within a few years the record showed an unmistakable upward trend with an annual sawtooth from Northern Hemisphere plant growth.³ The Keeling Curve converted a theoretical concern into an observed fact, and it is arguably the most consequential dataset in environmental science.

Warning, 1960–1990

Scientific understanding was essentially settled by the 1970s. Internal research at several oil companies during this period reached conclusions closely matching the academic consensus, which is documented and is the basis of subsequent litigation.

James Hansen's 1988 Congressional testimony that warming had begun and was attributable to human activity made it a public issue. The IPCC was established the same year, and its first assessment report followed in 1990.

Certainty, 1990–Present

Each assessment cycle narrowed the uncertainty. AR5 in 2013–14 called human causation "extremely likely." AR6 in 2021 dropped the hedging entirely.

The most important recent development is attribution science: the ability to quantify how much more likely a specific heat wave, flood, or drought was made by warming. This changed the public conversation, because it moved climate from a statistical abstraction about 2100 to a statement about the event that happened last month—and it moved climate into court, since attributable harm is the basis of a legal claim.

The Emissions Story

Global CO₂ emissions ran about 15 gigatons annually in 1970, roughly 34 by 2010, and about 40 today. The growth curve tells the political story better than any negotiation record.

Industrialization, 1900–1970. Coal built the industrial economies of Europe and North America; oil added transport and petrochemicals. Emissions tracked GDP almost exactly, because energy was fossil energy and there was no alternative.

Globalization, 1970–2010. The oil shocks of 1973 and 1979 produced brief conservation followed by resumed growth. From the 1980s, China's industrialization added coal capacity at unprecedented scale, and manufacturing shifted from developed economies to developing ones. This shift matters for accounting: some of the apparent decarbonization of wealthy countries is offshoring, and consumption-based emissions accounting shows a considerably less flattering picture than territorial accounting does.

The present period. Coal use is declining in the US and Europe. Solar and wind are the fastest-growing energy sources but grew from a small base, which is why their rapid percentage growth took years to register in absolute emissions. Global emissions have roughly plateaued rather than declined—which is progress against the counterfactual and nowhere near sufficient against the target.

The Political Story

Two decades of climate diplomacy produced almost no emissions effect.

Rio in 1992 established the framework with no binding targets. Kyoto in 1997 set binding targets for developed countries; the US never ratified and withdrew in 2001, and the agreement excluded developing countries including China, which by then was the fastest-growing emitter. Copenhagen in 2009 came close to collapse.

Paris in 2015 succeeded by abandoning the thing that had made its predecessors fail. Instead of negotiating binding allocations—which requires agreeing who has the right to emit, a question with no achievable answer—it let each country set its own target and subjected those targets to review and public comparison. It traded enforceability for participation.

Whether that trade was correct is the central question of climate governance, and the honest answer is mixed. Paris achieved near-universal participation and has not achieved sufficient emissions reductions. The implementation gap between pledges and policies, and between policies and outcomes, remains large.

What actually changed the trajectory happened outside the negotiating rooms. Solar photovoltaic costs fell roughly 90 percent over a decade. Battery costs fell comparably. At some point in the last several years, decarbonizing electricity stopped being an economic sacrifice and started being the cheapest option, and that shift has done more for emissions than thirty years of treaties.

Modern Bottlenecks

Scale and inertia. More than $100 trillion of fossil infrastructure exists and was built to operate for decades. Replacement is limited by manufacturing capacity, capital, and the ordinary lifespan of capital equipment.

Hard-to-abate sectors. Steel, cement, and chemicals require process heat and, in cement's case, release CO₂ from the chemistry itself regardless of the fuel. Aviation and shipping need energy-dense liquid fuels for which there is no drop-in replacement at scale. Agriculture produces methane from livestock and nitrous oxide from fertilizer, neither of which has a technological fix that does not involve changing what people eat or how they farm.

Grid and permitting. In many developed countries the binding constraint on renewable deployment is no longer cost or manufacturing but transmission capacity and the time required to obtain permission to build it. Interconnection queues in the US and Europe now hold more capacity than has been built. This is a legal and administrative bottleneck standing in front of a solved technical problem.

Equity. Roughly 750 million people lack electricity entirely, and their governments reasonably prioritize access over emissions. The countries that caused the accumulated problem are not the countries that will bear most of its costs, and every negotiation since 1992 has foundered on some version of this.

The AI Role

AI enters climate work in three distinct places.

Modeling. Machine learning emulators reproduce the output of physics-based climate models orders of magnitude faster, which permits running many more scenarios and, more importantly, generating the high-resolution regional projections that adaptation planning actually requires. Global average temperature is the wrong output for a city deciding how to size a storm drain.

Emissions reduction. Grid optimization for renewable variability is covered in Chapter 10; industrial process optimization, logistics routing, and materials discovery for batteries, catalysts, and capture sorbents all contribute. Each is incremental and they compound.

Monitoring and verification. This is the most underrated. Satellite data analyzed by machine learning now detects individual methane leaks from space, tracks deforestation in near real time, and estimates facility-level emissions independently of self-reporting. Climate agreements have always suffered from unverifiable claims; independent measurement changes the negotiating dynamic more than any additional pledge would.

Second-Order Impacts

Climate risk entered finance. Insurers have begun withdrawing from markets in Florida, California, and elsewhere where expected losses no longer support pricing. This is climate change arriving through the mortgage market rather than through weather, and it reprices property faster than any carbon tax would.

Attribution created legal exposure. The ability to quantify a specific harm's climate contribution, combined with documented early industry knowledge, has produced a growing body of litigation. This may end up mattering more than regulation.

Energy security and climate policy converged. The disruption of European gas supply after 2022 accelerated renewable deployment for reasons that had nothing to do with emissions, which suggests that framing decarbonization as security policy works better in some political environments than framing it as climate policy.

Migration pressure is building ahead of the models. Habitability declines gradually and then produces movement suddenly, and the institutions governing displacement do not recognize climate as a basis for protected status.

Conclusion

The physics has been understood for a century and a half. The measurements have been continuous for nearly seventy years. The predictions were made in advance and have been confirmed. There is no serious scientific dispute about the cause.

What was disputed was the response, and the disputes cost roughly thirty years.

The situation in 2026 is genuinely better than the situation in 2015, and the reason is instructive. It was not that international negotiation finally worked; Paris achieved participation by giving up on enforcement. It was that solar, wind, and battery costs fell far enough that the low-carbon option became the cheap option, at which point deployment stopped requiring anyone's virtue.

That is the mechanism worth understanding, because it is the one that generalizes. Emissions have fallen where clean alternatives became cheaper, and have not fallen where they did not. The remaining hard problems—cement chemistry, aviation fuel, agricultural methane, and the transmission lines nobody will permit—are exactly the places where no cheaper alternative yet exists.

Current trajectories point to 2.5–3°C. That is a bad outcome, considerably better than what was projected a decade ago, and considerably worse than what was promised. Whether it improves depends on whether the cost declines that transformed electricity can be repeated in the sectors that have so far resisted them.

One caution belongs alongside that number. Trajectory estimates assume the climate system responds smoothly to added forcing, and several components may not—ice sheets, permafrost carbon, Amazon dieback, and the Atlantic overturning circulation all have thresholds past which change becomes self-sustaining.¹⁰ Where those thresholds sit is poorly constrained, which is frequently presented as a reason for calm and is in fact the opposite: an uncertain threshold is one that cannot be approached carefully.

The next chapter takes up what happens if they cannot: removing carbon that has already been emitted, reflecting sunlight, and adapting to warming that is now unavoidable regardless of what is decided from here.

Endnotes — Chapter 55

1. CO₂ concentration measured at Mauna Loa Observatory; pre-industrial approximately 280 ppm, passing 420 ppm in 2023 and approximately 425 ppm currently, rising about 2.5 ppm annually.
2. Climate Action Tracker and the UNEP Emissions Gap Report estimate that current policies lead to approximately 2.5–3°C warming by 2100—an improvement over the 4°C+ trajectories projected a decade earlier.
3. The Keeling Curve: continuous CO₂ measurement at Mauna Loa initiated by Charles Keeling in 1958, showing sustained increase with an annual cycle driven by Northern Hemisphere vegetation.
4. IPCC AR6 Working Group I (2021): "It is unequivocal that human influence has warmed the atmosphere, ocean and land." Earlier assessments used progressively weaker formulations as confidence built.
5. The Paris Agreement was adopted in December 2015 and entered into force in November 2016; 196 parties have signed. Its architecture substitutes nationally determined contributions and periodic review for binding allocations.
6. Global CO₂ emissions from fossil fuels are approximately 40 Gt annually; total greenhouse gas emissions are approximately 53 Gt CO₂-equivalent including land use, methane, nitrous oxide, and fluorinated gases.
7. China accounts for roughly 30 percent of global CO₂ emissions and has committed to peak before 2030 and reach net-zero by 2060. It is also the dominant manufacturer of solar panels, batteries, and electric vehicles.
8. The US Inflation Reduction Act (2022) contains approximately $370 billion in climate-related provisions—the largest US climate investment to date, structured primarily as tax credits rather than regulation.
9. Clean energy investment exceeded $1.7 trillion globally in 2023 according to the IEA, surpassing fossil fuel investment for the first time.
10. Identified potential tipping points include the West Antarctic and Greenland ice sheets, Amazon rainforest dieback, permafrost carbon release, and weakening of the Atlantic meridional overturning circulation. Thresholds are poorly constrained, which is itself the argument for caution.