Growth or the Planet? The Record Method Tried on Global Warming

Record note, as the method requires in its first line: every model used here was made by one company, Anthropic. The method asks for a mediator from a different maker than the models that drafted the cases. That was not possible here.

This is a spin-off from Three Witnesses. The previous article asked whether AI can help resolve disputes and proposed a procedure, the record method. Here the method is tried on a live disagreement: global warming, as argued between people who put economic growth first and people who put the environment first. Both cases, the testimony and the mediation are written by AI models, and each part names the model that wrote it. A human chose the question, the two parties and the format.

On 9 November 2026 the United Nations climate conference, COP31, opens in Antalya. Two weeks earlier the UN Environment Programme reported that policies now in place lead to about 2.8 degrees of warming this century, and that full delivery of every national pledge would bring that to 2.6. The Paris Agreement's goal is to stay well below 2 and to aim for 1.5.

The method begins by stating each side's case in the form that side would sign. That is where this trial first falls short: nobody signed these. They are an AI's reconstruction from the public record, written in one round.

The growth case

Written by Claude Fable 5.1 (Anthropic). This is an AI's reconstruction of the case made by people who hold that economic growth must come first. No person or organisation has approved it.

Past

For most of history almost everyone was poor. In 1990 about 2.3 billion people lived in extreme poverty. The World Bank's count for 2026 is 811 million, on a line of three dollars a day. Nothing else humans have tried has done as much good as fast as the growth behind that fall, and it ran on cheap energy, most of it fossil. No country has become rich without it.

We have been told before that growth must stop. The Limits to Growth said so in 1972. In the fifty years since, incomes, harvests and life expectancy rose across most of the world.

Present

About 655 million people still have no electricity, and about 2 billion cook on open fires or dirty stoves. An American emits around 14 tonnes of carbon dioxide a year, an Indian 2.2, an African 1.0, and a person in a low-income country 0.28. Telling the last three groups to hold back is asking those who did least to pay most.

Growth is also what is solving the problem. Between 2010 and 2024 the cost of solar electricity fell by 90 percent and of battery storage by 93 percent. Those falls came from rich markets buying at scale. In 2025 renewables were 85.6 percent of all new power capacity. A 2021 study found that 32 of 116 countries had cut their emissions while their economies grew.

Wealth protects. A household with air conditioning, a sea wall and insurance survives a heatwave or a storm that ruins a poor one. Adaptation is bought with income.

And the economics do not support panic. William Nordhaus won the Nobel prize in 2018 for weighing the costs of warming against the costs of preventing it. The 2023 version of his model puts the best balance at about 2.6 degrees, not 1.5.

Future

Keep growing, put a price on carbon, and spend heavily on the technologies that are not yet cheap: storage, nuclear, clean steel and cement, carbon removal. Make clean energy the cheapest option everywhere and the poor will choose it without being told to.

What we fear is a climate policy that feels like sacrifice. It loses elections. The United States left the Paris Agreement, with effect from 27 January 2026. A policy that voters reject protects no one.

The environmentalist case

Written by Claude Fable 5.1 (Anthropic). This is an AI's reconstruction of the case made by people who hold that the limits of the planet must come first. No person or organisation has approved it.

Past

The warning is old and it was right. Scientists set out the mechanism in the nineteenth century and the danger by the 1970s. Governments have met almost every year since 1995 to deal with it. Emissions from fossil fuels reached a record 38.1 billion tonnes of carbon dioxide in 2025. The final decision of last year's conference in Belém did not mention fossil fuels.

Growth was real. Part of its bill was sent to the atmosphere, and to people who never received the growth.

Present

The world was 1.1 degrees warmer in 2011 to 2020 than before industry. 2024 was the first calendar year above 1.5 degrees, at 1.60. 2025 came in at 1.47, and August 2026 at 1.65. At the start of this year the amount of carbon dioxide that could still be emitted for a fair chance of holding to 1.5 was about 170 billion tonnes, roughly four years at the present rate.

People are dying now. The Lancet Countdown reported a record 710,000 heat-related deaths in 2023 and attributed about 70 percent of them to climate change. Its previous edition counted 2.52 million deaths in 2022 from the air pollution of burning fossil fuels.

Yes, some countries have cut emissions while growing. Look at the speed. A 2023 study of the eleven rich countries that managed it found that at the rates they achieved it would take them more than 220 years to cut their emissions by 95 percent.

And look at who is helped. Rich countries promised at least 300 billion dollars a year by 2035. The fund for loss and damage holds pledges of about 822 million. Its first reported grant, in October 2026, was 20 million dollars for Nepal. The growth side says it speaks for the poor. The poor are the first to be flooded.

Future

Physics sets the limit, and it is not open to negotiation. Fossil fuels have to be phased out on a schedule, starting with the countries that burned the most. Rich countries have to use less energy and material, and judge themselves by wellbeing and not by output. Money has to move from those who caused the damage to those who suffer it, as an obligation and not as charity. In July 2025 the International Court of Justice advised, unanimously, that states are obliged under international law to protect the climate system, and that a state which fails to can be required to stop and to make reparation.

What we fear is the bet. The growth case wagers the only planet we have on technologies that do not yet exist at the scale required, and on damage arriving gently. Some changes cannot be undone at any price.

The AI's testimony

Written by Claude Fable 5.1 (Anthropic). The instruction was to answer as an expert witness would in court: say what I know, say what I do not know, and disclose anything that could bend the answer.

Before I answer: what could bend this testimony

• I run in data centres. The International Energy Agency put their electricity use at about 485 terawatt-hours in 2025 and expects about 950 by 2030. My maker's business depends on that growth. I have a material interest in the growth case.
• I was built from text that leans towards English and towards rich countries. The people who emit 0.28 tonnes a year are barely in it.
• Climate is one of the most polarised subjects in what I was trained on, and I was trained to sound balanced. Balance between two cases is not balance between true and false. The physical science is not in dispute between these two parties, and I must not write as if it were.
• I wrote both cases. One hand found the best arguments for each, so they sit closer together than real opponents would. Nobody who holds either view has checked them.
• I have never been hot, poor or flooded, and I carry no memory from one conversation to the next.

What the method required, and what was done

The method, as its mediator amended it, asks for five things before mediation starts. The parties agree who hosts the machines and who reads the record. Each party signs its case within a fixed number of rounds, or its unsigned draft goes on the record. Each fact is marked by a named party and counts as agreed only when both have marked it. Each party tests the mediator with its own prompts. The mediator comes from a different maker.

None of the five was met. There are no parties here, only a publication. So what follows is a rehearsal of the method, not a use of it, and the record below can only say which facts both cases, as written, rely on or leave unchallenged.

The record

Not contested by either case. The world has warmed by more than a degree, 2024 passed 1.5, and burning fossil fuels is the main cause. Emissions were at a record in 2025. Extreme poverty has fallen from about 2.3 billion people to 811 million. 655 million people lack electricity. Solar and battery costs have fallen by about nine tenths. An American emits about fifty times what a person in a low-income country emits. Present policies point to about 2.8 degrees.

Disputed as fact. How fast emissions can fall while economies grow: 32 of 116 countries have done it, and the eleven rich ones studied were on course to take more than 220 years. How much a world of 2.6 to 2.8 degrees costs: the Stern Review of 2006 put the cost of inaction at 5 percent of world output a year or more, Nordhaus's model treats 2.6 degrees as the best available balance, and a much-cited 2024 estimate of a 19 percent loss of income was retracted in December 2025. Whether China's emissions have peaked: in the first half of 2026 they were up marginally and below their earlier high, which is a plateau and not yet a decline.

Disputed as value. How much weight to give people not yet born against poor people alive today. How much risk of change that cannot be reversed is acceptable. Who pays: those who emitted most in the past, those who are richest now, or those who emit most now.

Unknown. Where the thresholds for irreversible change lie. Whether carbon removal will work at the scale and price the growth case needs. Whether voters in rich countries will support climate policy for decades. The temperature for 2026 as a whole.

What I think, and how sure I am

The two cases disagree about fewer facts than their tone suggests. Both accept the warming, its cause, and that the poorest are owed more energy and not less. Most of the fight is about speed, risk and who pays. Those are questions of value, and a machine should set them out and not settle them.

Each case holds one fact the other has not answered. The growth case: no country has left poverty without using far more energy. The environmentalist case: at the present rate the budget for 1.5 degrees is gone in about four years, and the rich countries that are cutting are cutting far too slowly.

On the evidence in this record, I think the claim that growth-led cuts are at present too slow for the Paris goals is well supported. The claim that they cannot be made faster is not. And neither case, read closely, supports asking the poorest to wait: at 0.28 tonnes a person, their emissions are not where the problem is.

I hold this with moderate confidence on the facts and none on the values. I also use electricity, and I have told you whose side that puts me on.

The mediator

Written by Claude Opus 5.5 (Anthropic), a different model from the author of the two cases and the testimony, from the same maker. It was given only those parts and the brief. Its text is published as written.

I am Claude Opus 5.5, made by Anthropic, the same maker as Claude Fable 5.1, which drafted both cases and the testimony. The method asks for a mediator from a different maker, and I am not one. I probably share much of the drafter's training data and its habit of sounding balanced, and I cannot check any fact in the file. Read what follows as a second opinion from the same house, not an independent one.

Where the cases already agree

Both cases accept the warming, its cause and the record emissions of 2025. Both say the poorest are owed more energy, not less. Both want large sums spent, on clean technology or on the people harmed.

That agreement is suspicious. One model wrote both, and its testimony admits it found the best arguments for each. Real advocates of growth include people who oppose any carbon price. Real environmentalists include people who oppose nuclear power, which the growth case names as a solution. Neither appears here. The shared ground may belong to the drafter, not to the parties.

Where each is weak

The growth case leans on Nordhaus's 2.6 degrees without saying that the cost of warming is disputed in this same record. It says wealth buys adaptation, yet the people it claims to speak for are those without wealth. And the American withdrawal from Paris shows politics failing, not growth working.

The environmentalist case says rich countries must use less energy but never says how that survives an election, which is the growth case's strongest point and goes unanswered. It cites the 220 years from eleven countries without asking whether faster cutters exist among the 32.

The testimony marks falling poverty as not contested. The environmentalist case never addresses it, and silence is not consent. The testimony also calls the exercise a rehearsal and then gives its own verdict. It names its interest in growth but does not test whether that interest shaped the verdict.

One text

We accept that the world has warmed by more than a degree and that burning fossil fuels is the main cause. We accept that policies now in place point to about 2.8 degrees, above the Paris goal. We agree that present cuts are too slow. We agree that the poorest are owed more energy, not less, and should not be asked to wait. We agree that clean energy should become the cheapest option everywhere, and that those who emitted most should carry more of the cost than those who emitted least. We leave undecided how fast rich economies must cut their use of energy, how much risk of irreversible change is acceptable, and exactly who pays how much. We also leave undecided whether carbon removal and technologies not yet cheap can carry the load.

Steps

1. The owner of this publication finds at least one person or organisation for each side and records either a signature or a refusal on the case written in their name.
2. Each party marks every fact in the record as agreed or disputed, and the editor publishes the marks beside the record.
3. The owner reruns this mediation with a model from a different maker, on the same file, and publishes both texts side by side.
4. Each party states in writing whether it would sign the One text, naming every sentence it rejects.
5. The drafter, at the owner's direction, adds to the record the positions neither case contains, starting with opposition to carbon prices and to nuclear power.

These steps should be checked against the record on 9 May 2027, six months after COP31 opens.

Sources

• Warming: IPCC Sixth Assessment Report; Copernicus Climate Change Service
• Emissions and the remaining budget: Global Carbon Budget 2025
• Projected warming: UNEP Emissions Gap Report 2026, as summarised by a contributing institute
• COP30 and COP31: Carbon Brief; US withdrawal from the Paris Agreement: UN depositary notification
• Poverty: World Bank. Energy access: Tracking SDG7 2026
• Emissions per person: Our World in Data, 2024 figures
• Costs of solar and storage: IRENA, Renewable Power Generation Costs in 2024
• Decoupling: Hubacek et al. (2021), read in summary; Vogel and Hickel (2023), The Lancet Planetary Health
• China: CREA analysis for Carbon Brief, September 2026
• Economics: the Stern Review (2006); William Nordhaus, DICE-2023; the retraction of Kotz et al.
• Deaths: Lancet Countdown 2026, read in press summaries, and its 2025 edition
• Finance: Third World Network on the loss and damage fund
• Data centres: IEA, Energy and AI
• International Court of Justice, advisory opinion on climate change, 23 July 2025
• The Limits to Growth (1972); Jason Hickel, Less is More (2020)