Means, Motive and Opportunity: Bad Maths

The excitement around economic modelling in business, government and political circles in the early 1990s was similar to the hype about generative AI today. They are, of course, different innovations from different eras, but the core proposition remains the same. AI is presented as a knowing machine that takes the work out of thought. In reality, the machine cannot comprehend; it just recognises patterns. In doing so, it can only approximate someone else’s knowledge by spitting out what it thinks is most likely to occur next in the sequence, based on the data on which it was trained, to achieve the goal it has been given. Large language models, in particular, project the next word based on the pattern before it to generate what looks like a cogent response. That is enough to create the perception of insight.

Eventually, the argument goes, one AI will be used to train the next until we eventually build “god in a box—during an interview in July 2026, Anthropic CEO Sam Altman described AI as “a genie that can grant any wish”. This oracle will allegedly be so intelligent it will be capable of solving problems better and faster than any human. Of course, there is a trade-off. The physical infrastructure needed to support this computing power involves huge quantities of plastic, steel and cement. AI datacentres are power-hungry and, depending on the build, thirsty. A fleet of gas turbines is required to power them, creating a demand expected to keep petroleum companies in business for years to come.

The carbon emissions from all this activity will be so large that some researchers have described it as “devastating” when “applied at scale”. All of that will be worth it, the idea goes, because humanity could then ask the AI oracle how to solve its biggest problems, like climate change. Whatever the cost in dollars or damage caused by carbon emissions from continuing to burn fossil fuels, it will all be justified by the inconceivably wonderous future AI will bring. In fact, according to figures like techno-futurist Eric Schmidt, humanity must burn as much oil, gas and coal as it takes to create this god if it ever hopes to solve climate change.

“Knowing that climate change is real and knowing that it’s a problem is that the intelligence revolution, the ability to do planning and discovery will allow us as Americans to develop new materials, new energy sources and so forth because of the AI data centers,” he told a hearing of the US senate in April 2025. “So our core argument is invest in the way we can now because the future will be so much cleaner and so much more efficient as a result of these algorithms.”

Schmidt is not alone in embracing this logic. Billionaires Bill Gates and Jeff Bezos have sunk hundreds of millions of dollars into developing AI as a “climate solution”. Their similarly credulous faith in technology to solve climate change has been trickling down into public discourse.

What is curious is the extent to which history repeats. Before AI was being used by billionaires to justify delaying any energy transition, company executives and political leaders were using economic modelling to argue any phase-out of oil, gas and coal would be a disaster. Economic models, as sufficiently complex spreadsheets, can be thought of as an evolutionary ancestor to machine-learning algorithms and large language models. Likewise, climate change has always functioned as a convenient laboratory for the deployment of these speculative technologies. Both originate from the same basic impulse: the illusory promise that technology will solve a social, political and economic problem.

With the benefit of hindsight, the history of economic modelling shows these elaborate spreadsheets largely served the interests of the powerful. In Australia, it was ABARE that pioneered their use, with the support of oil, gas and coal producers.

It makes sense that economic models would attract the attention of the powerful. If knowledge is power, numbers are bricks. This makes numbers powerful in turn. Numbers suggest authority. They offer the illusion of certainty, free from subjectivity. They carry none of the messiness and vagaries of human analysis and interpretation. There can be no argument about whether six means two. A number just is. Numbers can be placed together to form a wall, built into a skyscraper, or torn up from the street and hurled through a plate glass window. An economic model, as a collection of numbers and the formulas required to generate them, ends arguments.

Applying economic modelling to climate policy has evolved over time, but what matters is this: an economic model aspires to be a world on paper. It is far more complex than a run-of-the-mill budget, but not so complex that it can reliably function as a forecasting tool. More specifically, it is an abstract representation, illustrated in numbers, of whatever economic driver it measures—be it across a segment of the economy, a state, a nation or the entire world. Huge reams of data go into building these models and, once sufficiently advanced, the inputs can be changed to generate hypothetical worlds or futures—the same method later deployed by Steve Brown, whose work was rejected by the Federal Court in the “Rocky Hill” case decades later.

There are legitimate uses for these sorts of tools, but measuring relationships and testing hypotheses were not how they would be deployed when it came to climate change. To a politician or senior bureaucrat with an ounce of power and the necessary resources, the potential seemed so much more.

On the one hand, modelling offered the illusion of risk-free decision-making. Those more perceptive understood how it might also be used as a powerful tool for persuasion. Because it was impossible to replicate the real world in the abstract, building any given model required making an educated guess or three. Who made these “assumptions” and who chose the inputs had enormous power to influence the result. Printed in a 100-page report with an executive summary, a methodology full of equations and pages of appendices, no one would question the conclusion. With time, the powerful would learn how these models could be made to justify whatever it was they wanted to do. Numbers convey authority. A computer never lies.

The idea of applying economic modelling to climate change was popularised by economist William Nordhaus. Around this time, Nordhaus infamously suggested a little climate change might not be so bad. In 1992, he combined economic and climate modelling in something called the “Dynamic Integrated Climate-Economy”, or DICE, model. The year before, an early effort concluded that the damage to the US economy would not be the worst—measuring about 13%. The apparent precision was remarkable. The basis for this conclusion? Two words: air conditioning. Nordhaus’s model assumed that, as most modern work takes place in “carefully controlled environments”, the effects of extreme weather events like fire, flood, drought and heat would be limited; office blocks, apparently, neither burn nor flood nor swelter.

In Australia, a group of academics and consultants, including economists at the Melbourne University Institute of Applied Economics and Social Research, had been playing with their own economic model. In 1989, resource giant Conzinc Riotinto of Australia (CRA), later renamed Rio Tinto, hired them to assess the implications of a plan to reduce emissions by 20% by 2005. CRA was a major producer of coal and uranium and it marked the first significant use of the modelling technology by a company of this scale. CRA would similarly maintain an interest in ABARE’s work in this area during Dr Fisher’s tenure as leader.

ABARE’s internal documents show the evolution of the organisation’s thinking as the bureau investigated the cost and effectiveness of available databases and models. None seemed to go far or deep enough. They were either too limited, too expensive or didn’t have the capability ABARE wanted. By 1993, ABARE was at work developing its own model, MEGABARE, with the help of consulting firm ACIL Economics. Unlike other models, MEGABARE’s advantage was that it married trade flows with a model of the national economy.

That year was a busy one for ABARE. Barry Jones published an article in the journal Energy Policy titled “Reducing Australian energy sector greenhouse gas emissions”, which outlined an early attempt to put these models to work on climate policy. As lead author in the three-person ABARE team, Jones used economic models of the Australian energy sector to estimate the cost of climate action. His conclusions would enter the mainstream and become a well-known objection to any suggestions that Australia do something. Jones’s reasoning was it would cost far too much.

“The changes to the electricity system alone [. . .] would result in a reduction in annual growth in domestic product of about 0.3%, and a somewhat larger reduction in growth in real wages, accompanied by a substantial devaluation of the Australian dollar,” he and his co-authors wrote. “Detailed analysis indicates that the target can be met by the energy sector, but only with substantial change to the structure of the sector, achieved at considerable cost.”

ABARE would later cite this same article as an authority when it was contracted in 1994 by the Department of Foreign Affairs and Trade (DFAT) to run a more ambitious study on the implications of addressing climate change using its very own model, MEGABARE. Before it could embark on this work, it first had to actually build the thing.

Royce Kurmelovs is an Australian freelance journalist and author covering the intersection of class, finance and climate change.

This is an edited extract of Means, Motive and Opportunity (Vantage Point Issue 7) by Royce Kurmelovs, published by Australia Institute Press. Available from the 9th of November 2026.