The year past has seen the most unbelievable acceleration in global warming ever recorded, one that has sent climate scientists reeling. Scientists are divided between those who see this as a bad, but expected, short-term fluctuation, and those who see it as a breaking point, a sign that our historical models are failing to adequately capture our present and our future. In the March 2024 paper, Global Warming Acceleration: Hope vs Hopium, James Hansen, Makiko Sato, and Pushker Kharecha argue that the acceleration is genuine:
Accumulating evidence supports the interpretation in our Pipeline paper: decreasing human-made aerosols increased Earth’s energy imbalance and accelerated global warming in the past decade. Climate sensitivity and aerosol forcing, physically independent quantities, were tied together by United Nations IPCC climate assessments that rely excessively on global climate models (GCMs) and fail to measure climate forcing by aerosols. IPCC’s best estimates for climate sensitivity and aerosol forcing both understate reality. Preservation of global shorelines and global climate patterns – the world humanity is adapted to – likely will require at least partly reversing global warming.
The global atmospheric CO2 for March 2024 is 425.38 ppm, an astonishing 4.39 ppm increase over the same time last year. I’m not going to even try to document the thousands of temperature records falling around the globe. Suffice to say that we are already pushing at the 1.5C threshold that we were supposed to meet in the 30s.
Now, one persistent claim of the AI advocates is that it will solve climate change. Countless articles tell you this.
There is no doubt investment in AI by sincere people who genuinely believe that it will help. But what if we focus on the chief leaders of the movement, those who direct the industry? Their desires are not your desires. Those who adhere to the philosophies known as “accelerationism” and “longtermism”, which are highly influential in AI circles, genuinely don’t care about climate collapse, because for them it is not an existential risk. In almost every climate collapse scenario, humanity in some form will survive, so for them this doesn’t rate beside the true threats, AI itself at the top of them. So if we have to drive our world through climate collapse to reach the glorious future promised by genuine AGI, it’s more than worth it.
Meanwhile, fossil fuel companies are massively investing in AI to drive extraction and profits. This is being enabled by Microsoft, Google, all the usual suspects. The Financial Times reports that the “AI revolution will be boon for natural gas, say fossil fuel bosses”.
The immediate problem is that these models take a vast amount of energy to run. How much, exactly? Once again, we find that, for all their promises of digital upotias of perfect knowledge and infinite data, the companies are shy about telling us any actual figures. In 2024, the IEA forecasts that:
Electricity consumption from data centres, artificial intelligence (AI) and the cryptocurrency sector could double by 2026. … This demand is roughly equivalent to the electricity consumption of Japan.
These systems are creating a cascading waterfall of damage, so it is hard to keep up. Many articles are recorded in the curated list at this link, of which these are just a few. According to the BBC, the CEO of Britain’s National Grid predicts that data center power usage over a decade will expand sixfold. Such centers use massive amounts of water, which is sometimes pulled from the desert. Environmental harms are outsourced to poor countries. An MIT report emphasizes the multi-dimensional environmental harms of the AI datacenters, which include the dirty energy used for running them, the materials for building, the water, the noise, and the waste.
Data is matter in motion, and it is moved by energy.
Defenders argue that AI’s usage is only a negligible percentage of total energy usage. It’s an unserious argument. Energy usage is already too great and growing, and as a result, global atmospheric CO2 growth is accelerating. We’re going the wrong way. We need to stop doing the things we are already doing, not add yet another whole new industry.
AI applications are typically more energy intensive than the conventional means they replace. From Vox:
According to the IEA, a single Google search takes 0.3 watt-hours of electricity, while a ChatGPT request takes 2.9 watt-hours. (An incandescent light bulb draws an average of 60 watt-hours of juice.) If ChatGPT were integrated into the 9 billion searches done each day, the IEA says, the electricity demand would increase by 10 terawatt-hours a year—the amount consumed by about 1.5 million European Union residents.
We’re going the wrong way.
But this is just the beginning. An incisive analysis by Paul SchĂĽtze shows how the realities of AI harm go far beyond energy usage:
AI futurism constitutes a mode of societal organization which hides the global material realities of actually-existing AI capitalism behind the curtain of promising projections
While AI has yet to fulfill its promise of solving climate change, it turns out to be great at spreading harmful climate disinformation.
But the fundamental problem lies in the very prosperity that AI promises. CO2e emissions are driven by economic growth. If, as AI acolytes continually promise, AI creates a world of abundance for all, that abundance is going to kill us. They have no plan for this. They just imagine that their machines will solve all the problems. Either we’ll magick up infinite energy or we’ll settle the stars before it all goes belly up.
We’re going the wrong way.
At a bare minimum, AI clients should be required by law to disclose their energy usage. This should contain three details:
- The total energy use of the project
- The human equivalent (eg. “equivalent to the electricity usage of 1.5 million EU residents”)
- The amount used for each operation compared to a conventional application.