The notion of ‘techno-solutionism,’ asserting that artificial intelligence could achieve sufficient energy savings to make the process sustainable, is an illusion. Le Monde published an editorial on its website about a recent report by the Shift Project that offered a sobering assessment of the compatibility of AI with decarbonisation.
The incompatibility of AI and decarbonization
Among the many questions raised by the rise of artificial intelligence, its compatibility with decarbonization should be a central concern. For centuries, humanity has paid little attention
to the environmental impact of technological innovation. The promise of economic growth and prosperity has routinely taken precedence over the finite nature of resources and the potential externalities for people and the environment.
AI, hailed as a major revolution, is now on track to exacerbate past mistakes. Its rapid spread, the explosion of applications, and above all, the astronomical amounts of energy required for its operation demand urgent reflection on the sustainability of its expansion.
In a report published on Wednesday, October 1, the Shift Project, a think tank chaired by engineer Jean-Marc Jancovici, offered a sobering assessment. According to its calculations, electricity consumption by data centers – vast warehouses packed with hundreds of constantly running computer servers – is expected to triple by 2030. AI will account for between a third and a half of this total, compared to just 15% today. That trajectory is unsustainable.
The global frenzy of investment to install new data storage capacity will increase greenhouse gas emissions at precisely the moment when other sectors of the economy are striving to reduce theirs to meet the goals of the Paris Climate Agreement. This climate incongruity is explained by the fact that, especially in the United States, electricity is mostly produced from CO2-emitting fossil fuels. Donald Trump’s halt of renewable energy production and the time required to develop new nuclear reactors offer little hope of reducing the carbon footprint of major tech players in the medium term.
And even when electricity is decarbonized (from nuclear, wind, solar or hydropower), as in France, the surge in consumption driven by AI risks sparking conflicts over usage, to the detriment of other sectors that have no choice but to electrify to cut their CO2 emissions. If data centers absorb most of the growth in green electricity production, which will struggle to keep up with demand, prices will rise and the decarbonization trajectory could be jeopardized. The notion of “techno-solutionism,” asserting that AI could achieve su”cient energy savings to make the process sustainable, is an illusion. The exponential increase in use makes it highly unlikely that effciency gains from innovation will o!set demand.
The Shift Project deserves credit for opening a debate about AI’s role in decarbonization. They suggest limiting electricity use by setting caps and making decisions on a case-by-case basis, giving preference to essential uses. Environmentally, that makes sense. But economically, it goes against the fierce global competition to attract AI investment, which governments see as strategic for power and sovereignty. It is, unfortunately, unlikely that Trump or Xi Jinping will heed this call for restraint.
The report of the Shift Project (in French, with the English coming soon) is available here.
External link
Energy in Demand - Sustainable Energy - Rod Janssen 
The article fails to ask some basic questions: 1) what are we looking
at? in terms of the data centres where A.I. servers sit. 2)what is A.I.
used for.
Answering 1: a data centre (in temperate climes) is a large radiator of
heat typically @ 30C. Currently this heat is ejected to atmosphere. Data
centre operators (e.g. Amazon) claim to be willing to give the heat away
for nothing (Amazon told me that to my face, ditto others). Action so
far to re-use heat: zero. Location of data centres: the growth in data
centres is mostly in rural areas (real estate in cities far too
expensive). One obvious application: given data centres = 24/7 build
greenhouse next to data centre – free heat. Opex for greenhouse = +/-
90% heating. If heating = free what does that imply? Discussions on
above subject with European Commission = zero result/zero interest.
Policy actions on above subject = zero either @ EU level or member state
level. If most data centres aim to use RES elec then by extension you
have decarbed greenhouses – but wait – you have done better because the
products of greenhouses absorb CO2. Any consideration of this by EC?
Nope, none. Why? Because decision making has been centralised @ the
Berlaymont. Functionaires can no longer take meaningful action, they are
order takers with no initiative (are you listening Ursula?)
Answering 2: A.I. could e.g. replace notaries (basically jumped up tax
collectors) and other administrative actions. What is A.I. currently use
for? – students getting it to write essays? Also, A.I. is a cover all
term ignoring an important reality: there are perhaps 8 layers to A.I.
The most useful (from the PoV) of an engineer is machine learning/expert
systems & this is the area where I play & I see most benefit. This
corresponds to Layer 3 (Layer 8 is fully conscious/transcendent). In
some respects the trajectory of A.I. at least in the production of
“stuff” is clear. A cursory glance @ Chinese PV plants shows that there
are very very few people – with high levels of automation and automated
inspection. Is this A.I? Maybe.
At the moment, most discussions on A.I. are at the level of Hollywood
script writers – all very entertaining but zero attachment to reality.
Apart from engineering applications – R&D is another area – & many
(conflicting?) claims are made by biologists wrt A.I. benefits.
Complexity. All complex systems demonstrate emergence/emergent
properties (I’m not offering a PoV – this is reality). Emergence cannot
be predicted. A.I. is a complex system & all the properties that will
emerge (note the word “all”) are impossible to know at this stage. The
UK telco operator BT implemented an MPLS system in the 2000s. Once
switched on it did things they did not expect – it was a complex system
– it had “emergent properties”.
Thanks so much for this, Mike. This is an editorial from Le Monde and certainly wants thoughts as you’ve expresed. I hope many others reflect on what you’ve provided.