Some have argued that the wide use of Artificial Intelligence (AI) will require substantial amounts of energy and that this will overwhelm any environmental gains from improved energy use efficiency and emission reduction.
However, there are other views. LSE's Grantham Institute and Systemiq produced a study last year claiming that AI applications in just three sectors (power, food, and mobility) could reduce global greenhouse gas emissions by 3.2 to 5.4 billion metric tons annually by 2035. That would more than offset all projected data centre emissions from AI across the entire global economy in the same period, and move us 36% closer to an ambitious 2035 emissions reduction trajectory versus business as usual.
A recent overview of the issues said that ‘This would be a huge win, and powered by only three sectors’. But it also reported that a more recent paper from MIT is much more cautious: ‘AI is likely to worsen climate change even if it stimulates large emissions reductions in the future. The conditions under which a net-positive outcome might be achieved - such as global carbon pricing, mandatory disclosure of full system-wide environmental costs, and aggressive state intervention to direct compute toward public goods - are not currently in place’.
Clearly there are still some big uncertainties and a need for careful assessment and management, as Terry Cook and I argue in a forthcoming paper. As we note, there have already been issues in the USA and problems seem to be continuing in many places, for example with AI use adding to power bills. That problem might be avoided over time if use is made of renewables, but nor every potential AI site will be suited to that. Or, to put it the other way around, to avoid transmission costs, AI sites will have go where there are good renewable sources. So far though that is not the way it’s going - for example, in the UK, there are plans for 100 new data centres to use power produced from burning fossil gas.
However, the debate continues, for example, in the UK, in relation to Elsham Tech Park Ltd in Lincolnshire, where a large new £10bn data centre is planned. Last year the Guardian reported that the developer had ‘ ruled out on-site renewables as impractical’, arguing that ‘if the system ran on biomass energy it would require the daily delivery of 100 large lorry loads of wood chips. Wind energy would require 10,000 20-metre wind turbines, while an area five times the size of the Glastonbury festival site would be needed if it were to be powered by photovoltaic panels’. But it has now got the go ahead, with, according to the Guardian, the local council concluding that despite the ‘large absolute energy demand’, the impact of emissions was not significant due to the datacentre’s proximity to clean energy sources in the Humber region, where, for example, there are lots of offshore wind capacity. But nuclear is also in this game, even though there are big uncertainties: Small Modular Reactors are still some way off and big nuclear plants are inflexible and more costly. So it’s slow going for nuclear.
By contrast, renewables are booming and creating a lot of new green jobs, as I noted in my last post. And there could be much more to come. For example, the ‘Workforce Foresighting for Offshore Wind 2030-2035’ report from Offshore Renewable Energy (ORE) Catapult, sponsored by RenewableUK, says that there are significant opportunities for UK jobseekers in the sector, from wind turbine technicians and high-voltage cable specialists to installation engineers and fabrication specialists. But it notes that the UK’s training systems are struggling to keep pace with technological progress: RUK says ‘we need to take action now so that the capabilities and capacity of our workforce will be sufficient to build the vast pipeline of projects which will be rolled out to meet the UK’s ambitious offshore wind targets.’
Terry Cook and I argue in our forthcoming paper that AI may help with this and with the expansion of renewables. Indeed, we note that that is already happening, with, for example, an expansion of AI-driven training programmes mounted via collaborations between AI tech companies & energy sector firms, via degree level apprenticeships. For example, in the UK, in E.ON, EDF & RWE. But, however it is done, much more is needed and soon. ORE Catapult says that, new training scheme are urgently needed: ‘There is little, if any, time to spare. If started now, the full cycle of developing course content, recruiting learners, re-skilling and providing new employees with the opportunities to gain on-the-job experience, will take until 2030’.
While attempts to expand AI training and the use of renewables may be good news, it is still not clear what the overall impact of AI will be in climate terms. A recent paper from a group of Indian academics suggests that ‘its impact will likely exceed available management solutions’- much as is said in the MIT paper mentioned earlier. That said that ‘absent complementary climate policies, efficiency gains enabled by AI are likely to trigger direct and indirect rebound effects that erode or reverse emissions savings. Critically, AI emissions are growing today, while AI-enabled emissions reductions will take time, creating an AI "carbon debt" that accelerates warming.’
Moreover, the MIT paper says that, even if the AI carbon debt ‘is eventually repaid by AI climate solutions, the additional warming and climate damage will persist for decades to millennia: carbon neutrality does not imply climate neutrality’. But the Indian paper is a bit more hopeful. It concludes that ‘AI’s ecological future is not predetermined and will ultimately be a product of the cumulative and collective choices regarding technology, policy, and ethics that lead AI development to long-term ecological viability’.
Well maybe AI expansion can and will be managed successfully, with, for example, flexible AI cutting its peak energy use by 40%. But most recent papers admit that outcomes are all still uncertain and it is clear that AI opens up some big ecological, social and political issues - not least likely major impacts on conventional employment patterns. This all needs to be faced: big changes are afoot and they need careful assessment and effective regulatory oversight. See the ‘This is not inevitable’ AI conference planned next week in Manchester…
Comments
Post a Comment