Many critics of AI technology rightly point to the enormous amounts of power required to train and run these systems.
Even among those who accept how impactful AI could be for productivity, at a time when the world is already facing a catastrophic climate crisis, sceptics are understandably opposed to heavy investments into such an energy-intensive technology.
A common response to this criticism of AI is that while it does require vast amounts of power to run, its ability to improve efficiency and accelerate innovation could apply to efforts against climate change, thus offsetting its environmental costs.
There have been a few recent studies on this subject, including from the likes of Microsoft, PwC and Google, though these are not peer-reviewed, do not disclose all the details of their methodologies and, notably, are conducted by firms with a major stake in AI.
Research into the exact potential of the technology as a tool to tackle climate change therefore remains limited, but researchers from the London School of Economics and Systemiq have attempted to fill this gap.
In the Green and intelligent: the role of AI in the climate transition report, the team identified a handful of key areas in which AI can effectively respond to climate threats, among them were:
Transforming complex systems
There are a number of complex systems currently contributing significantly to global emissions that the report suggests are ripe for optimisation with AI.
It offers the example of integrating renewable energy into power grids. One of the historic challenges of relying on sources like solar and wind is that supply of these energy types can fluctuate aggressively, meaning grids have to be able to adapt what sources are being used depending on availability and price.
AI has already started being used to optimise grid management. With intelligent automation, grids can forecast supply and demand more accurately and manage the distribution of energy accordingly.
Figures from DeepMind reveal AI can improve the economic value of wind energy by at least 20% by reducing the reliance on standby power sources.
The report notes that other complex systems, such as urban management of transportation and the construction of infrastructure all can be optimised for climate impact with AI forecasting, management and analysis.
Innovating technology discovery and resource efficiency
Another exciting area of opportunity that is already being explored in the early stages is AI’s ability to assist in the discovery and development of new technologies that can support the reduction of emissions.
The International Energy Agency has estimated that close to half of the emission reductions required to meet existing net zero commitments will come from technologies currently at the prototype or demonstration phase.
According to the report, the development cycle for these new technologies can be rapidly sped up as AI’s ability to analyse millions of datapoints can save vast amounts of time for researchers making climate-friendly materials, renewable energy storage techniques and alternative proteins to reduce livestock emissions.
Perhaps the most notable example of this in action is from DeepMind, which fostered the creation of the AlphaFold model that predicted the structure of 200 million proteins – and earned a Nobel Prize.
Nudging and behavioural change
According to the Intergovernmental Panel on Climate Change (IPCC), consumer and lifestyle behaviour change can reduce between 40% and 70% of greenhouse gas emissions by 2050.
This is easier said than done. Even among consumers who are happy to adapt their lifestyle to ease the burden on the environment, they are faced with conflicting information and, as the report notes, “inefficient market signals”.
The report suggests that AI can overcome any psychological or human error barriers by providing tailored recommendations for products, services and practices to consumers and businesses backed by real data.
“If AI is intentionally directed toward accelerating low-carbon solutions in key systems such as energy, transport, and food, the potential emission reductions could vastly outweigh the additional emissions from powering AI,” Mattia Romani, Systemiq partner and LSE researcher, told UKTN.
AI’s climate impact quantified
In an effort to quantify how incorporating various AI techniques could realistically reduce emissions, the researchers examined the climate impact of three sectors that collectively contribute and how it might be affected
Looking at the combined emissions of the power, meat and dairy and light road vehicles sectors in 2023, it was found to be 27.2 gigatonnes of carbon dioxide equivalent (GtCO₂e).
Considering existing efforts outside of AI and other factors, the report estimates this to fall to a range of 22.4 to 23.3 GtCO₂e by 2035, defined as the business as usual (BAU) projection.
When factoring in the potential influence of AI, from grid management, to the improved development of alternative proteins to shared mobility schemes, this figure falls further to a range of 17 to 20.1 GtCO₂e, demonstrating an accelerated reduction of emissions among these particularly polluting sectors.
Romani warned that there could be a risk that the demand for AI grows faster than its ability to optimise climate efficiencies, however, he is optimistic that this risk is “manageable”.
He said: “To ensure these trends scale fast enough, we need stronger policy frameworks and standards that incentivise clean, efficient AI.
“This is why in our paper we call for governments, the tech sector, and the energy industry to co-create solutions.”
