Page 49 - Frontier Technologies to Protect the Environment and Tackle Climate Change
P. 49
Frontier Technologies to Protect the Environment and Tackle Climate Change
The limitations of AI
While AI offers vast potential in the fight against climate change, as per the above examples, it must
also be noted that there are certain limitations and downsides pertaining to its carbon footprint.
These are explored in Box 9.
Box 9: The carbon footprint of AI and machine learning (ML)
137, 138, 139, 140
The carbon cost of AI and ML
Artificial Intelligence (AI) and machine learning (ML), like most technology, have the
potential to help in the fight against climate change. They can make systems more efficient
(e.g. prevent electricity loss during transmission), enable remote sensing and automatic
monitoring (e.g. pinpoint deforestation, gather data on buildings, and track personal energy
use), provide fast approximations to time-intensive simulations (e.g. climate models and
energy scheduling models), and has the potential to lead to the development of interpretable
or causal models (e.g. for understanding weather patterns, informing policy makers, and
planning for disasters). Figure 12 shows some of the applications of ML.
Figure 12: Applications of machine learning [xii]
It has been estimated, for instance, that ‘using AI for environmental applications could boost
the global economy by up to USD 5.2 trillion in 2030, a 4.4 per cent increase on the business-
as-usual scenario, while reducing GHG emissions worldwide by 4 per cent, equivalent to the
2030 annual emissions of Australia, Canada and Japan combined.’
The efficacy of ML and AI will, however, rely on bringing together several factors, including
ensuring their effective integration into other technologies and – because they require large
amounts of computing power – decarbonization of the energy system to ensure that AI and
ML can fulfil their sustainability potential. This is imperative as new studies are showing
that typical current ML processes can ‘emit more than 626 000 pounds of carbon dioxide
equivalent (CO e) – nearly five times the lifetime emissions of the average American car
2
(and this includes manufacture of the car itself)’.
The significance of this finding is immense, and until the issue of efficiency can be
addressed, AI research within academia may even be impacted due to lack of necessary
computational resources.
43