AI and climate change: The promise, the perils and pillars for action

This article was first published in Branch magazine, an online collaboration between EIT Climate-KIC, Mozilla Foundation and Climate Action.tech exploring the sustainability of the internet.

A global pandemic has shocked the world, leading to thousands of deaths, economic hardship and profound social disruption. While we worry about our immediate needs, we should remember that another crisis is looming: climate change.

The lockdown made it clear that staying at home and slowing down the economy is far from enough to solve the climate crisis. We’re still emitting more than 80 per cent as much CO2 as normal, despite having 17 per cent fewer emissions compared to 2019 — which is one of the most significant drops in recent years (1). If we don’t act decisively now, the economic damage caused by climate change in the next two decades will likely be as bad as a COVID-sized pandemic every ten years (2). 

Starting today, we need to accelerate our zero-carbon transition rapidly. This transition requires mitigation and adaptation measures that reduce greenhouse gas emissions and build resilience towards weather-related disasters. Despite vested interests, geopolitical competition and populist leaders, tremendous technological progress is being made towards tackling the climate crisis. In recent years, promising applications of artificial intelligence and data science have been developed to make sense of the vast amounts of data generated across sectors and better monitor Earth’s resources.

But several questions remain unanswered: to what extent can AI contribute to a net-zero economy? How quickly can this happen, given the urgency of the challenge? What is the net effect on the planet? And what can we do about this?

The promise

First, AI systems have the potential (3) to decouple economic growth from rising carbon emissions and environmental degradation. AI as a stack of data, learning algorithms and sensing devices can help with both impact and resource decoupling. 

Impact decoupling means decreasing environmental harm, including CO2 emissions, per unit of economic output (4). For instance, we can halt emissions in the energy sector by using AI technology to forecast the supply and demand of power in the grid, improve the scheduling of renewables, and reduce the life-cycle fossil fuel emissions through predictive maintenance. AI applications in transportation can enable more accurate traffic predictions, the optimisation of freight transportation, and better modelling of demand and shared mobility options. Other kinds of impacts include the waste that is disrupting ecosystems, pollutants that affect human and animal health and biodiversity loss. By harnessing the swaths of data from sensors and satellites, we can better predict climate change impacts and proactively steward these ecosystems. We can also actively increase the capacity of carbon sinks like peatlands and accelerate afforestation through locating appropriate planting sites, monitoring plant health and even controlling tree-planting drones.

Additionally, AI can be helpful in resource decoupling, which means the decoupling of economic output from the volume of resources used from the environment such as materials, water and land. By one example, AI applied in food systems can help better monitor crop yields, reduce the need for chemicals and excess water through precision agriculture and minimise food waste through forecasting demand and identifying spoiled produce. Lastly, AI systems used in buildings and cities can help automatically control heating and cooling as well as model energy used to decide which buildings to retrofit. 

The perils

AI for climate action has the potential to reduce global greenhouse gas emissions by up to 4.0 per cent (5). However, global data centres and predictive algorithms are also accelerating international chains of logistics, the extraction of resources and fossil fuel emissions in ways that we don’t see or understand. As we look to the future, we need to ensure that the benefits of using AI to tackle climate change outweigh the drawbacks. 

The services offered by big tech are compromising the path to a green transition. They are not about decommissioning, carbon capture or carbon sequestration.

Recently, the energy consumption of AI systems, specifically machine learning, has come under scrutiny. Between 2012 and 2018, the computations required for deep learning research have doubled, resulting in an estimated 300,000x increase. Several factors impact the carbon emitted by neural networks: the location of the server used for training; the energy grid that it plugs into; the size of the dataset; and the hardware where the training takes place (6). Even so, the use of increasingly energy-efficient processing units, as well as efficiencies in servers, storage, devices and hyperscale data centres offer some optimism for the future. 

What is more alarming is the special oil and gas divisions or big partnerships of big tech companies like Google, Amazon, and Microsoft with companies like Chevron, Total, Aramco, ExxonMobil, Shell and BP. The services offered by big tech are compromising the path to a green transition. They are not about decommissioning, carbon capture or carbon sequestration but instead helping oil companies identify wells, automate drilling, and make extraction as efficient and high-yield as possible. This indicates that the promise of decoupling may be relative, i.e. emissions will not decline fast enough, and it may concern only specific resources and locations.  

Since 2019, technology companies have responded to these concerns by adopting new green policies and initiatives. Examples include Microsoft’s pledge to be carbon negative by 2030 and remove all the carbon the company has emitted since 1975; the use of low-carbon aluminium by Apple; Google’s $5.75 billion sustainability bonds that will fund environmentally and socially responsible projects; and Amazon’s pledge to be net zero by 2040. In the next five years, we need to carefully monitor whether these commitments are industry marketing or genuine climate action plans. 

Pillars of action

AI brings lots of opportunities, but there are also tradeoffs and concerns. Shaping a positive scenario for our future will require collective action at multiple levels: integrating technology regulation with Green New Deal policies, developing new standards to mitigate environmental impacts, adopting green AI industry guidelines and training the next generation of responsible AI technologists. 

Any application of AI in climate change mitigation and adaptation will need to ensure that environmental impacts are not externalised onto the most marginalised populations

Moving forward, I want to suggest four pillars of action for technologists, data scientists, designers, engineers and technology activists:

DEVELOP ENABLING TECHNICAL ENVIRONMENTS FOR THE GREEN TRANSITION 

I invite technologists to apply their skills to climate change mitigation and work towards transforming how data-driven solutions are being developed and commercialised at scale. Industries like energy, food, manufacturing and finance need to transition within the next five years. A trustworthy data and AI environment will require, among others: open standards, shared frameworks for data sharing and robust data discovery and publishing practices between transition industries. These emerging data markets can give us a systemic picture of supply and demand at national and regional levels. Moreover, the integration of various forms of public, private, and citizen science data will require guidelines for public-private data collaborations that can be materialised through data commons and other novel data institutions. 

DEVELOP A CLIMATE AWARE DATA SCIENCE PRACTICE

AI and data science communities will need to follow the steps of computer scientists who have a long history of investigating sustainable computing. Researchers may advocate for making efficiency an evaluation criterion for research, use more computationally efficient hardware and algorithms and report the “price tag” of their models. Alternatively, Energy Usage Reports have been proposed as part of greener algorithmic accountability practices and tools like Machine Learning Emissions Calculator can help estimate the amount of carbon emissions produced by the training of AI models. Similarly, practitioners may start reporting the time to retrain models, share local infrastructure instead of relying on cloud computing and choose cloud providers who are offsetting their emissions.

FOCUS ON CLIMATE JUSTICE

A just transition requires that we pay attention to the struggles of various communities who are already defending their land, air, water, and livelihoods from extractive activities such as mining, fracking, gas flaring, etc. Any application of AI in climate change mitigation and adaptation will need to ensure that environmental impacts are not externalised onto the most marginalised populations, and that the gains are not only captured by digitally mature countries in the global north. This requires centring front-line communities and enabling them to take ownership of their data and bottom-up climate action plans. 

ORGANISE IN THE WORKPLACE 

In 2019, thousands of employees from Amazon, Google, Microsoft, Facebook and Twitter organised as the Tech Workers Coalition. They marched to demand from their employers to bring their emissions to zero by 2030, stop exploiting climate refugees and cancel contracts with fossil fuel companies. It will be of paramount importance for technology workers to raise awareness in their work about the climate impacts of technology. Technology companies need to be more transparent about their emissions and be pressured to provide this information to customers, regulators, and the public. This transparency will be the first step towards informing regulation and public discourse as well as incentivising practitioners to make more sustainable decisions.

Towards an ethics of planetary care

The planetary scale of our knowledge and technologies are revealing new interdependencies and feedback loops between environmental and engineered systems. This renewed understanding requires an updated ethical, ontological, and practical discourse—one that is not reductionist, but rather makes the moral responsibility for planetary custodianship even more urgent. Accordingly, the consideration of environmental impacts and the responsibility to care for our planet should be reflected in our technical infrastructure, our ways of working and our practices and policies for fair, accountable, transparent and ethical AI systems. AI will not be a substitute for more integrative ways of knowing or even degrowthist political agendas — but rather, when used responsibly, can be an enabler that helps us move faster to a safe and just post carbon world.

REFERENCES

1. Le Quéré, C., Jackson, R.B., Jones, M.W., Smith, A.J., Abernethy, S., Andrew, R.M., De-Gol, A.J., Willis, D.R., Shan, Y., Canadell, J.G. and Friedlingstein, P., 2020. Temporary reduction in daily global CO 2 emissions during the COVID-19 forced confinement. Nature Climate Change, pp.1-7.
   preprint arXiv:1907.10597.
2. Bill Gates, 2020, “COVID-19 is awful. Climate change could be worse” , Gates Notes, Website accessed 15th of August 2020
3. Rolnick, D., Donti, P.L., Kaack, L.H., Kochanski, K., Lacoste, A., Sankaran, K., Ross, A.S., Milojevic-Dupont, N., Jaques, N., Waldman-Brown, A. and Luccioni, A., 2019. Tackling climate change with machine learning. arXiv preprint arXiv:1906.05433.
4. Parrique, T., Barth, J., Briens, F., Kuokkanen, A. and Spangenberg, J.H., 2019. Evidence and arguments against green growth as a sole strategy for sustainability.
5. Joppa, L. and Herweijer, C., 2018. How AI can enable a sustainable future. Microsoft in association with PwC.
6. Schwartz, R., Dodge, J., Smith, N.A. and Etzioni, O., 2019. Green ai. arXiv

 About the author

Eirini Malliaraki is a design engineer and entrepreneur. She oversees project development on climate change and environmental science at the Alan Turing Institute, the UK’s National Institute for AI and Data Science.