AI tools are spurring climate action but also present a danger, CIFAR researcher says
January 3, 2024
Dr. David Rolnick, PhD (photo below), is using artificial intelligence in numerous applications – from accelerating climate modelling to monitoring insect biodiversity – to counter the impacts of climate change.
But the AI researcher is also keenly aware that AI presents significant risks related to climate change, in everything from self-driving vehicles to oil and natural gas extraction.
“The impacts of technology are not inevitable. They’re something that we choose, either implicitly or explicitly, across society,” Rolnick, assistant professor of computer science at McGill University and Canada CIFAR AI Chair at McGill and Mila-Quebec AI Institute, told Research Money.
“The choices are being made by technologists, by policy makers and by society at large. And also by the media, in highlighting different kinds of AI-enabled technology.”
In Rolnick’s view, media has paid far too much attention recently to generative AI incorporating large language models – such as ChatGPT – and far too little attention to the less “flashy” AI tools for helping the climate.
“I see a great growth and interest in AI,” he said. “But this is growth and interest is in the least productive kind of AI that is not helping the climate in any clear way – or at least the ways in which it helps the climate are more diffuse across society – and it is as likely to harm the climate as well as harm the society in many other ways.”
Rolnick’s research focuses on using and strengthening AI tools for fighting climate change. This encompasses AI-enabled applications across diverse areas, including biodiversity monitoring, crop yield prediction, climate model acceleration, carbon stock estimation in forests, new materials discovery for renewable energy, and many others.
“The unifying theme across these areas is there are opportunities for AI to help accelerate climate action through close partnership between stakeholders with different expertise and thinking carefully about the tools that are needed to make impact happen,” he said.
Rolnick recently returned from Panama, where he and his partners – including the U.K. Centre for Ecology and Hydrology, the Montreal Insectarium, and other organizations – were testing an AI tool for monitoring insect communities as an indicator of an ecosystem’s biodiversity.
Earth’s 1 million insect species represent one-half of the planet’s biodiversity and are essential to the other half. But climate change and other human activities significantly threaten this biodiversity, Rolnick said, “to the tune that an insect ‘apocalypse’ has been frequently cited as a catastrophic threat to global ecosystems.”
The AI-enabled tool he and other researchers use includes solar-powered ultraviolet light camera systems that attract and automatically photograph insects.
“We’ve developed algorithms that can track and identify different insect species, potentially thousands at a given location,” Rolnick said.
There aren’t enough entomologists in the field working on gathering data on insects, he added. “Our AI tool enables data gathering at a scale that was previously unprecedented, for ecology and land-use decisions, and to improve equity across both geographies and taxonomic groups that previously were quite under-served.”
The AI tool, which has been deployed across six nations and three continents, has led to the discovery of more than 100 new insect species.
Rolnick and his research group are currently focused on gathering data on moths, which represent about one-fifth of all insects and are significant pollinators important to agriculture.
The aim is to create a global repository of insect data that enables ecologists to use AI-enabled tools without having to understand anything about AI, he said. “That’s often the most important thing: making these AI tools accessible to people who need them.”
AI research requires multiple stakeholders to have real impact
Rolnick co-leads the Canada CIFAR AI Chairs Working Group in the priority area of AI for Energy and the Environment. He is also co-founder and chair of the nonprofit Climate Change AI, a global forum for thought leadership, research, funding and policy to better leverage AI to counter the impacts of climate change.
Environment and Climate Change Canada has funded another of Rolnick’s projects, which involves working with the federal department on AI algorithms to produce weather and climate models. Such models can run accurate weather and longer-timescale climate simulations, but often take months to run even on supercomputers. That’s too slow for guiding local responses to climate change.
“We have worked on particular algorithms that are designed to speed up that [modelling] process, not by improving on the physics but by quickly approximating it, with a slight loss in accuracy but a great improvement in speed,” Rolnick said.
Environment and Climate Change Canada is currently integrating the work into Canadian climate models.
Rolnick’s research attracts many partners and significant funding, including from major tech players IBM, Intel, Samsung, Microsoft and Google. His group has worked with NASA Harvest, a U.S.-based global food security and agriculture consortium advancing adoption of satellite-based Earth observations, and with the European Space Agency.
His group also is leading the Canadian research team, which received a $3.75- million grant from the Natural Sciences and Engineering Research Council, in a recently created, joint Canada-U.S. National Science Foundation Global Climate Center on AI and Biodiversity Change. The Canadian team involves researchers from McGill University, University of British Columbia and University of Guelph.
Rolnick and his group also are working with Montreal startup BrainBox AI, to use AI systems to better approximate and reduce energy usage of heating and cooling systems in commercial and industrial buildings.
AI-enabled applications have increasingly been connected to initiatives for sustainable agriculture and forestry, or to incentivize land-use decisions, such as managing soil to optimize carbon storage and take advantage of carbon market opportunities, Rolnick said.
Private sector interest also comes from the insurance industry, as well as from companies that provide risk-related information to either private or public entities looking to make risk-adjusted decisions for strategic planning.
Rolnick said CIFAR has been instrumental in supporting his work, “not just the monetary support but the broader ecosystem that CIFAR provides for interdisciplinary and multi-stakeholder work that crosses the public and private sectors and the often-siloed academic world.”
“For real impact to be achieved in research related to climate action, we need many stakeholders in the room.”
AI presents significant risks related to climate change
While AI can and is being used to help mitigate and adapt to climate change, the powerful technology also poses significant threats related to climate change, Rolnick said.
One of the biggest threats is AI being used to facilitate unsustainable practices, he said. He pointed to AI and machine learning applications being used extensively, in contracts between multinational oil companies and big tech players including Amazon, Microsoft and Google, to accelerate oil and gas exploration and extraction – including in Alberta’s oil sands.
AI is being used to produce more fossil fuels at a time when nearly 200 countries at the United Nations COP28 climate conference in November agreed to drive the energy transition away from fossil fuels, and when the International Energy Agency has said no new major oil and gas extractions are needed anywhere around the world.
A report by KBV Research forecast that the value of AI in the oil and gas market will increase from $US2.2 billion in 2021 to US$5.2 billion in 2028, generating considerable additional revenues for the global oil and gas industry.
Another risk is AI facilitating behaviour change in ways that aren’t aligned with climate action, Rolnick said. For example, AI is used in automated advertising systems designed to increase consumption, “which is an unmeasured but extremely significant contributor to global consumption of resources and energy.”
AI tools can have both positive and negative effects on behaviour patterns. AI-enabled autonomous driving technology, if used in personal vehicles, is actually expected to increase the impacts of climate change by decreasing the barriers to driving, he said. “If we were prioritizing self-driving buses, we would potentially have positive impacts from the same technology.”
Rolnick also sees a risk with AI being used within the information space. “There’s a danger that any shiny new tool can be seen as a replacement for the harder work and the harder decisions that need to happen across society,” he said. “AI is a useful tool, but it’s not a replacement for policy and it’s not a replacement for other mechanisms of climate action.”
AI can be used to create, promote and spread disinformation related to climate change, Rolnick said. “Increasingly we’ve seen the growth of systems that can generate especially the incorrect information very easily.”
“It is far easier to create disinformation than it is to create true content, especially given that the recent growth in large language models has not been designed to create truth. It has been designed to create text that looks true.”
Rolnick and his research group don’t work with large language models, nor does the group see many opportunities for such models to be particularly impactful in climate actions. “But its potential as a tool for counterproductive actions on climate is a danger,” he said.
AI offers many beneficial “medium-sized” impacts
Looking to the future of AI applications, Rolnick said it’s important to recognize that the impacts of AI across society – and specifically in the climate context – lie in many medium-sized impacts, rather than a single big impact.
“The opportunities are not concentrated in one space. They reside in stakeholders and players across energy, heavy industry, transportation, agriculture, forestry,” he said. “It’s using these tools, in combination with existing structures, and using them as accelerants for meaningful action.”
“There are some situations where AI can truly be a game-changer,” Rolnick said. But generally AI is a powerful tool that can help you solve problems you already had and already recognized, he said. “It can speed up but not change the kind of work that you do.”
“The idea of AI coming in and radically reshaping an industry is generally snake oil, being peddled by people who don’t have an idea of what AI is and is not capable of.”
As Rolnick and co-authors detail in the article “Tackling Climate Change with Machine Learning,” published in the journal AMC Computing Surveys, there are huge opportunities to use AI in myriad climate-related applications, such as speeding up data gathering in agriculture for crop yields and where crops are being grown. AI can be used to gather data on forests, peatlands and other areas where carbon is being stored naturally, to quantify the impacts of nature-based solution to climate change.
AI also can be used to spur innovation in materials science, in everything from EV battery storage to photovoltaic cells to solar fuels.
One area with great potential is using AI to optimize systems across difficult-to-decarbonize industrial sectors such as steel and cement manufacturing and freight transport, Rolnick said. “The role of AI in [producing] even slight percentages of increases in efficiency can be greatly felt in aggregate emissions, just because of the huge magnitude of the emissions involved and the way in which such industries have often been slow to adopt new technological tools.”
AI-enabled research and innovation requires powerful and project-appropriate, world-class supercomputing power, which as Research Money reported is in short supply in Canada.
Environment and Climate Change Canada, for example, has access to computing infrastructure that’s optimized for traditional, non-machine learning computation. However, the department doesn’t have access to supercomputers with GPUs, the graphics processing units that can process many pieces of data simultaneously and act like accelerators in high-performance computing systems. Advanced AI research and development requires GPUs.
Rolnick said there’s also a need for more data storage and data infrastructure, and the software expertise needed to maintain this. “Simply having access to tools, without the capacity for working with those tools, is not necessarily as helpful as one might expect.”
Click here for a presentation on machine learning and climate change by David Rolnick on YouTube.