AI, in essence, allows computers to make conclusions about a given problem, based on logical assumptions. ML, a subset of AI, attempts to use real-world observations to intuit solutions and gain new knowledge and insights in the process. The crux of their benefits lies in the ability of AI algorithms to crunch vast amounts of data, find efficiencies in various processes and predict how they might play out. Firms across the corporate landscape already aspire to harness such whizzy applications in their day-to-day operations, under the umbrella of “digital transformation”. The ostensible goal is typically to cut costs, but in many polluting sectors, what’s good for the bottom line can often help lessen the burden on the climate, too.

“You don’t have to choose between sustainability and profit,” says Sherif Elsayed-Ali, co-founder of Carbon Re, a London-based firm that works to cut emissions in heavy industry. Leveraging AI to analyse its clients’ data, Carbon Re recommends ways to greenify the cement- or steel-making process by uncovering the optimal process for the lowest possible carbon dioxide output or by determining the minimum level of energy needed to achieve a necessary chemical reaction. “The best analytic solutions to these problems are usually AI-based,” says Mr Elsayed-Ali.

Climate-related use cases like these are bubbling up in a diverse array of companies and contexts. SilviaTerra, a California-based start-up, uses remote sensing and AI to create a market for carbon credits and other benefits of forests beyond just timber. Its algorithms help to determine not only how much carbon is on a landowner’s property, but also how much of that carbon is at risk of being harvested within the next year. This marketplace allows landowners to state the price at which they would be willing to defer harvests for a single year, creating verifiable carbon credits that are then sold to firms looking to offset their emissions. This incentivises landowners to preserve their forests as carbon sinks. Zack Parisa, SilviaTerra’s co-founder, says that as forests have become a focal point for corporate conservationists over the past few years, creating markets for things such as carbon credits, wildlife habitats and fire-risk reduction can help to internalise the full value of forests in the supply chain. “[AI] tools can be leveraged to build credibility into these markets,” he says. “There’s no physical delivery of carbon credits; [our clients] won’t touch, taste or smell the carbon that they buy, so they have to buy it on faith. And that faith depends on data.”

Such examples are cropping up with increasing frequency. A global community of sustainability-minded AI developers has been coalescing over the past two years, in fields ranging from electricity management to city planning. Many instances of inefficiency in a system or lack of visibility over a process are opportunities to employ AI, a realisation that is gradually gaining steam in climate tech, according to David Rolnick, an assistant professor of computer science at McGill University in Montreal and co-founder of expert group Climate Change AI—even though the most effective AI interventions are often only incremental in nature. The constellation of use cases for AI as a weapon in the climate battle is vast; Mr Rolnick notes that its use in, for instance, moving people more smoothly through public-transit networks, or reducing the amount of fertiliser needed to grow crops, has knock-on benefits for emissions as well. “Increasing the efficiency of a process is not sexy, but it can be extremely impactful,” he says.

Token progress

There is seemingly no end to the parade of predictions, papers, analyses and prognostications about the power of AI to fight climate change. Yet its promise must be viewed in light of the fact that little hard data exists on the actual impact of AI on emissions reductions. Most of the ostensible effects are sector-dependent; Mr Elsayed-Ali, for example, estimates that using AI to cut 8% of emissions from cement production would forestall the release of 245m tonnes of carbon out of a global total of roughly 43bn tonnes annually. In the race to achieve net-zero emissions—which scientists believe must happen by mid-century in order to keep the rise in average global temperatures below 2 degrees Celsius—every bit counts, but no one expects AI to solve the climate crisis on its own.

“Right now we basically have a damaging externality which isn’t being measured,” says Olivier Corradi, founder of Tomorrow, a Danish firm that helps firms to lower the carbon intensity of electricity use. His company’s algorithms determine the best time of day to run energy-hungry operations, by detecting patterns in use based on reams of data and correlating them with external conditions—for example, recommending that a client undertake a particularly energy-intensive activity when the sun is shining, so it can be powered by solar energy rather than fossil fuels. Although any human could look out of the window to make this determination, delegating the task to machines automates the work and fine-tunes the decision-making to a degree that people would not be capable of. Mr Corradi reckons that when Tomorrow’s AI solution automatically directs flexible devices (such as heat pumps or electric vehicles) at optimal times, emissions are 10% lower than they would be otherwise.

Of course, not every climate-related problem has an AI solution. Lynn Kaack, co-founder of Climate Change AI and a researcher and lecturer with the Energy Politics Group at ETH Zürich, a Switzerland-based technical university, says that in some instances, it could make things worse. For example, a company might employ AI to increase the energy efficiency of its manufacturing process, which would lower the carbon emissions of the goods produced. But because this would lower the cost as well, the firm may simply increase production, partly offsetting gains from the reduction in carbon intensity. Ms Kaack calls for more work into building best practices in applying AI to fight climate change, in particular to understand what kind of problems AI can best be used to solve. “One has to be careful that it isn’t just a token for shiny new progress,” she says. “Not every problem lends itself to AI.”

Warrior cry

However, more climate warriors are likely to look to AI as the technology matures and understanding of its utility grows. Much of the momentum is concentrated in the research and start-up space, although big tech and governments are also getting involved. Microsoft ploughed $50m into its AI for Earth initiative, while Google has deployed its DeepMind AI programme to decarbonise its formidable energy consumption. Tackling climate change is cited as a priority in the UK’s most recent AI development policy, while the Canadian government recently announced a partnership with Microsoft to develop AI capabilities in the climate space. The US, meanwhile, is planning ambitious climate action that could further take advantage of AI, building upon work by the Department of Energy, which has invested $4.5bn in AI-enabled grid upgrades.

For all its positive impact, leveraging AI to optimise industrial processes runs the risk of papering over the fact that these activities still emit huge quantities of greenhouse gases, regardless of how efficient they are. Strengthening resilience to the worst effects of climate change—by, for example, helping to breed crops that are more resistant to drought or predicting migration patterns due to sea-level rise—could provide an even more powerful use case for AI. Yet—regardless of the application—as AI grows more capable, attention must be paid to concerns around data privacy, cyber-security, worker redundancy and other potential downsides that AI developers both in and out of the climate space must keep a watchful eye on. There is little doubt that AI can make a difference in how society confronts climate change, but only if done in a way that takes into account the complex considerations of this potentially revolutionary innovation.