Lorem ipsum dolor sit amet, consectetur adipiscing elit. Curabitur eleifend tortor nec augue pretium
There are numerous claims and counterclaims around the environmental impacts of artificial intelligence (AI). Some suggest the technology will slash greenhouse gas emissions by optimising energy, transport and food systems. Others point to the massive amounts of energy and water used to power and cool data centres, along with the unsustainable extraction of critical minerals and rare elements, damaging nature impacts and the production of electronic waste.
“It’s become a story of anecdotes and your word against mine,” Golestan Sally Radwan, chief digital officer at the UN Environment Programme (UNEP), tells me. “No one knows the truth, because we don’t have a common scientific understanding of what’s really going on, and we need to establish that before we can credibly make any statements around it being a net positive or negative.”
The World Economic Forum ranks “adverse outcomes from AI” as the fifth-greatest risk facing the planet over the next decade, just behind the closely related threat of “misinformation and disinformation”.
Radwan has spent decades working with governments and companies in the US, Europe and the Middle East to counter these AI risks, becoming a leading voice on ensuring that AI is a net positive for both society and the environment. She began her career at a London healthcare AI startup, later returning to Egypt to advise the minister of ICT on the national AI strategy, before moving to the UN.
Based at the UNEP headquarters in Nairobi, the computer scientist now leads a digital transformation of the agency that will enable it to better deliver its objectives for the environment, climate change and nature. “I work from two perspectives: how do we deploy digital technologies to help achieve those goals, and how do we prevent them from being too taxing on the environment? I look after that entire portfolio.”
Her work perfectly encapsulates the dilemma facing sustainability professionals today who are trying to understand whether the potential positives of AI outweigh the negatives. “We need to agree on what we mean by the environmental impact of AI. When we use Earth observation data to detect changes in land use or biodiversity, how do we quantify how much we’ve gained by using that?
“On the other side, how much has this really cost the environment? You need to break it down along many dimensions across the entire environmental lifecycle, which starts from the raw materials that go into manufacturing all these graphics processing units (GPUs), the construction of data centres, energy consumption, emissions, water use, e-waste generation, land degradation, and the rise in unsustainable consumption and production owing to the granular marketing that AI helps us do.”
A London School of Economics report last year suggested that AI could cut global greenhouse gas emissions by 3.2 to 5.4 billion tonnes annually by 2035. “Emissions reductions would outweigh increases from global power consumption of data centres and AI,” said the authors; arguing that “the world faces an unprecedented opportunity to leverage AI as a catalyst for the net-zero transition”.
However, Radwan says: “The problem with AI is similar to the problem we’ve always had with technology, which is that people think it’s just going to solve everything. AI hasn’t proved that it’s the magical solution for sustainability that everyone is hoping for. We all talk about potential applications, shifts and big changes, but that hasn’t really materialised yet.”
That’s not to say that these potential applications might not emerge soon. Research by the KTH Royal Institute of Technology in Sweden indicates that AI has the potential to help deliver 79% of targets within the UN Sustainable Development Goals.
“It’s interesting, and I’m sure there’s some truth to that, because any kind of digital technology can generate efficiencies. For example, if you use sensors and digital technologies to collect data for a country’s nationally determined contributions, of course, that’s better than sending 10,000 people with notepads to collect data manually. But where do we draw the line, and is there too much expectation of technology versus reality?”
There is a seemingly endless list of potential applications for AI, from early healthcare diagnostics to autonomous vehicles. However, there are just as many applications that could be catastrophic for the environment and society, which have flown under the radar.
Indeed, AI is quietly boosting fossil-fuel expansion, helping some of the world’s biggest polluters to discover more oil and gas and extract hard-to-reach reserves – thus driving additional greenhouse gases, known as ‘enabled emissions’. Microsoft’s potential market opportunity with fossil-fuel companies is reported to be between $35bn and $75bn annually.
“AI is a tool. You can use a knife to carve a piece of meat, and you can use a knife to kill a person, but we have laws to stop that,” Radwan continues. “You can use AI to build a factory with the lowest environmental footprint, or to find places to drill for more oil – the difference is the absence of governance that puts strict limits on what you can use AI for.”
Another challenge is AI’s ability to spread misinformation. Deepfake images, videos and fabricated text are proliferating rapidly, making it easier to create convincing, false content at scale to sway public opinion, damage reputations and undermine democratic processes.
In response, the UNEP has developed a chatbot called EnvironmentGPT, which provides verified responses from trusted scientific sources.
“The climate and environment domain is disproportionately affected by false information, so having more of these deep-fake detectors and misinformation or disinformation detectors to dispel myths is going to be really important going forward,” Radwan explains.
The chipmaker Nvidia hit a valuation of $5trn last year – just three months after becoming the world’s first $4trn company – while Microsoft and Apple have also each reached $4trn valuations; roughly the size of the UK’s entire GDP.
“Unfortunately, the narrative of an AI race makes it harder to work together, because every country now feels that they have the opportunity to leapfrog the other, which counteracts the idea of collaboration and international cooperation,” Radwan says.
She also dislikes the drive towards ‘sovereign AI’, where hardware vendors – driven by profit – encourage governments to build their own data centres, regardless of the environmental cost. Furthermore, she says that this ‘race’ is ultimately unfair, because countries are not starting from the same point.
“If you’re trying to teach the Finnish people about AI, for example, you could have the entire population educated in it within three years. If you try to do that in Egypt, you have people who still need to be taught how to read and write before you start to teach them mathematics, technology and AI.”
The UN says that 118 countries, mostly in the Global South, are currently absent from major AI governance discussions. However, Radwan admits there is a lack of capacity and even political will within these nations to help build a fairer system.
“Many of these conversations are driven by institutions that are based in the Global North. They invite people they know and rely on their own networks. In the Global South, there is a lack of well-established institutions with knowledge of AI to invite anyway.”
The UN General Assembly has been working on a new AI governance framework akin to the Paris Agreement for the past five years. However, Radwan says that ambitions keep getting watered down.
“In some ways, it’s the same tension you have between environmental sustainability and economic development. Many countries see AI as a driver for economic growth, and they see themselves as having a huge advantage over others.”
While working as the national AI lead for Egypt, she proposed an African Union for AI governance discussions to have a unified position, yet self-interests once again got in the way.
“When you go to a politician in the Global South and pitch the importance of having a voice in the global AI governance discussion, versus deploying whatever scarce resources they have towards AI for economic development, they opt for the latter, because it’s very difficult to quantify what participating in a global AI governance effort means to them.
“They’re reluctant to put in place things that will hamper their ability to innovate and lead, which makes it very difficult to make the argument that we need to regulate AI all over the world, put guardrails in place, and limits on use or proliferation and development.”
She reiterates that the promise of AI is not often rooted in scientific fact, and is instead an emotional response by people looking for a silver bullet for sustainability – and huge financial rewards.
As for her own feelings? “I think for simpler models and machine learning that are light on resources, there could be a net positive. Once you start getting into the more complex deep learning models and deep neural networks, then it is very questionable. We need to ask what the comparative gain and cost is, and in many cases, you’ll find that some basic data analysis will do an equally good job, if not better, because they’re more predictable and lighter on resources – an impact assessment needs to precede any kind of AI deployment.”
Having an “irrefutable scientific basis” of the environmental costs is key, as well as a consensus on what is an acceptable price to pay. “How do we quantify what’s acceptable consumption of minerals and metals? How do we mitigate that use? Can we define circularity standards? Can we encourage more research into reusable GPUs? Then it becomes a more practical conversation.”
For as long as the tech giants remain the gatekeepers of AI development, Radwan believes that the general public must demand that politicians and companies deliver the changes needed – in the same way citizens have fought for climate action.
“There are so many claims about the good of AI – much of which may be true – but we have no way of verifying those things, because there’s no data, there’s no transparency. You don’t know how they even did the analysis,” she continues.
“There has to be impartial third-party verification of those claims to hold people accountable and force transparency. That requires a lot of political will, but this must also come from public pressure.”
Chris Seekings is the Deputy Editor of ISEP’s Transform magazine, which is published biomonthly for ISEP members. Chris’s role involves writing sustainability-related news, features and interviews, as well as helping to plan and manage the magazine’s other day-to-day activities.
