Issue link: https://maltatoday.uberflip.com/i/1533584
Alexiei Dingli maltatoday | SUNDAY • 23 MARCH 2025 6 OPINION AI's dirty secret: The environmental trade-off no one talks about ARTIFICIAL Intelligence (AI) is often praised for its ability to revolutionise industries, boost efficiency, and drive innovation. However, a complex and re- source-intensive process lies behind the seamless experience of asking ChatGPT a question or generating an AI-powered image. AI models, particularly large-scale ones, consume vast amounts of energy, water, and computing power—raising pressing concerns about their environ- mental impact. While AI offers promis- ing solutions to ecological challenges, it contributes significantly to carbon emis- sions and resource depletion. So, is AI an environmental problem, or can it be part of the solution? Training large AI models requires massive computational power, trans- lating into significant energy use. For instance, training a single model like GPT-3 reportedly consumed around 1,287 megawatt-hours (MWh) of elec- tricity—enough to power hundreds of homes annually. The carbon footprint of such training processes is estimated at 500 metric tonnes of CO₂, rough- ly equivalent to the lifetime emissions of five average cars. To put this into perspective, some studies suggest that global data centres account for 2% of total carbon emissions—less than the energy or transport sectors but still a substantial and growing contributor. And this is only the beginning: as AI systems become more complex and widely adopted, their energy consump- tion is expected to rise. But energy is only part of the story. Water use is another hidden environ- mental cost of AI. Data centres, which house the massive servers needed for AI operations, require constant cool- ing to prevent overheating. Many use water-based cooling systems, meaning every ChatGPT query consumes about 1.7 millilitres of water—compared to just 1 millilitre for a Google search. On a larger scale, training a single model like GPT-3 is estimated to have used 700,000 litres of water. Researchers warn that as AI adoption grows, global AI-related water use could reach billions of cubic metres, comparable to the annual water consumption of entire countries. In a world where water scarcity is a growing crisis, this raises serious ethical and en- vironmental concerns. So, what can be done to minimise AI's environmental footprint? One key area of improvement is cooling technology. Companies like Microsoft are explor- ing advanced cooling solutions such as liquid cooling, drastically reducing water consumption. Another promis- ing approach is using recycled water or rainwater harvesting in data centres to reduce freshwater use. Some companies also strategically place data centres in cooler climates or near renewable ener- gy sources to optimise efficiency. Mean- while, AI researchers are working on model optimisation techniques, which can reduce the computational power needed to run large AI systems, making AI "leaner" and more energy-efficient. Beyond mitigation, AI offers tools to help other industries become more sus- tainable. AI-powered precision farming can optimise irrigation in agriculture, reducing unnecessary water use. Smart sensors can detect plant diseases early, allowing farmers to apply pesticides on- ly where needed, minimising chemical runoff into water sources. In forestry, AI-driven satellite imagery can track deforestation and even detect early signs of forest fires, helping authorities respond faster. Waste management, too, is benefiting from AI-powered sorting systems that separate recycla- bles more efficiently than humans, re- ducing landfill waste and promoting circular economy practices. AI can also provide real-time envi- ronmental impact assessments, helping industries and policymakers make da- ta-driven sustainability decisions. For example, AI can analyse air and water quality data, track carbon emissions, and predict environmental trends. This information allows governments and companies to act quickly—whether by enforcing stricter pollution controls or adjusting operations to reduce waste. Looking ahead, AI is already being used in biodiversity conservation, sus- tainable supply chains, and circular economy initiatives. Image and audio recognition tools help track endan- gered species and monitor ecosystems, while AI-powered logistics can reroute shipments to reduce fuel consump- tion. Automated recycling systems can identify and separate different materi- als more efficiently, ensuring fewer re- sources go to waste. These applications demonstrate AI's potential to drive environmental solutions—provided its resource consumption is kept in check. So, will AI ultimately be a net-positive or net-negative force for sustainability? The answer depends on how respon- sibly we develop and deploy it. On the one hand, AI's growing energy and wa- ter demands are a cause for concern, especially if the industry does not tran- sition to more sustainable practices. On the other hand, AI-driven innovations in renewable energy, conservation, and efficiency could help offset these im- pacts. Major tech companies like Goog- le, Microsoft, and Amazon have pledged to transition to 100% renewable energy, with Google aiming for carbon-free operations by 2030. While challenges remain—such as the intermittent na- ture of solar and wind power—ongoing investments in energy storage and grid improvements could make this shift feasible in the long run. Ultimately, AI is neither inherently good nor bad for the environment—it depends on how it is used. If we priori- tise energy-efficient AI models, sustain- able data centre practices, and environ- mentally focused applications, AI could become a powerful ally in the fight against climate change. But if left un- checked, its rising resource consump- tion could add to the very problems it seeks to solve. As with any technology, the key lies in balance: leveraging AI's strengths while minimising its foot- print, ensuring that innovation and sus- tainability go hand in hand. Alexiei Dingli is a professor of artificial intelligence