The environmental impact of AI: a complex balance

The artificial intelligence industry could consume as much energy as a country the size of the Netherlands by 2027.

When we think about protecting the environment, artificial intelligence (AI) might not be the first thing that comes to mind. Yet, AI is rapidly becoming a major player in shaping our future and impacting our planet.

In an article published by the Head of Department of Systems and Control Engineering, within the Faculty of Engineering at the University of Malta, Dr Alexandra Bonnici reminds us that while trees, healthy soil, and clean water are essential for a happy planet, AI has integrated into our daily lives and is shaping humanity’s future in various ways.

She observes that the environmental impact of AI is multifaceted. On one hand, it offers tremendous benefits such as improving supply chain efficiency, urban traffic management, and weather modelling. On the other hand, the cost of training large AI models is significant. Dr Bonnici warns that, even when AI is applied to climate-conscious technology, the costs of building and training these models could potentially leave society in a worse environmental situation.

Dr Alexandra Bonnici

The energy demands of AI training

“The artificial intelligence industry could consume as much energy as a country the size of the Netherlands by 2027.”
If this prediction is shocking to you, you should know that training AI models requires immense amounts of power and time. In her article, Dr Bonnici highlights that training a model like GPT-3 requires approximately 1,287 MWh of energy. And the energy consumption doesn’t stop there. Deployed AI models, such as ChatGPT, continue to use substantial energy. For instance, OpenAI’s ChatGPT consumes around 564 MWh every day to stay available and ready to respond.

The carbon footprint of training AI models is substantial due to the vast computational resources needed. GPT-3, with its 175 billion parameters, demands colossal power to find the optimal values for these parameters. Training such a model requires significant computational resources, translating to high energy consumption, warns Dr Bonnici.

Moreover, data centres where AI training takes place need constant power and cooling. These centres are major energy consumers, accounting for about 3% of global electricity supply and 2% of total greenhouse gas emissions.

Water usage, hardware, and e-waste

The power consumption of data centres also leads to a significant side effect: heat. Cooling these centres requires substantial water usage. Dr Bonnici notes that Google reportedly consumed 4.3 billion gallons of water in 2021 to cool its data centre servers. This water usage strains resources, especially in hotter, drier climates with increased drought risk.

AI’s environmental impact extends beyond energy and water consumption. The hardware used for AI training, such as GPUs and TPUs, has a limited lifespan. The rapid advancement in AI technologies often renders these components obsolete quickly, leading to increased electronic waste (e-waste), as Dr Bonnici highlights. Improper disposal of e-waste can contaminate soil and water, exacerbating environmental issues.

It’s not all bad

Despite its environmental costs, AI can also significantly reduce environmental impact across various sectors by optimising systems.

Consider your daily commute. AI can improve urban traffic management, reducing congestion and enhancing traffic flow. Within the Department of Systems and Control Engineering, AI technologies are used to optimise traffic by analysing real-time data from cameras and sensors. Dr Bonnici observes that AI can adjust traffic signals, reroute vehicles, and improve overall traffic efficiency. AI also predicts traffic patterns, accidents, and congestion hotspots, allowing authorities to address issues proactively.

Weather prediction and renewable energy optimisation

Accurate weather prediction enhances preparedness for extreme weather and improves resource management. AI algorithms process vast amounts of meteorological data to create precise weather models. Dr Bonnici explains that the speed of these algorithms allows models to be refined quickly with new data.” AI can also model climate change effects, providing crucial insights for policymakers.

Furthermore, AI’s modelling and predictive capabilities optimise energy generation, particularly from renewable sources. By predicting energy consumption and generation, AI helps balance renewable and non-renewable energy use, ensuring optimal availability while minimising waste. Dr Bonnici explains that AI-driven solutions can optimise the positioning of wind turbines and solar panels based on weather monitoring, maximising energy generation. This leads to the creation of truly smart grids.

Balancing the costs and benefits

While AI has environmental costs, its benefits can outweigh them with proper measures. Dr Bonnici emphasises that acknowledging AI’s negative impacts helps identify approaches to shift the balance toward a positive environmental impact. Companies like Google and Microsoft invest in green data centres powered by renewable energy with innovative cooling systems and water reuse practices.

Efforts are also made to design more efficient algorithms requiring less computational power. Techniques like model pruning and quantization reduce AI model size and complexity, and innovative methods inspired by the human brain could enhance training efficiency.

Policy and regulation

Relying solely on tech companies’ goodwill to reduce AI’s environmental impact is insufficient. Governments and regulatory bodies must ensure environmentally sustainable AI development. Policies promoting energy efficiency, supporting renewable energy adoption, and regulating e-waste recycling are essential.

Collaboration between academia, industry, and government is crucial for sustainable AI innovation. Dr Bonnici highlights that collaborative research initiatives can lead to breakthroughs in energy-efficient AI technologies. Government agencies working with academia and tech industries can develop science-based policies and guidelines promoting sustainable AI. Establishing industry standards for energy efficiency and sustainability will foster widespread adoption of best practices, creating benchmarks and certification for sustainable AI practices.

5 1 vote
Article Rating
Notify of
Inline Feedbacks
View all comments