A Snapshot: How Green is your AI?

However, while AI integrates itself into our everyday platforms and lives, do we have any idea of how much energy its using or the effect it has on our carbon footprint?  Some interesting statistics reported by the International Energy Agency (IEA) shows that a typical request to chatbot ChatGPT consumes 10 kilojoules — roughly ten times as much as a conventional Google search, and by 2026 the agency projects that data centres’ energy consumption will have increased by between 35% and 128%!

Here we look at some of the challenges around the environmental impact of AI and highlight some startling statistics, plus some tips to help you reduce your AI-related emissions!

• Training an AI model requires extensive computing power and energy consumption. For instance, The MIT Technology Review reported that training just one AI model can emit more than 626,00 pounds of carbon dioxide equivalent, which is nearly five times the lifetime emissions of an average car.
• Furthermore, training an AI model can lead to the evaporation of an astonishing amount of fresh water into the atmosphere for data centre heat rejection, potentially exacerbating stress on our already limited freshwater resources.
• Once trained, only a few dozen queries to an AI chatbot is roughly equivalent to boiling an electric kettle. While this may not sound like a lot, research shows that total electricity consumption from using AI could reach between 85–143 TWh by 2027, rivalling the consumption of a small country. Some, however, have commented this type of research is speculative as AI companies don’t publish enough data to confirm such claims.
• Furthermore, some research shows AI is less carbon intensive than humans are when compared on generating an illustration, or a piece of text, based on the use of laptop or the average carbon consumption per year.

5 Tips to help you reduce your carbon-related emissions:

1. Use smaller and more specialised AI models, often called Small Language Models (SLMs), as these can be fine-tuned to meet your needs. These models use less energy per task, resulting in lower carbon emissions. The AI industry is increasingly embracing this trend, as seen with models like GPT-4o mini, Google’s Gemini Flash, and Anthropic’s Claude Haiku. By being more efficient and compact, SLMs offer sustainable AI solutions with reduced energy consumption. Find a few examples of SLMs here: 5 Small-Scale Multimodal AI Models and What They Can Do – The New Stack.
2. Track carbon emissions and evaluate and monitor the environmental impact of your AI systems throughout their lifecycle. Consider using tools like the ML CO₂ Impact Calculator to assess the carbon footprint and adopt measures to minimise environmental impacts. Once you’ve entered your hardware, this nifty calculator will give you 2 numbers: the raw carbon emissions produced and the approximate offsetcarbon emissions.
3. Advance Sustainability Goals with AI: Depending on your business operations, AI can offer sustainability benefits too! For transport and logistics, intelligent transport scheduling packages can reduce your miles and fuel costs. For commercial kitchens, AI can monitor food wastage, highlighting particularly intensive waste sources.
4. Optimise your hardware for efficiency: Opt for laptops that prioritise energy efficiency by looking out for the Energy Star certificate. These indicate a higher energy efficiency and eco-friendliness rating compared to similar laptops without the classification. For laptops, an Energy Star certificate can indicates 25%-40% less energy consumption compared to a conventional laptop.
5. Develop an AI Ethics Policy: Use government guidelines to craft a clear policy that outlines your business’s commitment to ethical AI use. This policy should guide all AI-related initiatives and decisions. This process will provide you and your team with an accessible framework to enable you and your team members to explore and discuss the ethical aspects of AI.