Understanding AI's environmental footprint

At Mitigate, AI is central to our work. We use it extensively for research, to improve our ESG reporting platform, and in our development. AI is powerful, but it requires energy. For us, smart technology should also be environmentally responsible.

This topic is currently popular, as AI is advancing at an extremely fast pace and companies are fundamentally incorporating ESG considerations into their processes.

Making progress, but using resources wisely?

AI can bring significant improvements and help solve environmental problems in new ways. For example, AI can optimise energy use across a whole manufacturing company facilities and processes. However, AI also needs a lot of energy to function, which can harm the environment if not managed carefully. It's a balance: AI can help the planet, but we need to use it responsibly.

Training a large AI model can use as much electricity as several homes in a year! This shows that AI's capabilities come with a real energy cost. This post explains why this is important, what rules are being developed, and how organisations can use AI in a more sustainable way.

How AI impacts our environment

Here are key ways AI affects the environment:

• ⚡️ High energy use: Advanced AI models need a lot of computer power to learn and operate. This leads to significant electricity consumption, especially for complex AI like deep learning. Using these models daily also requires substantial energy.
  

• ⛽️ Greenhouse gas emissions: The electricity used for AI often comes from sources that release harmful gases into the atmosphere, contributing to climate change. High AI energy use can increase these emissions.
  

• 🌊 Water consumption: Large data centers that house AI computers need cooling systems, which often use large amounts of water. In areas with limited water, this can be a problem.
  

• 🗑️ Electronic waste: The computers and specialised hardware for AI have an environmental impact from their production to their disposal as electronic waste.

Think about AI that creates images. While it seems simple, it uses more energy than processing text. Also, AI designed for many tasks tends to use more energy than AI designed for specific, smaller tasks. Our choices about when and how we use AI have real environmental consequences.

Rules and Efforts for Greener AI

People are starting to create rules and guidelines to make AI more sustainable. For example, the European Union has a set of new rules for AI, and some of them also consider AI's environmental impact. The EU AI Act also considers sustainability in how AI is designed and used. Organisations like NIST are also working on ways to measure AI's environmental impact.

However, it's still difficult to accurately measure the energy used by AI across all its processes. This makes it challenging for companies to fully understand and reduce their AI's environmental footprint.

Where the main responsibility lies and what everyone can do

The large technology companies and those who develop the core AI models have the greatest ability to reduce AI's environmental impact. Their decisions about model design, training methods, and energy sources for their infrastructure are crucial.

For small and medium-sized businesses (SMEs), directly affecting the energy use of big AI models might be difficult. However, avoiding AI is not a good option, as it can put them at a disadvantage compared to larger companies. Instead, SMEs should focus on choosing efficient AI tools and asking their providers about their sustainability efforts. Collective demand for greener AI can drive change.

Working towards a more sustainable AI future

Regardless of company size, considering environmental impact when using AI is important. This includes:

• Understanding computer power needs: Understanding the computational requirements of your AI tools is crucial, particularly when using proprietary models, whether they are large or small.

• Monitoring energy use: Carefully monitor the electricity usage of your AI applications. Measure it as precisely as possible and include this data in your ESG reports. Inquire with your tech providers whether they track this information.

• Choose appropriate models for each task: While you can employ the most powerful models, such as LLMs (ChatGPT, Gemini, Claude), for all tasks, optimising their use by selecting smaller models for specific tasks or sub-tasks can be more efficient. Smaller models typically demand less computational power, which in turn reduces energy consumption, making them more environmentally friendly.

A smart future should also be a sustainable one. By understanding AI's environmental impact and working together – from big tech to small businesses – we can use AI responsibly. At Mitigate, we are committed to this. For example, in out ESG reporting platform we have over 20 AI-powered features that provide real value and save time for our users, and we continuously focus on making them as energy-efficient as possible.

Contact us if you're thinking about using AI for your ESG management or reporting requirements. We are always eager to share our knowledge and expertise.

🌎