Artificial Intelligence and the environment

Earlier, you learned a little more about how algorithms are trained and how this can introduce bias into artificial intelligence (AI) systems, but there is another flaw with how AI models are trained: it uses a lot of energy!

Energy and AI

When large AI systems are being trained, this often involves giving them a huge amount of data. This can be thousands of terabytes of data. This takes a lot of time, and uses a lot of computational power, and thus energy;  these factors mean that training AI models is very inefficient.

A system like ChatGPT requires a huge amount of training, and while this training happens it is using the equivalent power needed for a small data centre which is around 1 megawatt per hour – this is roughly the same as 1000 toasters running all at once. This process takes several weeks, so this is  hundreds of megawatts of energy!

It is estimated that one of the latest versions of ChatGPT 4 took between 51,773 MWh and 62,319 MWh to train.

A data centre server rack, BalticServers.com via Wikimedia Commons, CC BY-SA 3.0,

Once the training is over, the energy used reduces considerably, but every time the system is trained, this energy use will be repeated.

While the amount of energy used is less for smaller systems, people who design AI systems still need to be careful and make sure that they design their training data well so that they do not need to re-train their algorithms and waste a lot of energy.

Hardware and the environment

The hardware used in AI such as graphical processing units (GPUs) is also concerning, as these contain rare metals. These metals are often mined using methods which are harmful both to the environment, and people who live near them.

As hardware is replaced and upgraded frequently, this can cause a lot of electronic waste (e-waste). The rare metals in e-waste are often toxic and leech into the soil which causes further harm to the environment. In some cases this causes huge amounts of soil and water pollution which can be very harmful to humans and wildlife.

Positive impacts of AI on the environment

While the amount of energy used is a concern, many AI developers are actively trying to reduce this.

AI also has the potential to help the environment because it can be used to help researchers who are looking at climate change. By analysing data and modelling what might happen in future, AI can be used to help understand how we can prevent further climate change.

Using AI to automate things can also help companies to be more energy efficient. Using AI systems to turn off lights, central heating, and air conditioning systems can help offices to use less power. A lot of energy is wasted by offices which use energy overnight, with lights or computer equipment left on. AI can help to reduce this wasted energy.