Bear environmental impacts in mind

For the time being, many AI-related processes, such as machine learning, deep learning and training large language models, are very energy-intensive operations.

The main message of most environmental impact calculations is that systems that require high computational power can be a factor that significantly exacerbates the climate and environmental crisis if the consumed electricity is produced from fossil fuel.

The widespread everday massive use of generative AI in recent years has created a new level of energy requirements for the use of AI. According to the MIT Technological Review, published by the Massachusetts Institute of Technology, the energy hunger of generative AI is significant due to the high continuous usage rates. For example, with a powerful AI model, such as Stable Diffusion, creating one image can require as much energy as charging a smartphone from a dead battery to a full charge. On the other hand, generating text alone requires only a tiny fraction of the energy needed to create images. So a responsible user will take a moment to consider whether every fun idea is worth generating images with AI.

One significant environmental and ethical burden related to IT devices are the minerals required in their components, often mined without paying any heed to environmental impacts. People in these mines are often underpaid and operate under dangerous working conditions. These worksites may also use child labour.

Many minerals are also mined in countries whose population does not ultimately benefit from the work they do, as their rate of digitalisation is very low. The fruits of their labour are consumed in post-industrial countries.

Public sector organisations should be well equipped and willing to take these factors into account when making procurements. Developing environmental certificates as procurement criteria for AI systems would be a major step forward. Before this becomes reality, each organisation must strive to make conscious and responsible decisions independently.

Read more about sustainable industrial production chains in the data economy in this Sitra blog post.Opens in a new window.