ľ¹Ï¸£ÀûÓ°ÊÓ

Generative AI is strongly on the rise lately and is expected to also change production processes in the public sector. This technology can support various information-processing tasks within governmental organisations. A British research project shows more than 20 % of the professionals in the public sector use GenAI. However, critics point out the rising (sustainability) costs, such as energy and water consumption. These costs are often overlooked. Governments using GenAI have to find ways to limit these negative consequences, but there are few guidelines for managing (sustainability) costs. Researchers at Utrecht ľ¹Ï¸£ÀûÓ°ÊÓ have been commissioned by the Dutch Ministry of the Interior and Kingdom Relations to develop scenarios which provide insight in the expected costs and to come up with possible adaptation strategies.

Within the Dutch government, many innovative solutions which have the potential to change processes are being developed. That varies from processing large amounts of text and transcribing interviews to the development of a so-called ‘digital citizen’ – a digital application trained on earlier contacts with citizens, which can ‘think along’ with government employees on government plans in an early stage. On the other hand, there is an increasing use of external GenAI applications such as ChatGPT, MidJourney and Microsoft Co-Pilot. This is often not officially permitted, but it does happen. This is also called ‘Shadow IT’. But there is currently insufficient good oversight into the extent of this use.

Scenarios

In the report ‘Generative AI and sustainability’, Albert Meijer, Rosanna Nagtegaal and Alex de Vries have made an analysis of the sustainability costs and adaptation strategies. According to them, GenAI can develop in three ways in the upcoming years:

• Slow development: In this, GenAI applications will not be more integrated into government processes than is currently the case.
• Average development: In this, internally developed GenAI applications are scaled up and/or external applications are purchased. On top of that, GenAI can become a part of government processes by means of updates of existing software which is currently in use.
• Fast development: In this, GenAI is developed and purchased on a large scale, and all kinds of new applications beyond just processing texts are used.

The electricity consumption of GenAI used within the Dutch government can, depending on the development, reach up to respectively 14.6, 27.2 or even 53.3 GWh annually in three years. This means that in an average development, approximately 21 percent more electricity will be consumed by the government than in 2020.

Based on these costs, the CO₂ emission and water consumption were calculated. The expected electricity costs of GenAI within the Dutch government translate into respectively 3.9, 7.4 en 14.4 million kilograms of CO₂ emission and 13.7, 25.4 and 49.8 million litres of water consumption.

External organisations probably have other priorities than sustainability costs and are often not transparent about processes and costs

Adaptation strategies

If we want to use the potential of GenAI, we should be prepared for more energy and water consumption as well as more CO₂ emissions, the researchers write. This is why in order to get ahead of the negative consequences, we should pay attention to adaptation strategies now. This is not only important for sustainable use, but to save costs too. The public sector is facing the important choice to use resources for the internal development of GenAI applications, or to purchase external applications. External organisations providing GenAI applications probably have other priorities than limiting sustainability costs. On top of that, they are often not transparent about processes and costs.

It is more important that there is more attention for sustainability in policy on digitalization, and more monitoring of sustainability costs. Governmental organisations can make conscious choices for sustainable alternatives, such as reusing existing applications instead of developing new applications. A tool to provide insight in secondary terms of employment developed for one municipality could also be handy to another municipality. If municipalities collaborate, this tool or a part of it can be reused.

At a national level, central guidance by the Dutch Ministry of the Interior and Kingdom Relations on the developments in the use of GenAI applications and awareness of the costs are essential.

More information

Would you like to know more? If you do, please contact Rosanna Nagtegaal: r.nagtegaal@uu.nl or read the report ‘Generative AI and sustainability’.