By Jules van den Berg
The sixth item of article 4 focuses on sustainability and reads as follows: “f) ‘social and environmental well-being’ means that AI systems shall be developed and used in a sustainable and environmentally friendly manner as well as in a way to benefit all human beings, while monitoring and assessing the long-term impacts on the individual, society, and democracy.”
This raises questions about its implementation. How can you, as a policy officer, administrator, or IT professional in a government organization, ensure that the internal algorithms comply with this? In this blog, you will find concrete steps to make sustainable use of algorithms feasible in your organization.
How do you ensure that your algorithm is developed, deployed, and maintained in a sustainable manner? For a complete process, the following two points are essential.
1. Start: Sustainable by design
Right from the beginning of your algorithm development, it's important to consider sustainability in the decisions to be made, much like with information security and privacy (or as it should be). This means that sustainability should be a factor from the drawing board. By establishing a flexible, future-proof architecture that incorporates electricity and water consumption in the design, you make sustainable choices at a later moment possible. Subsequently, a data scientist should consider the environmental requirements posed by different types of models and algorithms. A significant portion of the costs of new software applications comes from managing them. Therefore, it's important to involve administrators early in the process, to incorporate their knowledge into the trade-offs between performance and consumption during development, but also in management.
In each of these disciplines, there are ways to reduce the environmental impact of an algorithm in production, but to achieve these benefits, all aspects must be considered. It is therefore necessary for sustainable algorithms to involve expertise from the entire lifecycle of a model.
2. On the way: Take responsibility
The decision maker sets the goals and lets IT optimize. This means that sustainability should be discussed in all parts of a software organization, as well as the requirements you impose on your algorithms on a daily basis.
For example, it may not be necessary to continuously analyze if someone, for instance, only looks at the results once a day. Discuss this so that unnecessary environmental burden can be reduced. If the context of your data points is not relevant, do not include them in your calculations. If you can omit parameters or layers in your model and still maintain high accuracy, consider leaving them out. If you need the same data from the same sources every year or every month, set up a data governance structure to streamline this process, and thus make it more environmentally friendly. There are dozens of ways to reduce the environmental impact of algorithms in use, but this requires coordination between those setting the goals and those operating the controls.
In short, by incorporating sustainability into the design of an algorithm from the very beginning, much environmental damage can be prevented. If the genie is already out of the bottle, regular coordination between IT and decision-makers helps to jointly create the most sustainable algorithm possible.