Energy- and resource-efficient artificial intelligence – reference models, metrics, measurement methods, criteria, recommendations for action and case studies from logistics

Artificial intelligence (AI) methods and applications consume a lot of energy and resources. This applies, for example, to the collection and processing of data, the training of AI and the concrete application and adaptation of AI models. The resulting ecological footprint is at risk of becoming ever larger. At the same time, AI also has the potential to provide very practical solutions for protecting the environment and climate. For this to happen, however, AI itself should be as resource-efficient as possible. Important prerequisites for this are that the consumption of energy and resources is transparent and that valid measurement methods and metrics exist.

In the KIRA project, an attempt is being made to record the resource efficiency of Kl systems and to make this measurable. This would make it possible to validly map the environmental impacts of Kl-based systems for the first time. The reference model can be applied universally in connection with AI systems and has the potential to optimise them with regard to their resource efficiency. This promises advantages in ecological and economic terms. Last but not least, the open-source character of the project promises a high dissemination of the KIRA assessment method.