AI-Reference Modell for Energy- and Ressource Efficiency and its Industrial Application (KIRA)

Artificial intelligence (AI) causes energy and resource consumption in all steps of the AI lifecycle (problem definition, data collection and preparation, model development, training, application, adaptation, etc.). Often required large amounts of data bring elaborate training and application scenarios. However, the abundance of available (environmentally relevant) data also offers opportunities to use AI to solve environmental and sustainability problems. Therefore, it is all the more relevant that AI itself does not become a driver for resource consumption, that energy and resource consumption is made transparent, and that valid measurement methods and metrics exist. Due to the complexity of the systems in the development process and practical application, simple methods for determining resource efficiency are difficult to use. In the project, we develop a reference model specifically geared towards AI design and application in order to structure interrelationships and dependencies and make them transparent. We develop and test criteria and metrics to optimize AI systems along their life cycle with respect to their resource and energy requirements. This includes, for example, data management, the choice of methods and frameworks, of infrastructure, efficient training, application/adaptation of the systems, and the question whether local model computations or computations in the cloud reduce the resource consumption for a given scenario. The project also aims to quantify AI processes in terms of resource consumption, at least roughly, and to develop measurement methods for this. To this end, we analyze and compare existing methods and procedures. In addition, we want to systematically include components such as sensors, actuators, energy harvesting components, etc. in the AI reference model. Based on an intralogistics case study and other openly available case studies, we evaluate and disseminate the model.