Artificial intelligence (AI) systems impact society, the environment, and the economy. Documenting them is a prerequisite to understanding and assessing such impact. Additionally, real-world AI systems are highly complex: they may consist of multiple models and datasets, may be embedded into a software application, and may run on different hardware structures. Existing documentation practices do not sufficiently mirror this complexity or focus only on specific aspects of the possible impact of such a system.
To close this gap, we propose the AI through the lens of Multidimensional Sustainability Templates (AIMS-Templates). The framework of multidimensional sustainability, i.e., acknowledging that sustainability has a social, environmental, and economic dimension, thereby allows us to cover all relevant areas of impact of an AI system. The AIMS-Templates consist of three sub-templates, covering the AI system, the AI model, and the dataset, and were developed in an iterative process including feedback from practitioners and from applying it in two use cases. They are designed for any type of AI model and dataset, for internal and external use, and with a focus on feasibility. In addition to presenting the AIMS-Templates, we discuss first learnings from using them in practice. As such, the AIMS-Templates are a concrete proposal for achieving the often-called-for greater transparency of AI systems and a basis to understand their potential positive impact for sustainability.
