Technical dynamics and advances in the field of so-called artificial intelligence (AI) and especially machine learning (ML) have led to its ubiquitous application in many societal domains and industrial sectors. Due to its pervasiveness, it is sometimes even referred to as a general purpose technology. The increasing use of AI systems raises debates about their societal, environmental and economic impacts, such as non-transparent decision-making processes, discrimination, increasing inequalities, rising energy consumption and greenhouse gas emissions in AI model development and application. Other discussions also relate to broader consequences on labor markets, consumption patterns or the market power of large corporations.
From a sustainability perspective, AI systems are thus associated with multiple challenges. We argue that these socio-technical systems, with their design, their organizational embedding and implementation processes, can unfold significant impacts on society, the environment and economic actions. The goal of this paper is to expand current discussions on responsible AI into an overarching perspective on sustainable AI. This includes a systematic overview of the social, environmental, and economic impacts along the entire AI lifecycle as well as along the organizational embedding of these systems.
This publication is based on work conducted in the research project 'SustAIn – Sustainability Index for Artificial Intelligence'. It presents conceptional ideas for a comprehensive sustainability assessment along the AI lifecycle. For this purpose, first the conceptual foundations and a sustainability perspective on AI are being explained. We then suggest thirteen sustainability criteria with corresponding indicators based on an analysis of existing scientific and societal discourses on the social, ecological and economic impacts of AI. Eventually, we will shortly sketch how in the ongoing research process these indicators will be translated into assessment tools. This paper aims at strengthening discourses on the sustainability of AI as well as at enabling a systematic assessment of the sustainability of AI systems in order to support their development and application in a sustainable manner.
Link to publication