AI is often seen as an important tool for addressing the climate crisis. The potential applications of AI systems are wideranging: They are designed to make resource consumption more efficient, bring about more efficient transportation and more effective urban planning, create a more sustainable energy system and even facilitate research into new materials. Rare earths, necessary for chips and circuit boards, must be mined at great expense. The finished hardware also has to be transported to the site where it will be used. Energy is consumed at data centers to develop and deploy AI systems, and at the same time, IT infrastructure must be cooled to protect components from damage. And the hardware also must be replaced regularly, which in turn leads to new material requirements and a large amount of electronic waste. In the worstcase scenario, improper disposal can also release harmful chemicals.
The aim of environmental sustainability is to preserve nature in order to keep our planet inhabitable for future generations. The “planetary boundaries” developed among other by Johan Rockström define thresholds that, if exceeded, would result in irreversible environmental damage. AI systems impact many of these boundaries, either directly or indirectly. AI systems are the opposite of environmentally sustainable if they consume more resources than are saved or even reproduced through their use. In addition to the material consumption for hardware, their immense energy consumption and the associated emissions are an obstacle on the road to environmental sustainability.
Energy efficiency should be monitored during AI develop- ment and, if necessary, optimized through appropriate methods such as model compression.
CO2 and Greenhouse Gas Emissions
CO2 efficiency can be increased through the use of a sustainable energy mix, the appropriate choice of time and location for training, and by offsetting the CO2 emissions generated.
Sustainability Potential in Application
AI systems can have a sustainable impact if they take sustainability into account in their decision-making – if, for example, they promote sustainable products or they minimize the broader consumption of resources.
Indirect Resource Consumption
The hardware necessary for AI systems requires additional energy and resources. Here, it is criti- cal to consider environmentally friendly production and disposal.
Economic sustainability expands the horizon of economic activities: Rather than focusing the economy only on satisfying the needs of people living today, an economically sustainable perspective also plans for meeting the needs of humanity in the future. This change in consciousness is urgent against the backdrop of the “Grand Challenges” of climate change, ongoing loss of biodiversity and species, and the growing scarcity of resources. In the course of a socio-ecological transformation, fundamental questions of fairness arise because production and consumption cycles determine whether the distribution of natural resources can be reconciled with a decent and self-determined life. Economic sustainability embeds the economy between social and environmental guard rails. The effects of AI systems must also be viewed in this context. Systems that have farreaching effects on the distribution of wealth in society (for example, in the allocation of social benefits, loans or housing) as well as on economic structures and dynamics must be used in a particularly responsible manner.
Market Diversity and Exploitation of Innovation Potential
To prevent conentration in AI markets, fair access must be established for AI development through, for example, open data pools, open source code or even interfaces (APIs).
Distribution Effect in Target Markets
Access to AI applications is not available to all economic actors, leading in some cases to competitive distortions or even market concentrations. Inclusivity could be expanded by enabling models to work with small sets of data, or by enabling small and medium-sized companies to use AI through funding opportunities.
Working Conditions and Jobs
Fair working conditions should be ensured along the entire value chain of AI development. If AI is deployed in the workplace, the impact on workers should be assessed in advance and, where necessary, compensated for.
We have differentiated and operationalized these criteria in over 40 indicators
They can be read in the discussion paper:
… is a research associate at the Distributed Artificial Intelligence Laboratory at TU Berlin, where he is researching applications of Machine Learning methods for load forecasting and the sustainability of AI systems.
… is senior policy and advocacy manager at AlgorithmWatch. She works on the sustainability of automated decision- making (ADM) systems. Other areas of focus include ADM systems in the workplace and in the public sector.
… conducts research into sustainability at the Institute for Ecological Economy Research (IÖW). She is completing her Ph.D. at the TU Berlin and works on sociotechnical futures in the context of digital transformation, social innovations and algorithmic decision-making systems.
… is a research associate at the Institute for Ecological Economy Research. In the research field of environmental economics and policy, she focuses on the topics of digitalization and social change as well as economic and institutional analysis of environmental policies.