Designing artificial intelligence to be environmentally-friendly
Potenziale von KI frühzeitig heben, um Fehlentwicklungen vorzubeugen
© plainpicture / Erik Isakson
Debate about artificial intelligence (AI) is currently focused on energy consumption and carbon emissions. However, the more crucial ecological challenges lie deeper: algorithm-based decision-making systems (ADS), which are increasingly based on AI, are shaping human behaviour and social processes. This is giving rise to socio-technical mediation of environmental impacts; such mediated impacts have been largely overlooked to date by existing environmental legislation. In a new study, Oeko-Institut, along with the Independent Institute for Environmental Issues (UfU), the Society for Institutional Analysis (sofia) and experts from the German Research Center for Artificial Intelligence (DFKI), Jade University and IOW Rostock, has analysed how environmental law can be adapted and extended to respond to this development.
What is missing in the legal framework
According to the study, neither the European Union’s AI Regulation nor environmental, product or liability law adequately reflect the environmental risks and potential of digital control systems. Technology and environmental law have blank spots in this area: environmental risks often go unnoticed and sustainable innovation opportunities are barely tapped. ADS are already influencing key processes that are relevant to the environment and determining how commodity flows are directed, how fields are irrigated and how consumer offers are placed. They change routines, shift incentives and reinforce behaviour patterns. The study shows that there is also a constitutional dimension to such digitally-influenced path dependencies: the German Federal Constitutional Court's climate ruling requires risks of climate change to be limited at an early stage and technological potentials to be exploited so as not to unduly restrict the freedom of future generations.
Identifying gaps in environmental law
The study develops an innovative methodological approach: ADS are regarded as technologies with positive and negative environmental impacts that can be legally shaped. The assessment grid developed compares the environmental law objectives and instruments with the functioning of algorithmic systems. It contrasts legally-required behaviour with realistically expected system behaviour, thus revealing the lacunae of legal instruments. The aim is to determine the ‘ADS fitness’ of environmental law: where do existing instruments of environmental law already fit the specific control logic of ADS, and where should they be adapted? ‘This provides legal policy with a tool for systematically evaluating and developing the effectiveness of existing environmental law instruments,’ explains Gailhofer.
The authors of the study recommend a learning and adaptable regulatory system that starts generating knowledge about the risks and potential of ADS and reducing uncertainties at an early stage. This leaves room for regulatory flexibility. On this basis, a toolbox is being created that aligns the further development of environmental law with digital control processes. Instead of one “large” overarching regulation, existing specialist laws should be expanded so that they can deal with ADS, i.e. tailored and compatible with existing enforcement structures.
Adapting environmental law to artificial intelligence
Data governance plays an important role in the further development of environmental law. ‘ADS are only as good as the data on which they are based,’ says Gailhofer. ‘That’s why their environmental orientation depends largely on what data is available, how it is shared and who has access to it.’ The study therefore proposes considering a data governance regulation for environmental laws as well, to ensure that environmental risks and the potentials for reducing the burden on the environment are taken into account via training, testing and validation data.
Further proposals concern, for example, precautionary procedures for evaluating the environmental impacts of AI applications, legal requirements for sustainability by design and participatory and information-related elements. The recommendations focus on making sector-specific additions to existing environmental regulations.
The study highlights the areas in which environmental law should be extended so as to steer technological developments in a sustainable direction in good time. In this way, ecological and digital transformation can be interlinked: as a genuine “twin transition” that limits risks and taps potentials in a sustainable manner.
This study was conducted by the Öko-Institut e.V.
